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
39 #include <linux/magic.h>
40 #include <linux/fs_context.h>
41 #include <linux/fs_parser.h>
42
43 static const struct super_operations xfs_super_operations;
44
45 static struct kset *xfs_kset; /* top-level xfs sysfs dir */
46 #ifdef DEBUG
47 static struct xfs_kobj xfs_dbg_kobj; /* global debug sysfs attrs */
48 #endif
49
50 enum xfs_dax_mode {
51 XFS_DAX_INODE = 0,
52 XFS_DAX_ALWAYS = 1,
53 XFS_DAX_NEVER = 2,
54 };
55
56 static void
xfs_mount_set_dax_mode(struct xfs_mount * mp,enum xfs_dax_mode mode)57 xfs_mount_set_dax_mode(
58 struct xfs_mount *mp,
59 enum xfs_dax_mode mode)
60 {
61 switch (mode) {
62 case XFS_DAX_INODE:
63 mp->m_flags &= ~(XFS_MOUNT_DAX_ALWAYS | XFS_MOUNT_DAX_NEVER);
64 break;
65 case XFS_DAX_ALWAYS:
66 mp->m_flags |= XFS_MOUNT_DAX_ALWAYS;
67 mp->m_flags &= ~XFS_MOUNT_DAX_NEVER;
68 break;
69 case XFS_DAX_NEVER:
70 mp->m_flags |= XFS_MOUNT_DAX_NEVER;
71 mp->m_flags &= ~XFS_MOUNT_DAX_ALWAYS;
72 break;
73 }
74 }
75
76 static const struct constant_table dax_param_enums[] = {
77 {"inode", XFS_DAX_INODE },
78 {"always", XFS_DAX_ALWAYS },
79 {"never", XFS_DAX_NEVER },
80 {}
81 };
82
83 /*
84 * Table driven mount option parser.
85 */
86 enum {
87 Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
88 Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
89 Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
90 Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
91 Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
92 Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
93 Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
94 Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
95 Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
96 };
97
98 static const struct fs_parameter_spec xfs_fs_parameters[] = {
99 fsparam_u32("logbufs", Opt_logbufs),
100 fsparam_string("logbsize", Opt_logbsize),
101 fsparam_string("logdev", Opt_logdev),
102 fsparam_string("rtdev", Opt_rtdev),
103 fsparam_flag("wsync", Opt_wsync),
104 fsparam_flag("noalign", Opt_noalign),
105 fsparam_flag("swalloc", Opt_swalloc),
106 fsparam_u32("sunit", Opt_sunit),
107 fsparam_u32("swidth", Opt_swidth),
108 fsparam_flag("nouuid", Opt_nouuid),
109 fsparam_flag("grpid", Opt_grpid),
110 fsparam_flag("nogrpid", Opt_nogrpid),
111 fsparam_flag("bsdgroups", Opt_bsdgroups),
112 fsparam_flag("sysvgroups", Opt_sysvgroups),
113 fsparam_string("allocsize", Opt_allocsize),
114 fsparam_flag("norecovery", Opt_norecovery),
115 fsparam_flag("inode64", Opt_inode64),
116 fsparam_flag("inode32", Opt_inode32),
117 fsparam_flag("ikeep", Opt_ikeep),
118 fsparam_flag("noikeep", Opt_noikeep),
119 fsparam_flag("largeio", Opt_largeio),
120 fsparam_flag("nolargeio", Opt_nolargeio),
121 fsparam_flag("attr2", Opt_attr2),
122 fsparam_flag("noattr2", Opt_noattr2),
123 fsparam_flag("filestreams", Opt_filestreams),
124 fsparam_flag("quota", Opt_quota),
125 fsparam_flag("noquota", Opt_noquota),
126 fsparam_flag("usrquota", Opt_usrquota),
127 fsparam_flag("grpquota", Opt_grpquota),
128 fsparam_flag("prjquota", Opt_prjquota),
129 fsparam_flag("uquota", Opt_uquota),
130 fsparam_flag("gquota", Opt_gquota),
131 fsparam_flag("pquota", Opt_pquota),
132 fsparam_flag("uqnoenforce", Opt_uqnoenforce),
133 fsparam_flag("gqnoenforce", Opt_gqnoenforce),
134 fsparam_flag("pqnoenforce", Opt_pqnoenforce),
135 fsparam_flag("qnoenforce", Opt_qnoenforce),
136 fsparam_flag("discard", Opt_discard),
137 fsparam_flag("nodiscard", Opt_nodiscard),
138 fsparam_flag("dax", Opt_dax),
139 fsparam_enum("dax", Opt_dax_enum, dax_param_enums),
140 {}
141 };
142
143 struct proc_xfs_info {
144 uint64_t flag;
145 char *str;
146 };
147
148 static int
xfs_fs_show_options(struct seq_file * m,struct dentry * root)149 xfs_fs_show_options(
150 struct seq_file *m,
151 struct dentry *root)
152 {
153 static struct proc_xfs_info xfs_info_set[] = {
154 /* the few simple ones we can get from the mount struct */
155 { XFS_MOUNT_IKEEP, ",ikeep" },
156 { XFS_MOUNT_WSYNC, ",wsync" },
157 { XFS_MOUNT_NOALIGN, ",noalign" },
158 { XFS_MOUNT_SWALLOC, ",swalloc" },
159 { XFS_MOUNT_NOUUID, ",nouuid" },
160 { XFS_MOUNT_NORECOVERY, ",norecovery" },
161 { XFS_MOUNT_ATTR2, ",attr2" },
162 { XFS_MOUNT_FILESTREAMS, ",filestreams" },
163 { XFS_MOUNT_GRPID, ",grpid" },
164 { XFS_MOUNT_DISCARD, ",discard" },
165 { XFS_MOUNT_LARGEIO, ",largeio" },
166 { XFS_MOUNT_DAX_ALWAYS, ",dax=always" },
167 { XFS_MOUNT_DAX_NEVER, ",dax=never" },
168 { 0, NULL }
169 };
170 struct xfs_mount *mp = XFS_M(root->d_sb);
171 struct proc_xfs_info *xfs_infop;
172
173 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
174 if (mp->m_flags & xfs_infop->flag)
175 seq_puts(m, xfs_infop->str);
176 }
177
178 seq_printf(m, ",inode%d",
179 (mp->m_flags & XFS_MOUNT_SMALL_INUMS) ? 32 : 64);
180
181 if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
182 seq_printf(m, ",allocsize=%dk",
183 (1 << mp->m_allocsize_log) >> 10);
184
185 if (mp->m_logbufs > 0)
186 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
187 if (mp->m_logbsize > 0)
188 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
189
190 if (mp->m_logname)
191 seq_show_option(m, "logdev", mp->m_logname);
192 if (mp->m_rtname)
193 seq_show_option(m, "rtdev", mp->m_rtname);
194
195 if (mp->m_dalign > 0)
196 seq_printf(m, ",sunit=%d",
197 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
198 if (mp->m_swidth > 0)
199 seq_printf(m, ",swidth=%d",
200 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
201
202 if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
203 seq_puts(m, ",usrquota");
204 else if (mp->m_qflags & XFS_UQUOTA_ACCT)
205 seq_puts(m, ",uqnoenforce");
206
207 if (mp->m_qflags & XFS_PQUOTA_ACCT) {
208 if (mp->m_qflags & XFS_PQUOTA_ENFD)
209 seq_puts(m, ",prjquota");
210 else
211 seq_puts(m, ",pqnoenforce");
212 }
213 if (mp->m_qflags & XFS_GQUOTA_ACCT) {
214 if (mp->m_qflags & XFS_GQUOTA_ENFD)
215 seq_puts(m, ",grpquota");
216 else
217 seq_puts(m, ",gqnoenforce");
218 }
219
220 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
221 seq_puts(m, ",noquota");
222
223 return 0;
224 }
225
226 /*
227 * Set parameters for inode allocation heuristics, taking into account
228 * filesystem size and inode32/inode64 mount options; i.e. specifically
229 * whether or not XFS_MOUNT_SMALL_INUMS is set.
230 *
231 * Inode allocation patterns are altered only if inode32 is requested
232 * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
233 * If altered, XFS_MOUNT_32BITINODES is set as well.
234 *
235 * An agcount independent of that in the mount structure is provided
236 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
237 * to the potentially higher ag count.
238 *
239 * Returns the maximum AG index which may contain inodes.
240 */
241 xfs_agnumber_t
xfs_set_inode_alloc(struct xfs_mount * mp,xfs_agnumber_t agcount)242 xfs_set_inode_alloc(
243 struct xfs_mount *mp,
244 xfs_agnumber_t agcount)
245 {
246 xfs_agnumber_t index;
247 xfs_agnumber_t maxagi = 0;
248 xfs_sb_t *sbp = &mp->m_sb;
249 xfs_agnumber_t max_metadata;
250 xfs_agino_t agino;
251 xfs_ino_t ino;
252
253 /*
254 * Calculate how much should be reserved for inodes to meet
255 * the max inode percentage. Used only for inode32.
256 */
257 if (M_IGEO(mp)->maxicount) {
258 uint64_t icount;
259
260 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
261 do_div(icount, 100);
262 icount += sbp->sb_agblocks - 1;
263 do_div(icount, sbp->sb_agblocks);
264 max_metadata = icount;
265 } else {
266 max_metadata = agcount;
267 }
268
269 /* Get the last possible inode in the filesystem */
270 agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
271 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
272
273 /*
274 * If user asked for no more than 32-bit inodes, and the fs is
275 * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
276 * the allocator to accommodate the request.
277 */
278 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
279 mp->m_flags |= XFS_MOUNT_32BITINODES;
280 else
281 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
282
283 for (index = 0; index < agcount; index++) {
284 struct xfs_perag *pag;
285
286 ino = XFS_AGINO_TO_INO(mp, index, agino);
287
288 pag = xfs_perag_get(mp, index);
289
290 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
291 if (ino > XFS_MAXINUMBER_32) {
292 pag->pagi_inodeok = 0;
293 pag->pagf_metadata = 0;
294 } else {
295 pag->pagi_inodeok = 1;
296 maxagi++;
297 if (index < max_metadata)
298 pag->pagf_metadata = 1;
299 else
300 pag->pagf_metadata = 0;
301 }
302 } else {
303 pag->pagi_inodeok = 1;
304 pag->pagf_metadata = 0;
305 }
306
307 xfs_perag_put(pag);
308 }
309
310 return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
311 }
312
313 STATIC int
xfs_blkdev_get(xfs_mount_t * mp,const char * name,struct block_device ** bdevp)314 xfs_blkdev_get(
315 xfs_mount_t *mp,
316 const char *name,
317 struct block_device **bdevp)
318 {
319 int error = 0;
320
321 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
322 mp);
323 if (IS_ERR(*bdevp)) {
324 error = PTR_ERR(*bdevp);
325 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
326 }
327
328 return error;
329 }
330
331 STATIC void
xfs_blkdev_put(struct block_device * bdev)332 xfs_blkdev_put(
333 struct block_device *bdev)
334 {
335 if (bdev)
336 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
337 }
338
339 void
xfs_blkdev_issue_flush(xfs_buftarg_t * buftarg)340 xfs_blkdev_issue_flush(
341 xfs_buftarg_t *buftarg)
342 {
343 blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS);
344 }
345
346 STATIC void
xfs_close_devices(struct xfs_mount * mp)347 xfs_close_devices(
348 struct xfs_mount *mp)
349 {
350 struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
351
352 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
353 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
354 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
355
356 xfs_free_buftarg(mp->m_logdev_targp);
357 xfs_blkdev_put(logdev);
358 fs_put_dax(dax_logdev);
359 }
360 if (mp->m_rtdev_targp) {
361 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
362 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
363
364 xfs_free_buftarg(mp->m_rtdev_targp);
365 xfs_blkdev_put(rtdev);
366 fs_put_dax(dax_rtdev);
367 }
368 xfs_free_buftarg(mp->m_ddev_targp);
369 fs_put_dax(dax_ddev);
370 }
371
372 /*
373 * The file system configurations are:
374 * (1) device (partition) with data and internal log
375 * (2) logical volume with data and log subvolumes.
376 * (3) logical volume with data, log, and realtime subvolumes.
377 *
378 * We only have to handle opening the log and realtime volumes here if
379 * they are present. The data subvolume has already been opened by
380 * get_sb_bdev() and is stored in sb->s_bdev.
381 */
382 STATIC int
xfs_open_devices(struct xfs_mount * mp)383 xfs_open_devices(
384 struct xfs_mount *mp)
385 {
386 struct block_device *ddev = mp->m_super->s_bdev;
387 struct dax_device *dax_ddev = fs_dax_get_by_bdev(ddev);
388 struct dax_device *dax_logdev = NULL, *dax_rtdev = NULL;
389 struct block_device *logdev = NULL, *rtdev = NULL;
390 int error;
391
392 /*
393 * Open real time and log devices - order is important.
394 */
395 if (mp->m_logname) {
396 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
397 if (error)
398 goto out;
399 dax_logdev = fs_dax_get_by_bdev(logdev);
400 }
401
402 if (mp->m_rtname) {
403 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
404 if (error)
405 goto out_close_logdev;
406
407 if (rtdev == ddev || rtdev == logdev) {
408 xfs_warn(mp,
409 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
410 error = -EINVAL;
411 goto out_close_rtdev;
412 }
413 dax_rtdev = fs_dax_get_by_bdev(rtdev);
414 }
415
416 /*
417 * Setup xfs_mount buffer target pointers
418 */
419 error = -ENOMEM;
420 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
421 if (!mp->m_ddev_targp)
422 goto out_close_rtdev;
423
424 if (rtdev) {
425 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
426 if (!mp->m_rtdev_targp)
427 goto out_free_ddev_targ;
428 }
429
430 if (logdev && logdev != ddev) {
431 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
432 if (!mp->m_logdev_targp)
433 goto out_free_rtdev_targ;
434 } else {
435 mp->m_logdev_targp = mp->m_ddev_targp;
436 }
437
438 return 0;
439
440 out_free_rtdev_targ:
441 if (mp->m_rtdev_targp)
442 xfs_free_buftarg(mp->m_rtdev_targp);
443 out_free_ddev_targ:
444 xfs_free_buftarg(mp->m_ddev_targp);
445 out_close_rtdev:
446 xfs_blkdev_put(rtdev);
447 fs_put_dax(dax_rtdev);
448 out_close_logdev:
449 if (logdev && logdev != ddev) {
450 xfs_blkdev_put(logdev);
451 fs_put_dax(dax_logdev);
452 }
453 out:
454 fs_put_dax(dax_ddev);
455 return error;
456 }
457
458 /*
459 * Setup xfs_mount buffer target pointers based on superblock
460 */
461 STATIC int
xfs_setup_devices(struct xfs_mount * mp)462 xfs_setup_devices(
463 struct xfs_mount *mp)
464 {
465 int error;
466
467 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
468 if (error)
469 return error;
470
471 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
472 unsigned int log_sector_size = BBSIZE;
473
474 if (xfs_sb_version_hassector(&mp->m_sb))
475 log_sector_size = mp->m_sb.sb_logsectsize;
476 error = xfs_setsize_buftarg(mp->m_logdev_targp,
477 log_sector_size);
478 if (error)
479 return error;
480 }
481 if (mp->m_rtdev_targp) {
482 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
483 mp->m_sb.sb_sectsize);
484 if (error)
485 return error;
486 }
487
488 return 0;
489 }
490
491 STATIC int
xfs_init_mount_workqueues(struct xfs_mount * mp)492 xfs_init_mount_workqueues(
493 struct xfs_mount *mp)
494 {
495 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
496 WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_super->s_id);
497 if (!mp->m_buf_workqueue)
498 goto out;
499
500 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
501 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_super->s_id);
502 if (!mp->m_unwritten_workqueue)
503 goto out_destroy_buf;
504
505 mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
506 WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND,
507 0, mp->m_super->s_id);
508 if (!mp->m_cil_workqueue)
509 goto out_destroy_unwritten;
510
511 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
512 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_super->s_id);
513 if (!mp->m_reclaim_workqueue)
514 goto out_destroy_cil;
515
516 mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
517 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_super->s_id);
518 if (!mp->m_eofblocks_workqueue)
519 goto out_destroy_reclaim;
520
521 mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s", WQ_FREEZABLE, 0,
522 mp->m_super->s_id);
523 if (!mp->m_sync_workqueue)
524 goto out_destroy_eofb;
525
526 return 0;
527
528 out_destroy_eofb:
529 destroy_workqueue(mp->m_eofblocks_workqueue);
530 out_destroy_reclaim:
531 destroy_workqueue(mp->m_reclaim_workqueue);
532 out_destroy_cil:
533 destroy_workqueue(mp->m_cil_workqueue);
534 out_destroy_unwritten:
535 destroy_workqueue(mp->m_unwritten_workqueue);
536 out_destroy_buf:
537 destroy_workqueue(mp->m_buf_workqueue);
538 out:
539 return -ENOMEM;
540 }
541
542 STATIC void
xfs_destroy_mount_workqueues(struct xfs_mount * mp)543 xfs_destroy_mount_workqueues(
544 struct xfs_mount *mp)
545 {
546 destroy_workqueue(mp->m_sync_workqueue);
547 destroy_workqueue(mp->m_eofblocks_workqueue);
548 destroy_workqueue(mp->m_reclaim_workqueue);
549 destroy_workqueue(mp->m_cil_workqueue);
550 destroy_workqueue(mp->m_unwritten_workqueue);
551 destroy_workqueue(mp->m_buf_workqueue);
552 }
553
554 static void
xfs_flush_inodes_worker(struct work_struct * work)555 xfs_flush_inodes_worker(
556 struct work_struct *work)
557 {
558 struct xfs_mount *mp = container_of(work, struct xfs_mount,
559 m_flush_inodes_work);
560 struct super_block *sb = mp->m_super;
561
562 if (down_read_trylock(&sb->s_umount)) {
563 sync_inodes_sb(sb);
564 up_read(&sb->s_umount);
565 }
566 }
567
568 /*
569 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
570 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
571 * for IO to complete so that we effectively throttle multiple callers to the
572 * rate at which IO is completing.
573 */
574 void
xfs_flush_inodes(struct xfs_mount * mp)575 xfs_flush_inodes(
576 struct xfs_mount *mp)
577 {
578 /*
579 * If flush_work() returns true then that means we waited for a flush
580 * which was already in progress. Don't bother running another scan.
581 */
582 if (flush_work(&mp->m_flush_inodes_work))
583 return;
584
585 queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
586 flush_work(&mp->m_flush_inodes_work);
587 }
588
589 /* Catch misguided souls that try to use this interface on XFS */
590 STATIC struct inode *
xfs_fs_alloc_inode(struct super_block * sb)591 xfs_fs_alloc_inode(
592 struct super_block *sb)
593 {
594 BUG();
595 return NULL;
596 }
597
598 #ifdef DEBUG
599 static void
xfs_check_delalloc(struct xfs_inode * ip,int whichfork)600 xfs_check_delalloc(
601 struct xfs_inode *ip,
602 int whichfork)
603 {
604 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
605 struct xfs_bmbt_irec got;
606 struct xfs_iext_cursor icur;
607
608 if (!ifp || !xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got))
609 return;
610 do {
611 if (isnullstartblock(got.br_startblock)) {
612 xfs_warn(ip->i_mount,
613 "ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]",
614 ip->i_ino,
615 whichfork == XFS_DATA_FORK ? "data" : "cow",
616 got.br_startoff, got.br_blockcount);
617 }
618 } while (xfs_iext_next_extent(ifp, &icur, &got));
619 }
620 #else
621 #define xfs_check_delalloc(ip, whichfork) do { } while (0)
622 #endif
623
624 /*
625 * Now that the generic code is guaranteed not to be accessing
626 * the linux inode, we can inactivate and reclaim the inode.
627 */
628 STATIC void
xfs_fs_destroy_inode(struct inode * inode)629 xfs_fs_destroy_inode(
630 struct inode *inode)
631 {
632 struct xfs_inode *ip = XFS_I(inode);
633
634 trace_xfs_destroy_inode(ip);
635
636 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
637 XFS_STATS_INC(ip->i_mount, vn_rele);
638 XFS_STATS_INC(ip->i_mount, vn_remove);
639
640 xfs_inactive(ip);
641
642 if (!XFS_FORCED_SHUTDOWN(ip->i_mount) && ip->i_delayed_blks) {
643 xfs_check_delalloc(ip, XFS_DATA_FORK);
644 xfs_check_delalloc(ip, XFS_COW_FORK);
645 ASSERT(0);
646 }
647
648 XFS_STATS_INC(ip->i_mount, vn_reclaim);
649
650 /*
651 * We should never get here with one of the reclaim flags already set.
652 */
653 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
654 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
655
656 /*
657 * We always use background reclaim here because even if the inode is
658 * clean, it still may be under IO and hence we have wait for IO
659 * completion to occur before we can reclaim the inode. The background
660 * reclaim path handles this more efficiently than we can here, so
661 * simply let background reclaim tear down all inodes.
662 */
663 xfs_inode_set_reclaim_tag(ip);
664 }
665
666 static void
xfs_fs_dirty_inode(struct inode * inode,int flag)667 xfs_fs_dirty_inode(
668 struct inode *inode,
669 int flag)
670 {
671 struct xfs_inode *ip = XFS_I(inode);
672 struct xfs_mount *mp = ip->i_mount;
673 struct xfs_trans *tp;
674
675 if (!(inode->i_sb->s_flags & SB_LAZYTIME))
676 return;
677 if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
678 return;
679
680 if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
681 return;
682 xfs_ilock(ip, XFS_ILOCK_EXCL);
683 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
684 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
685 xfs_trans_commit(tp);
686 }
687
688 /*
689 * Slab object creation initialisation for the XFS inode.
690 * This covers only the idempotent fields in the XFS inode;
691 * all other fields need to be initialised on allocation
692 * from the slab. This avoids the need to repeatedly initialise
693 * fields in the xfs inode that left in the initialise state
694 * when freeing the inode.
695 */
696 STATIC void
xfs_fs_inode_init_once(void * inode)697 xfs_fs_inode_init_once(
698 void *inode)
699 {
700 struct xfs_inode *ip = inode;
701
702 memset(ip, 0, sizeof(struct xfs_inode));
703
704 /* vfs inode */
705 inode_init_once(VFS_I(ip));
706
707 /* xfs inode */
708 atomic_set(&ip->i_pincount, 0);
709 spin_lock_init(&ip->i_flags_lock);
710
711 mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
712 "xfsino", ip->i_ino);
713 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
714 "xfsino", ip->i_ino);
715 }
716
717 /*
718 * We do an unlocked check for XFS_IDONTCACHE here because we are already
719 * serialised against cache hits here via the inode->i_lock and igrab() in
720 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
721 * racing with us, and it avoids needing to grab a spinlock here for every inode
722 * we drop the final reference on.
723 */
724 STATIC int
xfs_fs_drop_inode(struct inode * inode)725 xfs_fs_drop_inode(
726 struct inode *inode)
727 {
728 struct xfs_inode *ip = XFS_I(inode);
729
730 /*
731 * If this unlinked inode is in the middle of recovery, don't
732 * drop the inode just yet; log recovery will take care of
733 * that. See the comment for this inode flag.
734 */
735 if (ip->i_flags & XFS_IRECOVERY) {
736 ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
737 return 0;
738 }
739
740 return generic_drop_inode(inode);
741 }
742
743 static void
xfs_mount_free(struct xfs_mount * mp)744 xfs_mount_free(
745 struct xfs_mount *mp)
746 {
747 kfree(mp->m_rtname);
748 kfree(mp->m_logname);
749 kmem_free(mp);
750 }
751
752 STATIC int
xfs_fs_sync_fs(struct super_block * sb,int wait)753 xfs_fs_sync_fs(
754 struct super_block *sb,
755 int wait)
756 {
757 struct xfs_mount *mp = XFS_M(sb);
758
759 /*
760 * Doing anything during the async pass would be counterproductive.
761 */
762 if (!wait)
763 return 0;
764
765 xfs_log_force(mp, XFS_LOG_SYNC);
766 if (laptop_mode) {
767 /*
768 * The disk must be active because we're syncing.
769 * We schedule log work now (now that the disk is
770 * active) instead of later (when it might not be).
771 */
772 flush_delayed_work(&mp->m_log->l_work);
773 }
774
775 return 0;
776 }
777
778 STATIC int
xfs_fs_statfs(struct dentry * dentry,struct kstatfs * statp)779 xfs_fs_statfs(
780 struct dentry *dentry,
781 struct kstatfs *statp)
782 {
783 struct xfs_mount *mp = XFS_M(dentry->d_sb);
784 xfs_sb_t *sbp = &mp->m_sb;
785 struct xfs_inode *ip = XFS_I(d_inode(dentry));
786 uint64_t fakeinos, id;
787 uint64_t icount;
788 uint64_t ifree;
789 uint64_t fdblocks;
790 xfs_extlen_t lsize;
791 int64_t ffree;
792
793 statp->f_type = XFS_SUPER_MAGIC;
794 statp->f_namelen = MAXNAMELEN - 1;
795
796 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
797 statp->f_fsid = u64_to_fsid(id);
798
799 icount = percpu_counter_sum(&mp->m_icount);
800 ifree = percpu_counter_sum(&mp->m_ifree);
801 fdblocks = percpu_counter_sum(&mp->m_fdblocks);
802
803 spin_lock(&mp->m_sb_lock);
804 statp->f_bsize = sbp->sb_blocksize;
805 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
806 statp->f_blocks = sbp->sb_dblocks - lsize;
807 spin_unlock(&mp->m_sb_lock);
808
809 /* make sure statp->f_bfree does not underflow */
810 statp->f_bfree = max_t(int64_t, fdblocks - mp->m_alloc_set_aside, 0);
811 statp->f_bavail = statp->f_bfree;
812
813 fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
814 statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
815 if (M_IGEO(mp)->maxicount)
816 statp->f_files = min_t(typeof(statp->f_files),
817 statp->f_files,
818 M_IGEO(mp)->maxicount);
819
820 /* If sb_icount overshot maxicount, report actual allocation */
821 statp->f_files = max_t(typeof(statp->f_files),
822 statp->f_files,
823 sbp->sb_icount);
824
825 /* make sure statp->f_ffree does not underflow */
826 ffree = statp->f_files - (icount - ifree);
827 statp->f_ffree = max_t(int64_t, ffree, 0);
828
829
830 if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
831 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
832 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
833 xfs_qm_statvfs(ip, statp);
834
835 if (XFS_IS_REALTIME_MOUNT(mp) &&
836 (ip->i_d.di_flags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
837 statp->f_blocks = sbp->sb_rblocks;
838 statp->f_bavail = statp->f_bfree =
839 sbp->sb_frextents * sbp->sb_rextsize;
840 }
841
842 return 0;
843 }
844
845 STATIC void
xfs_save_resvblks(struct xfs_mount * mp)846 xfs_save_resvblks(struct xfs_mount *mp)
847 {
848 uint64_t resblks = 0;
849
850 mp->m_resblks_save = mp->m_resblks;
851 xfs_reserve_blocks(mp, &resblks, NULL);
852 }
853
854 STATIC void
xfs_restore_resvblks(struct xfs_mount * mp)855 xfs_restore_resvblks(struct xfs_mount *mp)
856 {
857 uint64_t resblks;
858
859 if (mp->m_resblks_save) {
860 resblks = mp->m_resblks_save;
861 mp->m_resblks_save = 0;
862 } else
863 resblks = xfs_default_resblks(mp);
864
865 xfs_reserve_blocks(mp, &resblks, NULL);
866 }
867
868 /*
869 * Trigger writeback of all the dirty metadata in the file system.
870 *
871 * This ensures that the metadata is written to their location on disk rather
872 * than just existing in transactions in the log. This means after a quiesce
873 * there is no log replay required to write the inodes to disk - this is the
874 * primary difference between a sync and a quiesce.
875 *
876 * We cancel log work early here to ensure all transactions the log worker may
877 * run have finished before we clean up and log the superblock and write an
878 * unmount record. The unfreeze process is responsible for restarting the log
879 * worker correctly.
880 */
881 void
xfs_quiesce_attr(struct xfs_mount * mp)882 xfs_quiesce_attr(
883 struct xfs_mount *mp)
884 {
885 int error = 0;
886
887 cancel_delayed_work_sync(&mp->m_log->l_work);
888
889 /* force the log to unpin objects from the now complete transactions */
890 xfs_log_force(mp, XFS_LOG_SYNC);
891
892
893 /* Push the superblock and write an unmount record */
894 error = xfs_log_sbcount(mp);
895 if (error)
896 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
897 "Frozen image may not be consistent.");
898 xfs_log_quiesce(mp);
899 }
900
901 /*
902 * Second stage of a freeze. The data is already frozen so we only
903 * need to take care of the metadata. Once that's done sync the superblock
904 * to the log to dirty it in case of a crash while frozen. This ensures that we
905 * will recover the unlinked inode lists on the next mount.
906 */
907 STATIC int
xfs_fs_freeze(struct super_block * sb)908 xfs_fs_freeze(
909 struct super_block *sb)
910 {
911 struct xfs_mount *mp = XFS_M(sb);
912 unsigned int flags;
913 int ret;
914
915 /*
916 * The filesystem is now frozen far enough that memory reclaim
917 * cannot safely operate on the filesystem. Hence we need to
918 * set a GFP_NOFS context here to avoid recursion deadlocks.
919 */
920 flags = memalloc_nofs_save();
921 xfs_stop_block_reaping(mp);
922 xfs_save_resvblks(mp);
923 xfs_quiesce_attr(mp);
924 ret = xfs_sync_sb(mp, true);
925 memalloc_nofs_restore(flags);
926 return ret;
927 }
928
929 STATIC int
xfs_fs_unfreeze(struct super_block * sb)930 xfs_fs_unfreeze(
931 struct super_block *sb)
932 {
933 struct xfs_mount *mp = XFS_M(sb);
934
935 xfs_restore_resvblks(mp);
936 xfs_log_work_queue(mp);
937 xfs_start_block_reaping(mp);
938 return 0;
939 }
940
941 /*
942 * This function fills in xfs_mount_t fields based on mount args.
943 * Note: the superblock _has_ now been read in.
944 */
945 STATIC int
xfs_finish_flags(struct xfs_mount * mp)946 xfs_finish_flags(
947 struct xfs_mount *mp)
948 {
949 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
950
951 /* Fail a mount where the logbuf is smaller than the log stripe */
952 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
953 if (mp->m_logbsize <= 0 &&
954 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
955 mp->m_logbsize = mp->m_sb.sb_logsunit;
956 } else if (mp->m_logbsize > 0 &&
957 mp->m_logbsize < mp->m_sb.sb_logsunit) {
958 xfs_warn(mp,
959 "logbuf size must be greater than or equal to log stripe size");
960 return -EINVAL;
961 }
962 } else {
963 /* Fail a mount if the logbuf is larger than 32K */
964 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
965 xfs_warn(mp,
966 "logbuf size for version 1 logs must be 16K or 32K");
967 return -EINVAL;
968 }
969 }
970
971 /*
972 * V5 filesystems always use attr2 format for attributes.
973 */
974 if (xfs_sb_version_hascrc(&mp->m_sb) &&
975 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
976 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
977 "attr2 is always enabled for V5 filesystems.");
978 return -EINVAL;
979 }
980
981 /*
982 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
983 * told by noattr2 to turn it off
984 */
985 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
986 !(mp->m_flags & XFS_MOUNT_NOATTR2))
987 mp->m_flags |= XFS_MOUNT_ATTR2;
988
989 /*
990 * prohibit r/w mounts of read-only filesystems
991 */
992 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
993 xfs_warn(mp,
994 "cannot mount a read-only filesystem as read-write");
995 return -EROFS;
996 }
997
998 if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
999 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
1000 !xfs_sb_version_has_pquotino(&mp->m_sb)) {
1001 xfs_warn(mp,
1002 "Super block does not support project and group quota together");
1003 return -EINVAL;
1004 }
1005
1006 return 0;
1007 }
1008
1009 static int
xfs_init_percpu_counters(struct xfs_mount * mp)1010 xfs_init_percpu_counters(
1011 struct xfs_mount *mp)
1012 {
1013 int error;
1014
1015 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1016 if (error)
1017 return -ENOMEM;
1018
1019 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1020 if (error)
1021 goto free_icount;
1022
1023 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1024 if (error)
1025 goto free_ifree;
1026
1027 error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1028 if (error)
1029 goto free_fdblocks;
1030
1031 return 0;
1032
1033 free_fdblocks:
1034 percpu_counter_destroy(&mp->m_fdblocks);
1035 free_ifree:
1036 percpu_counter_destroy(&mp->m_ifree);
1037 free_icount:
1038 percpu_counter_destroy(&mp->m_icount);
1039 return -ENOMEM;
1040 }
1041
1042 void
xfs_reinit_percpu_counters(struct xfs_mount * mp)1043 xfs_reinit_percpu_counters(
1044 struct xfs_mount *mp)
1045 {
1046 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1047 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1048 percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1049 }
1050
1051 static void
xfs_destroy_percpu_counters(struct xfs_mount * mp)1052 xfs_destroy_percpu_counters(
1053 struct xfs_mount *mp)
1054 {
1055 percpu_counter_destroy(&mp->m_icount);
1056 percpu_counter_destroy(&mp->m_ifree);
1057 percpu_counter_destroy(&mp->m_fdblocks);
1058 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1059 percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1060 percpu_counter_destroy(&mp->m_delalloc_blks);
1061 }
1062
1063 static void
xfs_fs_put_super(struct super_block * sb)1064 xfs_fs_put_super(
1065 struct super_block *sb)
1066 {
1067 struct xfs_mount *mp = XFS_M(sb);
1068
1069 /* if ->fill_super failed, we have no mount to tear down */
1070 if (!sb->s_fs_info)
1071 return;
1072
1073 xfs_notice(mp, "Unmounting Filesystem");
1074 xfs_filestream_unmount(mp);
1075 xfs_unmountfs(mp);
1076
1077 xfs_freesb(mp);
1078 free_percpu(mp->m_stats.xs_stats);
1079 xfs_destroy_percpu_counters(mp);
1080 xfs_destroy_mount_workqueues(mp);
1081 xfs_close_devices(mp);
1082
1083 sb->s_fs_info = NULL;
1084 xfs_mount_free(mp);
1085 }
1086
1087 static long
xfs_fs_nr_cached_objects(struct super_block * sb,struct shrink_control * sc)1088 xfs_fs_nr_cached_objects(
1089 struct super_block *sb,
1090 struct shrink_control *sc)
1091 {
1092 /* Paranoia: catch incorrect calls during mount setup or teardown */
1093 if (WARN_ON_ONCE(!sb->s_fs_info))
1094 return 0;
1095 return xfs_reclaim_inodes_count(XFS_M(sb));
1096 }
1097
1098 static long
xfs_fs_free_cached_objects(struct super_block * sb,struct shrink_control * sc)1099 xfs_fs_free_cached_objects(
1100 struct super_block *sb,
1101 struct shrink_control *sc)
1102 {
1103 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1104 }
1105
1106 static const struct super_operations xfs_super_operations = {
1107 .alloc_inode = xfs_fs_alloc_inode,
1108 .destroy_inode = xfs_fs_destroy_inode,
1109 .dirty_inode = xfs_fs_dirty_inode,
1110 .drop_inode = xfs_fs_drop_inode,
1111 .put_super = xfs_fs_put_super,
1112 .sync_fs = xfs_fs_sync_fs,
1113 .freeze_fs = xfs_fs_freeze,
1114 .unfreeze_fs = xfs_fs_unfreeze,
1115 .statfs = xfs_fs_statfs,
1116 .show_options = xfs_fs_show_options,
1117 .nr_cached_objects = xfs_fs_nr_cached_objects,
1118 .free_cached_objects = xfs_fs_free_cached_objects,
1119 };
1120
1121 static int
suffix_kstrtoint(const char * s,unsigned int base,int * res)1122 suffix_kstrtoint(
1123 const char *s,
1124 unsigned int base,
1125 int *res)
1126 {
1127 int last, shift_left_factor = 0, _res;
1128 char *value;
1129 int ret = 0;
1130
1131 value = kstrdup(s, GFP_KERNEL);
1132 if (!value)
1133 return -ENOMEM;
1134
1135 last = strlen(value) - 1;
1136 if (value[last] == 'K' || value[last] == 'k') {
1137 shift_left_factor = 10;
1138 value[last] = '\0';
1139 }
1140 if (value[last] == 'M' || value[last] == 'm') {
1141 shift_left_factor = 20;
1142 value[last] = '\0';
1143 }
1144 if (value[last] == 'G' || value[last] == 'g') {
1145 shift_left_factor = 30;
1146 value[last] = '\0';
1147 }
1148
1149 if (kstrtoint(value, base, &_res))
1150 ret = -EINVAL;
1151 kfree(value);
1152 *res = _res << shift_left_factor;
1153 return ret;
1154 }
1155
1156 /*
1157 * Set mount state from a mount option.
1158 *
1159 * NOTE: mp->m_super is NULL here!
1160 */
1161 static int
xfs_fc_parse_param(struct fs_context * fc,struct fs_parameter * param)1162 xfs_fc_parse_param(
1163 struct fs_context *fc,
1164 struct fs_parameter *param)
1165 {
1166 struct xfs_mount *mp = fc->s_fs_info;
1167 struct fs_parse_result result;
1168 int size = 0;
1169 int opt;
1170
1171 opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1172 if (opt < 0)
1173 return opt;
1174
1175 switch (opt) {
1176 case Opt_logbufs:
1177 mp->m_logbufs = result.uint_32;
1178 return 0;
1179 case Opt_logbsize:
1180 if (suffix_kstrtoint(param->string, 10, &mp->m_logbsize))
1181 return -EINVAL;
1182 return 0;
1183 case Opt_logdev:
1184 kfree(mp->m_logname);
1185 mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1186 if (!mp->m_logname)
1187 return -ENOMEM;
1188 return 0;
1189 case Opt_rtdev:
1190 kfree(mp->m_rtname);
1191 mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1192 if (!mp->m_rtname)
1193 return -ENOMEM;
1194 return 0;
1195 case Opt_allocsize:
1196 if (suffix_kstrtoint(param->string, 10, &size))
1197 return -EINVAL;
1198 mp->m_allocsize_log = ffs(size) - 1;
1199 mp->m_flags |= XFS_MOUNT_ALLOCSIZE;
1200 return 0;
1201 case Opt_grpid:
1202 case Opt_bsdgroups:
1203 mp->m_flags |= XFS_MOUNT_GRPID;
1204 return 0;
1205 case Opt_nogrpid:
1206 case Opt_sysvgroups:
1207 mp->m_flags &= ~XFS_MOUNT_GRPID;
1208 return 0;
1209 case Opt_wsync:
1210 mp->m_flags |= XFS_MOUNT_WSYNC;
1211 return 0;
1212 case Opt_norecovery:
1213 mp->m_flags |= XFS_MOUNT_NORECOVERY;
1214 return 0;
1215 case Opt_noalign:
1216 mp->m_flags |= XFS_MOUNT_NOALIGN;
1217 return 0;
1218 case Opt_swalloc:
1219 mp->m_flags |= XFS_MOUNT_SWALLOC;
1220 return 0;
1221 case Opt_sunit:
1222 mp->m_dalign = result.uint_32;
1223 return 0;
1224 case Opt_swidth:
1225 mp->m_swidth = result.uint_32;
1226 return 0;
1227 case Opt_inode32:
1228 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1229 return 0;
1230 case Opt_inode64:
1231 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1232 return 0;
1233 case Opt_nouuid:
1234 mp->m_flags |= XFS_MOUNT_NOUUID;
1235 return 0;
1236 case Opt_largeio:
1237 mp->m_flags |= XFS_MOUNT_LARGEIO;
1238 return 0;
1239 case Opt_nolargeio:
1240 mp->m_flags &= ~XFS_MOUNT_LARGEIO;
1241 return 0;
1242 case Opt_filestreams:
1243 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
1244 return 0;
1245 case Opt_noquota:
1246 mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1247 mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1248 mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
1249 return 0;
1250 case Opt_quota:
1251 case Opt_uquota:
1252 case Opt_usrquota:
1253 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
1254 XFS_UQUOTA_ENFD);
1255 return 0;
1256 case Opt_qnoenforce:
1257 case Opt_uqnoenforce:
1258 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
1259 mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1260 return 0;
1261 case Opt_pquota:
1262 case Opt_prjquota:
1263 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
1264 XFS_PQUOTA_ENFD);
1265 return 0;
1266 case Opt_pqnoenforce:
1267 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
1268 mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1269 return 0;
1270 case Opt_gquota:
1271 case Opt_grpquota:
1272 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
1273 XFS_GQUOTA_ENFD);
1274 return 0;
1275 case Opt_gqnoenforce:
1276 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
1277 mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1278 return 0;
1279 case Opt_discard:
1280 mp->m_flags |= XFS_MOUNT_DISCARD;
1281 return 0;
1282 case Opt_nodiscard:
1283 mp->m_flags &= ~XFS_MOUNT_DISCARD;
1284 return 0;
1285 #ifdef CONFIG_FS_DAX
1286 case Opt_dax:
1287 xfs_mount_set_dax_mode(mp, XFS_DAX_ALWAYS);
1288 return 0;
1289 case Opt_dax_enum:
1290 xfs_mount_set_dax_mode(mp, result.uint_32);
1291 return 0;
1292 #endif
1293 /* Following mount options will be removed in September 2025 */
1294 case Opt_ikeep:
1295 xfs_warn(mp, "%s mount option is deprecated.", param->key);
1296 mp->m_flags |= XFS_MOUNT_IKEEP;
1297 return 0;
1298 case Opt_noikeep:
1299 xfs_warn(mp, "%s mount option is deprecated.", param->key);
1300 mp->m_flags &= ~XFS_MOUNT_IKEEP;
1301 return 0;
1302 case Opt_attr2:
1303 xfs_warn(mp, "%s mount option is deprecated.", param->key);
1304 mp->m_flags |= XFS_MOUNT_ATTR2;
1305 return 0;
1306 case Opt_noattr2:
1307 xfs_warn(mp, "%s mount option is deprecated.", param->key);
1308 mp->m_flags &= ~XFS_MOUNT_ATTR2;
1309 mp->m_flags |= XFS_MOUNT_NOATTR2;
1310 return 0;
1311 default:
1312 xfs_warn(mp, "unknown mount option [%s].", param->key);
1313 return -EINVAL;
1314 }
1315
1316 return 0;
1317 }
1318
1319 static int
xfs_fc_validate_params(struct xfs_mount * mp)1320 xfs_fc_validate_params(
1321 struct xfs_mount *mp)
1322 {
1323 /*
1324 * no recovery flag requires a read-only mount
1325 */
1326 if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
1327 !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1328 xfs_warn(mp, "no-recovery mounts must be read-only.");
1329 return -EINVAL;
1330 }
1331
1332 if ((mp->m_flags & XFS_MOUNT_NOALIGN) &&
1333 (mp->m_dalign || mp->m_swidth)) {
1334 xfs_warn(mp,
1335 "sunit and swidth options incompatible with the noalign option");
1336 return -EINVAL;
1337 }
1338
1339 if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1340 xfs_warn(mp, "quota support not available in this kernel.");
1341 return -EINVAL;
1342 }
1343
1344 if ((mp->m_dalign && !mp->m_swidth) ||
1345 (!mp->m_dalign && mp->m_swidth)) {
1346 xfs_warn(mp, "sunit and swidth must be specified together");
1347 return -EINVAL;
1348 }
1349
1350 if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1351 xfs_warn(mp,
1352 "stripe width (%d) must be a multiple of the stripe unit (%d)",
1353 mp->m_swidth, mp->m_dalign);
1354 return -EINVAL;
1355 }
1356
1357 if (mp->m_logbufs != -1 &&
1358 mp->m_logbufs != 0 &&
1359 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1360 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1361 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1362 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1363 return -EINVAL;
1364 }
1365
1366 if (mp->m_logbsize != -1 &&
1367 mp->m_logbsize != 0 &&
1368 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1369 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1370 !is_power_of_2(mp->m_logbsize))) {
1371 xfs_warn(mp,
1372 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1373 mp->m_logbsize);
1374 return -EINVAL;
1375 }
1376
1377 if ((mp->m_flags & XFS_MOUNT_ALLOCSIZE) &&
1378 (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1379 mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1380 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1381 mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1382 return -EINVAL;
1383 }
1384
1385 return 0;
1386 }
1387
1388 static int
xfs_fc_fill_super(struct super_block * sb,struct fs_context * fc)1389 xfs_fc_fill_super(
1390 struct super_block *sb,
1391 struct fs_context *fc)
1392 {
1393 struct xfs_mount *mp = sb->s_fs_info;
1394 struct inode *root;
1395 int flags = 0, error;
1396
1397 mp->m_super = sb;
1398
1399 error = xfs_fc_validate_params(mp);
1400 if (error)
1401 goto out_free_names;
1402
1403 sb_min_blocksize(sb, BBSIZE);
1404 sb->s_xattr = xfs_xattr_handlers;
1405 sb->s_export_op = &xfs_export_operations;
1406 #ifdef CONFIG_XFS_QUOTA
1407 sb->s_qcop = &xfs_quotactl_operations;
1408 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1409 #endif
1410 sb->s_op = &xfs_super_operations;
1411
1412 /*
1413 * Delay mount work if the debug hook is set. This is debug
1414 * instrumention to coordinate simulation of xfs mount failures with
1415 * VFS superblock operations
1416 */
1417 if (xfs_globals.mount_delay) {
1418 xfs_notice(mp, "Delaying mount for %d seconds.",
1419 xfs_globals.mount_delay);
1420 msleep(xfs_globals.mount_delay * 1000);
1421 }
1422
1423 if (fc->sb_flags & SB_SILENT)
1424 flags |= XFS_MFSI_QUIET;
1425
1426 error = xfs_open_devices(mp);
1427 if (error)
1428 goto out_free_names;
1429
1430 error = xfs_init_mount_workqueues(mp);
1431 if (error)
1432 goto out_close_devices;
1433
1434 error = xfs_init_percpu_counters(mp);
1435 if (error)
1436 goto out_destroy_workqueues;
1437
1438 /* Allocate stats memory before we do operations that might use it */
1439 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1440 if (!mp->m_stats.xs_stats) {
1441 error = -ENOMEM;
1442 goto out_destroy_counters;
1443 }
1444
1445 error = xfs_readsb(mp, flags);
1446 if (error)
1447 goto out_free_stats;
1448
1449 error = xfs_finish_flags(mp);
1450 if (error)
1451 goto out_free_sb;
1452
1453 error = xfs_setup_devices(mp);
1454 if (error)
1455 goto out_free_sb;
1456
1457 /* V4 support is undergoing deprecation. */
1458 if (!xfs_sb_version_hascrc(&mp->m_sb)) {
1459 #ifdef CONFIG_XFS_SUPPORT_V4
1460 xfs_warn_once(mp,
1461 "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1462 #else
1463 xfs_warn(mp,
1464 "Deprecated V4 format (crc=0) not supported by kernel.");
1465 error = -EINVAL;
1466 goto out_free_sb;
1467 #endif
1468 }
1469
1470 /*
1471 * XFS block mappings use 54 bits to store the logical block offset.
1472 * This should suffice to handle the maximum file size that the VFS
1473 * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1474 * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1475 * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1476 * to check this assertion.
1477 *
1478 * Avoid integer overflow by comparing the maximum bmbt offset to the
1479 * maximum pagecache offset in units of fs blocks.
1480 */
1481 if (XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE) > XFS_MAX_FILEOFF) {
1482 xfs_warn(mp,
1483 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1484 XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1485 XFS_MAX_FILEOFF);
1486 error = -EINVAL;
1487 goto out_free_sb;
1488 }
1489
1490 error = xfs_filestream_mount(mp);
1491 if (error)
1492 goto out_free_sb;
1493
1494 /*
1495 * we must configure the block size in the superblock before we run the
1496 * full mount process as the mount process can lookup and cache inodes.
1497 */
1498 sb->s_magic = XFS_SUPER_MAGIC;
1499 sb->s_blocksize = mp->m_sb.sb_blocksize;
1500 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1501 sb->s_maxbytes = MAX_LFS_FILESIZE;
1502 sb->s_max_links = XFS_MAXLINK;
1503 sb->s_time_gran = 1;
1504 if (xfs_sb_version_hasbigtime(&mp->m_sb)) {
1505 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1506 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1507 } else {
1508 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1509 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1510 }
1511 trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1512 sb->s_iflags |= SB_I_CGROUPWB;
1513
1514 set_posix_acl_flag(sb);
1515
1516 /* version 5 superblocks support inode version counters. */
1517 if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1518 sb->s_flags |= SB_I_VERSION;
1519
1520 if (xfs_sb_version_hasbigtime(&mp->m_sb))
1521 xfs_warn(mp,
1522 "EXPERIMENTAL big timestamp feature in use. Use at your own risk!");
1523
1524 if (mp->m_flags & XFS_MOUNT_DAX_ALWAYS) {
1525 bool rtdev_is_dax = false, datadev_is_dax;
1526
1527 xfs_warn(mp,
1528 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1529
1530 datadev_is_dax = bdev_dax_supported(mp->m_ddev_targp->bt_bdev,
1531 sb->s_blocksize);
1532 if (mp->m_rtdev_targp)
1533 rtdev_is_dax = bdev_dax_supported(
1534 mp->m_rtdev_targp->bt_bdev, sb->s_blocksize);
1535 if (!rtdev_is_dax && !datadev_is_dax) {
1536 xfs_alert(mp,
1537 "DAX unsupported by block device. Turning off DAX.");
1538 xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
1539 }
1540 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1541 xfs_alert(mp,
1542 "DAX and reflink cannot be used together!");
1543 error = -EINVAL;
1544 goto out_filestream_unmount;
1545 }
1546 }
1547
1548 if (mp->m_flags & XFS_MOUNT_DISCARD) {
1549 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1550
1551 if (!blk_queue_discard(q)) {
1552 xfs_warn(mp, "mounting with \"discard\" option, but "
1553 "the device does not support discard");
1554 mp->m_flags &= ~XFS_MOUNT_DISCARD;
1555 }
1556 }
1557
1558 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1559 if (mp->m_sb.sb_rblocks) {
1560 xfs_alert(mp,
1561 "reflink not compatible with realtime device!");
1562 error = -EINVAL;
1563 goto out_filestream_unmount;
1564 }
1565
1566 if (xfs_globals.always_cow) {
1567 xfs_info(mp, "using DEBUG-only always_cow mode.");
1568 mp->m_always_cow = true;
1569 }
1570 }
1571
1572 if (xfs_sb_version_hasrmapbt(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1573 xfs_alert(mp,
1574 "reverse mapping btree not compatible with realtime device!");
1575 error = -EINVAL;
1576 goto out_filestream_unmount;
1577 }
1578
1579 if (xfs_sb_version_hasinobtcounts(&mp->m_sb))
1580 xfs_warn(mp,
1581 "EXPERIMENTAL inode btree counters feature in use. Use at your own risk!");
1582
1583 error = xfs_mountfs(mp);
1584 if (error)
1585 goto out_filestream_unmount;
1586
1587 root = igrab(VFS_I(mp->m_rootip));
1588 if (!root) {
1589 error = -ENOENT;
1590 goto out_unmount;
1591 }
1592 sb->s_root = d_make_root(root);
1593 if (!sb->s_root) {
1594 error = -ENOMEM;
1595 goto out_unmount;
1596 }
1597
1598 return 0;
1599
1600 out_filestream_unmount:
1601 xfs_filestream_unmount(mp);
1602 out_free_sb:
1603 xfs_freesb(mp);
1604 out_free_stats:
1605 free_percpu(mp->m_stats.xs_stats);
1606 out_destroy_counters:
1607 xfs_destroy_percpu_counters(mp);
1608 out_destroy_workqueues:
1609 xfs_destroy_mount_workqueues(mp);
1610 out_close_devices:
1611 xfs_close_devices(mp);
1612 out_free_names:
1613 sb->s_fs_info = NULL;
1614 xfs_mount_free(mp);
1615 return error;
1616
1617 out_unmount:
1618 xfs_filestream_unmount(mp);
1619 xfs_unmountfs(mp);
1620 goto out_free_sb;
1621 }
1622
1623 static int
xfs_fc_get_tree(struct fs_context * fc)1624 xfs_fc_get_tree(
1625 struct fs_context *fc)
1626 {
1627 return get_tree_bdev(fc, xfs_fc_fill_super);
1628 }
1629
1630 static int
xfs_remount_rw(struct xfs_mount * mp)1631 xfs_remount_rw(
1632 struct xfs_mount *mp)
1633 {
1634 struct xfs_sb *sbp = &mp->m_sb;
1635 int error;
1636
1637 if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1638 xfs_warn(mp,
1639 "ro->rw transition prohibited on norecovery mount");
1640 return -EINVAL;
1641 }
1642
1643 if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1644 xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1645 xfs_warn(mp,
1646 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1647 (sbp->sb_features_ro_compat &
1648 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1649 return -EINVAL;
1650 }
1651
1652 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1653
1654 /*
1655 * If this is the first remount to writeable state we might have some
1656 * superblock changes to update.
1657 */
1658 if (mp->m_update_sb) {
1659 error = xfs_sync_sb(mp, false);
1660 if (error) {
1661 xfs_warn(mp, "failed to write sb changes");
1662 return error;
1663 }
1664 mp->m_update_sb = false;
1665 }
1666
1667 /*
1668 * Fill out the reserve pool if it is empty. Use the stashed value if
1669 * it is non-zero, otherwise go with the default.
1670 */
1671 xfs_restore_resvblks(mp);
1672 xfs_log_work_queue(mp);
1673
1674 /* Recover any CoW blocks that never got remapped. */
1675 error = xfs_reflink_recover_cow(mp);
1676 if (error) {
1677 xfs_err(mp,
1678 "Error %d recovering leftover CoW allocations.", error);
1679 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1680 return error;
1681 }
1682 xfs_start_block_reaping(mp);
1683
1684 /* Create the per-AG metadata reservation pool .*/
1685 error = xfs_fs_reserve_ag_blocks(mp);
1686 if (error && error != -ENOSPC)
1687 return error;
1688
1689 return 0;
1690 }
1691
1692 static int
xfs_remount_ro(struct xfs_mount * mp)1693 xfs_remount_ro(
1694 struct xfs_mount *mp)
1695 {
1696 int error;
1697
1698 /*
1699 * Cancel background eofb scanning so it cannot race with the final
1700 * log force+buftarg wait and deadlock the remount.
1701 */
1702 xfs_stop_block_reaping(mp);
1703
1704 /* Get rid of any leftover CoW reservations... */
1705 error = xfs_icache_free_cowblocks(mp, NULL);
1706 if (error) {
1707 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1708 return error;
1709 }
1710
1711 /* Free the per-AG metadata reservation pool. */
1712 error = xfs_fs_unreserve_ag_blocks(mp);
1713 if (error) {
1714 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1715 return error;
1716 }
1717
1718 /*
1719 * Before we sync the metadata, we need to free up the reserve block
1720 * pool so that the used block count in the superblock on disk is
1721 * correct at the end of the remount. Stash the current* reserve pool
1722 * size so that if we get remounted rw, we can return it to the same
1723 * size.
1724 */
1725 xfs_save_resvblks(mp);
1726
1727 xfs_quiesce_attr(mp);
1728 mp->m_flags |= XFS_MOUNT_RDONLY;
1729
1730 return 0;
1731 }
1732
1733 /*
1734 * Logically we would return an error here to prevent users from believing
1735 * they might have changed mount options using remount which can't be changed.
1736 *
1737 * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1738 * arguments in some cases so we can't blindly reject options, but have to
1739 * check for each specified option if it actually differs from the currently
1740 * set option and only reject it if that's the case.
1741 *
1742 * Until that is implemented we return success for every remount request, and
1743 * silently ignore all options that we can't actually change.
1744 */
1745 static int
xfs_fc_reconfigure(struct fs_context * fc)1746 xfs_fc_reconfigure(
1747 struct fs_context *fc)
1748 {
1749 struct xfs_mount *mp = XFS_M(fc->root->d_sb);
1750 struct xfs_mount *new_mp = fc->s_fs_info;
1751 xfs_sb_t *sbp = &mp->m_sb;
1752 int flags = fc->sb_flags;
1753 int error;
1754
1755 /* version 5 superblocks always support version counters. */
1756 if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1757 fc->sb_flags |= SB_I_VERSION;
1758
1759 error = xfs_fc_validate_params(new_mp);
1760 if (error)
1761 return error;
1762
1763 sync_filesystem(mp->m_super);
1764
1765 /* inode32 -> inode64 */
1766 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1767 !(new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1768 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1769 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1770 }
1771
1772 /* inode64 -> inode32 */
1773 if (!(mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1774 (new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1775 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1776 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1777 }
1778
1779 /* ro -> rw */
1780 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(flags & SB_RDONLY)) {
1781 error = xfs_remount_rw(mp);
1782 if (error)
1783 return error;
1784 }
1785
1786 /* rw -> ro */
1787 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (flags & SB_RDONLY)) {
1788 error = xfs_remount_ro(mp);
1789 if (error)
1790 return error;
1791 }
1792
1793 return 0;
1794 }
1795
xfs_fc_free(struct fs_context * fc)1796 static void xfs_fc_free(
1797 struct fs_context *fc)
1798 {
1799 struct xfs_mount *mp = fc->s_fs_info;
1800
1801 /*
1802 * mp is stored in the fs_context when it is initialized.
1803 * mp is transferred to the superblock on a successful mount,
1804 * but if an error occurs before the transfer we have to free
1805 * it here.
1806 */
1807 if (mp)
1808 xfs_mount_free(mp);
1809 }
1810
1811 static const struct fs_context_operations xfs_context_ops = {
1812 .parse_param = xfs_fc_parse_param,
1813 .get_tree = xfs_fc_get_tree,
1814 .reconfigure = xfs_fc_reconfigure,
1815 .free = xfs_fc_free,
1816 };
1817
xfs_init_fs_context(struct fs_context * fc)1818 static int xfs_init_fs_context(
1819 struct fs_context *fc)
1820 {
1821 struct xfs_mount *mp;
1822
1823 mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1824 if (!mp)
1825 return -ENOMEM;
1826
1827 spin_lock_init(&mp->m_sb_lock);
1828 spin_lock_init(&mp->m_agirotor_lock);
1829 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1830 spin_lock_init(&mp->m_perag_lock);
1831 mutex_init(&mp->m_growlock);
1832 INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1833 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1834 INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1835 INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
1836 mp->m_kobj.kobject.kset = xfs_kset;
1837 /*
1838 * We don't create the finobt per-ag space reservation until after log
1839 * recovery, so we must set this to true so that an ifree transaction
1840 * started during log recovery will not depend on space reservations
1841 * for finobt expansion.
1842 */
1843 mp->m_finobt_nores = true;
1844
1845 /*
1846 * These can be overridden by the mount option parsing.
1847 */
1848 mp->m_logbufs = -1;
1849 mp->m_logbsize = -1;
1850 mp->m_allocsize_log = 16; /* 64k */
1851
1852 /*
1853 * Copy binary VFS mount flags we are interested in.
1854 */
1855 if (fc->sb_flags & SB_RDONLY)
1856 mp->m_flags |= XFS_MOUNT_RDONLY;
1857 if (fc->sb_flags & SB_DIRSYNC)
1858 mp->m_flags |= XFS_MOUNT_DIRSYNC;
1859 if (fc->sb_flags & SB_SYNCHRONOUS)
1860 mp->m_flags |= XFS_MOUNT_WSYNC;
1861
1862 fc->s_fs_info = mp;
1863 fc->ops = &xfs_context_ops;
1864
1865 return 0;
1866 }
1867
1868 static struct file_system_type xfs_fs_type = {
1869 .owner = THIS_MODULE,
1870 .name = "xfs",
1871 .init_fs_context = xfs_init_fs_context,
1872 .parameters = xfs_fs_parameters,
1873 .kill_sb = kill_block_super,
1874 .fs_flags = FS_REQUIRES_DEV,
1875 };
1876 MODULE_ALIAS_FS("xfs");
1877
1878 STATIC int __init
xfs_init_zones(void)1879 xfs_init_zones(void)
1880 {
1881 xfs_log_ticket_zone = kmem_cache_create("xfs_log_ticket",
1882 sizeof(struct xlog_ticket),
1883 0, 0, NULL);
1884 if (!xfs_log_ticket_zone)
1885 goto out;
1886
1887 xfs_bmap_free_item_zone = kmem_cache_create("xfs_bmap_free_item",
1888 sizeof(struct xfs_extent_free_item),
1889 0, 0, NULL);
1890 if (!xfs_bmap_free_item_zone)
1891 goto out_destroy_log_ticket_zone;
1892
1893 xfs_btree_cur_zone = kmem_cache_create("xfs_btree_cur",
1894 sizeof(struct xfs_btree_cur),
1895 0, 0, NULL);
1896 if (!xfs_btree_cur_zone)
1897 goto out_destroy_bmap_free_item_zone;
1898
1899 xfs_da_state_zone = kmem_cache_create("xfs_da_state",
1900 sizeof(struct xfs_da_state),
1901 0, 0, NULL);
1902 if (!xfs_da_state_zone)
1903 goto out_destroy_btree_cur_zone;
1904
1905 xfs_ifork_zone = kmem_cache_create("xfs_ifork",
1906 sizeof(struct xfs_ifork),
1907 0, 0, NULL);
1908 if (!xfs_ifork_zone)
1909 goto out_destroy_da_state_zone;
1910
1911 xfs_trans_zone = kmem_cache_create("xf_trans",
1912 sizeof(struct xfs_trans),
1913 0, 0, NULL);
1914 if (!xfs_trans_zone)
1915 goto out_destroy_ifork_zone;
1916
1917
1918 /*
1919 * The size of the zone allocated buf log item is the maximum
1920 * size possible under XFS. This wastes a little bit of memory,
1921 * but it is much faster.
1922 */
1923 xfs_buf_item_zone = kmem_cache_create("xfs_buf_item",
1924 sizeof(struct xfs_buf_log_item),
1925 0, 0, NULL);
1926 if (!xfs_buf_item_zone)
1927 goto out_destroy_trans_zone;
1928
1929 xfs_efd_zone = kmem_cache_create("xfs_efd_item",
1930 (sizeof(struct xfs_efd_log_item) +
1931 (XFS_EFD_MAX_FAST_EXTENTS - 1) *
1932 sizeof(struct xfs_extent)),
1933 0, 0, NULL);
1934 if (!xfs_efd_zone)
1935 goto out_destroy_buf_item_zone;
1936
1937 xfs_efi_zone = kmem_cache_create("xfs_efi_item",
1938 (sizeof(struct xfs_efi_log_item) +
1939 (XFS_EFI_MAX_FAST_EXTENTS - 1) *
1940 sizeof(struct xfs_extent)),
1941 0, 0, NULL);
1942 if (!xfs_efi_zone)
1943 goto out_destroy_efd_zone;
1944
1945 xfs_inode_zone = kmem_cache_create("xfs_inode",
1946 sizeof(struct xfs_inode), 0,
1947 (SLAB_HWCACHE_ALIGN |
1948 SLAB_RECLAIM_ACCOUNT |
1949 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
1950 xfs_fs_inode_init_once);
1951 if (!xfs_inode_zone)
1952 goto out_destroy_efi_zone;
1953
1954 xfs_ili_zone = kmem_cache_create("xfs_ili",
1955 sizeof(struct xfs_inode_log_item), 0,
1956 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1957 NULL);
1958 if (!xfs_ili_zone)
1959 goto out_destroy_inode_zone;
1960
1961 xfs_icreate_zone = kmem_cache_create("xfs_icr",
1962 sizeof(struct xfs_icreate_item),
1963 0, 0, NULL);
1964 if (!xfs_icreate_zone)
1965 goto out_destroy_ili_zone;
1966
1967 xfs_rud_zone = kmem_cache_create("xfs_rud_item",
1968 sizeof(struct xfs_rud_log_item),
1969 0, 0, NULL);
1970 if (!xfs_rud_zone)
1971 goto out_destroy_icreate_zone;
1972
1973 xfs_rui_zone = kmem_cache_create("xfs_rui_item",
1974 xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
1975 0, 0, NULL);
1976 if (!xfs_rui_zone)
1977 goto out_destroy_rud_zone;
1978
1979 xfs_cud_zone = kmem_cache_create("xfs_cud_item",
1980 sizeof(struct xfs_cud_log_item),
1981 0, 0, NULL);
1982 if (!xfs_cud_zone)
1983 goto out_destroy_rui_zone;
1984
1985 xfs_cui_zone = kmem_cache_create("xfs_cui_item",
1986 xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
1987 0, 0, NULL);
1988 if (!xfs_cui_zone)
1989 goto out_destroy_cud_zone;
1990
1991 xfs_bud_zone = kmem_cache_create("xfs_bud_item",
1992 sizeof(struct xfs_bud_log_item),
1993 0, 0, NULL);
1994 if (!xfs_bud_zone)
1995 goto out_destroy_cui_zone;
1996
1997 xfs_bui_zone = kmem_cache_create("xfs_bui_item",
1998 xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
1999 0, 0, NULL);
2000 if (!xfs_bui_zone)
2001 goto out_destroy_bud_zone;
2002
2003 return 0;
2004
2005 out_destroy_bud_zone:
2006 kmem_cache_destroy(xfs_bud_zone);
2007 out_destroy_cui_zone:
2008 kmem_cache_destroy(xfs_cui_zone);
2009 out_destroy_cud_zone:
2010 kmem_cache_destroy(xfs_cud_zone);
2011 out_destroy_rui_zone:
2012 kmem_cache_destroy(xfs_rui_zone);
2013 out_destroy_rud_zone:
2014 kmem_cache_destroy(xfs_rud_zone);
2015 out_destroy_icreate_zone:
2016 kmem_cache_destroy(xfs_icreate_zone);
2017 out_destroy_ili_zone:
2018 kmem_cache_destroy(xfs_ili_zone);
2019 out_destroy_inode_zone:
2020 kmem_cache_destroy(xfs_inode_zone);
2021 out_destroy_efi_zone:
2022 kmem_cache_destroy(xfs_efi_zone);
2023 out_destroy_efd_zone:
2024 kmem_cache_destroy(xfs_efd_zone);
2025 out_destroy_buf_item_zone:
2026 kmem_cache_destroy(xfs_buf_item_zone);
2027 out_destroy_trans_zone:
2028 kmem_cache_destroy(xfs_trans_zone);
2029 out_destroy_ifork_zone:
2030 kmem_cache_destroy(xfs_ifork_zone);
2031 out_destroy_da_state_zone:
2032 kmem_cache_destroy(xfs_da_state_zone);
2033 out_destroy_btree_cur_zone:
2034 kmem_cache_destroy(xfs_btree_cur_zone);
2035 out_destroy_bmap_free_item_zone:
2036 kmem_cache_destroy(xfs_bmap_free_item_zone);
2037 out_destroy_log_ticket_zone:
2038 kmem_cache_destroy(xfs_log_ticket_zone);
2039 out:
2040 return -ENOMEM;
2041 }
2042
2043 STATIC void
xfs_destroy_zones(void)2044 xfs_destroy_zones(void)
2045 {
2046 /*
2047 * Make sure all delayed rcu free are flushed before we
2048 * destroy caches.
2049 */
2050 rcu_barrier();
2051 kmem_cache_destroy(xfs_bui_zone);
2052 kmem_cache_destroy(xfs_bud_zone);
2053 kmem_cache_destroy(xfs_cui_zone);
2054 kmem_cache_destroy(xfs_cud_zone);
2055 kmem_cache_destroy(xfs_rui_zone);
2056 kmem_cache_destroy(xfs_rud_zone);
2057 kmem_cache_destroy(xfs_icreate_zone);
2058 kmem_cache_destroy(xfs_ili_zone);
2059 kmem_cache_destroy(xfs_inode_zone);
2060 kmem_cache_destroy(xfs_efi_zone);
2061 kmem_cache_destroy(xfs_efd_zone);
2062 kmem_cache_destroy(xfs_buf_item_zone);
2063 kmem_cache_destroy(xfs_trans_zone);
2064 kmem_cache_destroy(xfs_ifork_zone);
2065 kmem_cache_destroy(xfs_da_state_zone);
2066 kmem_cache_destroy(xfs_btree_cur_zone);
2067 kmem_cache_destroy(xfs_bmap_free_item_zone);
2068 kmem_cache_destroy(xfs_log_ticket_zone);
2069 }
2070
2071 STATIC int __init
xfs_init_workqueues(void)2072 xfs_init_workqueues(void)
2073 {
2074 /*
2075 * The allocation workqueue can be used in memory reclaim situations
2076 * (writepage path), and parallelism is only limited by the number of
2077 * AGs in all the filesystems mounted. Hence use the default large
2078 * max_active value for this workqueue.
2079 */
2080 xfs_alloc_wq = alloc_workqueue("xfsalloc",
2081 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
2082 if (!xfs_alloc_wq)
2083 return -ENOMEM;
2084
2085 xfs_discard_wq = alloc_workqueue("xfsdiscard", WQ_UNBOUND, 0);
2086 if (!xfs_discard_wq)
2087 goto out_free_alloc_wq;
2088
2089 return 0;
2090 out_free_alloc_wq:
2091 destroy_workqueue(xfs_alloc_wq);
2092 return -ENOMEM;
2093 }
2094
2095 STATIC void
xfs_destroy_workqueues(void)2096 xfs_destroy_workqueues(void)
2097 {
2098 destroy_workqueue(xfs_discard_wq);
2099 destroy_workqueue(xfs_alloc_wq);
2100 }
2101
2102 STATIC int __init
init_xfs_fs(void)2103 init_xfs_fs(void)
2104 {
2105 int error;
2106
2107 xfs_check_ondisk_structs();
2108
2109 printk(KERN_INFO XFS_VERSION_STRING " with "
2110 XFS_BUILD_OPTIONS " enabled\n");
2111
2112 xfs_dir_startup();
2113
2114 error = xfs_init_zones();
2115 if (error)
2116 goto out;
2117
2118 error = xfs_init_workqueues();
2119 if (error)
2120 goto out_destroy_zones;
2121
2122 error = xfs_mru_cache_init();
2123 if (error)
2124 goto out_destroy_wq;
2125
2126 error = xfs_buf_init();
2127 if (error)
2128 goto out_mru_cache_uninit;
2129
2130 error = xfs_init_procfs();
2131 if (error)
2132 goto out_buf_terminate;
2133
2134 error = xfs_sysctl_register();
2135 if (error)
2136 goto out_cleanup_procfs;
2137
2138 xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2139 if (!xfs_kset) {
2140 error = -ENOMEM;
2141 goto out_sysctl_unregister;
2142 }
2143
2144 xfsstats.xs_kobj.kobject.kset = xfs_kset;
2145
2146 xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2147 if (!xfsstats.xs_stats) {
2148 error = -ENOMEM;
2149 goto out_kset_unregister;
2150 }
2151
2152 error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2153 "stats");
2154 if (error)
2155 goto out_free_stats;
2156
2157 #ifdef DEBUG
2158 xfs_dbg_kobj.kobject.kset = xfs_kset;
2159 error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2160 if (error)
2161 goto out_remove_stats_kobj;
2162 #endif
2163
2164 error = xfs_qm_init();
2165 if (error)
2166 goto out_remove_dbg_kobj;
2167
2168 error = register_filesystem(&xfs_fs_type);
2169 if (error)
2170 goto out_qm_exit;
2171 return 0;
2172
2173 out_qm_exit:
2174 xfs_qm_exit();
2175 out_remove_dbg_kobj:
2176 #ifdef DEBUG
2177 xfs_sysfs_del(&xfs_dbg_kobj);
2178 out_remove_stats_kobj:
2179 #endif
2180 xfs_sysfs_del(&xfsstats.xs_kobj);
2181 out_free_stats:
2182 free_percpu(xfsstats.xs_stats);
2183 out_kset_unregister:
2184 kset_unregister(xfs_kset);
2185 out_sysctl_unregister:
2186 xfs_sysctl_unregister();
2187 out_cleanup_procfs:
2188 xfs_cleanup_procfs();
2189 out_buf_terminate:
2190 xfs_buf_terminate();
2191 out_mru_cache_uninit:
2192 xfs_mru_cache_uninit();
2193 out_destroy_wq:
2194 xfs_destroy_workqueues();
2195 out_destroy_zones:
2196 xfs_destroy_zones();
2197 out:
2198 return error;
2199 }
2200
2201 STATIC void __exit
exit_xfs_fs(void)2202 exit_xfs_fs(void)
2203 {
2204 xfs_qm_exit();
2205 unregister_filesystem(&xfs_fs_type);
2206 #ifdef DEBUG
2207 xfs_sysfs_del(&xfs_dbg_kobj);
2208 #endif
2209 xfs_sysfs_del(&xfsstats.xs_kobj);
2210 free_percpu(xfsstats.xs_stats);
2211 kset_unregister(xfs_kset);
2212 xfs_sysctl_unregister();
2213 xfs_cleanup_procfs();
2214 xfs_buf_terminate();
2215 xfs_mru_cache_uninit();
2216 xfs_destroy_workqueues();
2217 xfs_destroy_zones();
2218 xfs_uuid_table_free();
2219 }
2220
2221 module_init(init_xfs_fs);
2222 module_exit(exit_xfs_fs);
2223
2224 MODULE_AUTHOR("Silicon Graphics, Inc.");
2225 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2226 MODULE_LICENSE("GPL");
2227