?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
WIP FPC-III support
[linux/fpc-iii.git] / fs / xfs / xfs_super.c
blob813be879a5e51d928f8b10523b41f835c308c523
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
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
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) {
203 if (mp->m_qflags & XFS_UQUOTA_ENFD)
204 seq_puts(m, ",usrquota");
205 else
206 seq_puts(m, ",uqnoenforce");
207 }
208
209 if (mp->m_qflags & XFS_PQUOTA_ACCT) {
210 if (mp->m_qflags & XFS_PQUOTA_ENFD)
211 seq_puts(m, ",prjquota");
212 else
213 seq_puts(m, ",pqnoenforce");
214 }
215 if (mp->m_qflags & XFS_GQUOTA_ACCT) {
216 if (mp->m_qflags & XFS_GQUOTA_ENFD)
217 seq_puts(m, ",grpquota");
218 else
219 seq_puts(m, ",gqnoenforce");
220 }
221
222 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
223 seq_puts(m, ",noquota");
224
225 return 0;
226 }
227
228 /*
229 * Set parameters for inode allocation heuristics, taking into account
230 * filesystem size and inode32/inode64 mount options; i.e. specifically
231 * whether or not XFS_MOUNT_SMALL_INUMS is set.
232 *
233 * Inode allocation patterns are altered only if inode32 is requested
234 * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
235 * If altered, XFS_MOUNT_32BITINODES is set as well.
236 *
237 * An agcount independent of that in the mount structure is provided
238 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
239 * to the potentially higher ag count.
240 *
241 * Returns the maximum AG index which may contain inodes.
242 */
243 xfs_agnumber_t
244 xfs_set_inode_alloc(
245 struct xfs_mount *mp,
246 xfs_agnumber_t agcount)
247 {
248 xfs_agnumber_t index;
249 xfs_agnumber_t maxagi = 0;
250 xfs_sb_t *sbp = &mp->m_sb;
251 xfs_agnumber_t max_metadata;
252 xfs_agino_t agino;
253 xfs_ino_t ino;
254
255 /*
256 * Calculate how much should be reserved for inodes to meet
257 * the max inode percentage. Used only for inode32.
258 */
259 if (M_IGEO(mp)->maxicount) {
260 uint64_t icount;
261
262 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
263 do_div(icount, 100);
264 icount += sbp->sb_agblocks - 1;
265 do_div(icount, sbp->sb_agblocks);
266 max_metadata = icount;
267 } else {
268 max_metadata = agcount;
269 }
270
271 /* Get the last possible inode in the filesystem */
272 agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
273 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
274
275 /*
276 * If user asked for no more than 32-bit inodes, and the fs is
277 * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
278 * the allocator to accommodate the request.
279 */
280 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
281 mp->m_flags |= XFS_MOUNT_32BITINODES;
282 else
283 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
284
285 for (index = 0; index < agcount; index++) {
286 struct xfs_perag *pag;
287
288 ino = XFS_AGINO_TO_INO(mp, index, agino);
289
290 pag = xfs_perag_get(mp, index);
291
292 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
293 if (ino > XFS_MAXINUMBER_32) {
294 pag->pagi_inodeok = 0;
295 pag->pagf_metadata = 0;
296 } else {
297 pag->pagi_inodeok = 1;
298 maxagi++;
299 if (index < max_metadata)
300 pag->pagf_metadata = 1;
301 else
302 pag->pagf_metadata = 0;
303 }
304 } else {
305 pag->pagi_inodeok = 1;
306 pag->pagf_metadata = 0;
307 }
308
309 xfs_perag_put(pag);
310 }
311
312 return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
313 }
314
315 STATIC int
316 xfs_blkdev_get(
317 xfs_mount_t *mp,
318 const char *name,
319 struct block_device **bdevp)
320 {
321 int error = 0;
322
323 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
324 mp);
325 if (IS_ERR(*bdevp)) {
326 error = PTR_ERR(*bdevp);
327 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
328 }
329
330 return error;
331 }
332
333 STATIC void
334 xfs_blkdev_put(
335 struct block_device *bdev)
336 {
337 if (bdev)
338 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
339 }
340
341 void
342 xfs_blkdev_issue_flush(
343 xfs_buftarg_t *buftarg)
344 {
345 blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS);
346 }
347
348 STATIC void
349 xfs_close_devices(
350 struct xfs_mount *mp)
351 {
352 struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
353
354 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
355 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
356 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
357
358 xfs_free_buftarg(mp->m_logdev_targp);
359 xfs_blkdev_put(logdev);
360 fs_put_dax(dax_logdev);
361 }
362 if (mp->m_rtdev_targp) {
363 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
364 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
365
366 xfs_free_buftarg(mp->m_rtdev_targp);
367 xfs_blkdev_put(rtdev);
368 fs_put_dax(dax_rtdev);
369 }
370 xfs_free_buftarg(mp->m_ddev_targp);
371 fs_put_dax(dax_ddev);
372 }
373
374 /*
375 * The file system configurations are:
376 * (1) device (partition) with data and internal log
377 * (2) logical volume with data and log subvolumes.
378 * (3) logical volume with data, log, and realtime subvolumes.
379 *
380 * We only have to handle opening the log and realtime volumes here if
381 * they are present. The data subvolume has already been opened by
382 * get_sb_bdev() and is stored in sb->s_bdev.
383 */
384 STATIC int
385 xfs_open_devices(
386 struct xfs_mount *mp)
387 {
388 struct block_device *ddev = mp->m_super->s_bdev;
389 struct dax_device *dax_ddev = fs_dax_get_by_bdev(ddev);
390 struct dax_device *dax_logdev = NULL, *dax_rtdev = NULL;
391 struct block_device *logdev = NULL, *rtdev = NULL;
392 int error;
393
394 /*
395 * Open real time and log devices - order is important.
396 */
397 if (mp->m_logname) {
398 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
399 if (error)
400 goto out;
401 dax_logdev = fs_dax_get_by_bdev(logdev);
402 }
403
404 if (mp->m_rtname) {
405 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
406 if (error)
407 goto out_close_logdev;
408
409 if (rtdev == ddev || rtdev == logdev) {
410 xfs_warn(mp,
411 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
412 error = -EINVAL;
413 goto out_close_rtdev;
414 }
415 dax_rtdev = fs_dax_get_by_bdev(rtdev);
416 }
417
418 /*
419 * Setup xfs_mount buffer target pointers
420 */
421 error = -ENOMEM;
422 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
423 if (!mp->m_ddev_targp)
424 goto out_close_rtdev;
425
426 if (rtdev) {
427 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
428 if (!mp->m_rtdev_targp)
429 goto out_free_ddev_targ;
430 }
431
432 if (logdev && logdev != ddev) {
433 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
434 if (!mp->m_logdev_targp)
435 goto out_free_rtdev_targ;
436 } else {
437 mp->m_logdev_targp = mp->m_ddev_targp;
438 }
439
440 return 0;
441
442 out_free_rtdev_targ:
443 if (mp->m_rtdev_targp)
444 xfs_free_buftarg(mp->m_rtdev_targp);
445 out_free_ddev_targ:
446 xfs_free_buftarg(mp->m_ddev_targp);
447 out_close_rtdev:
448 xfs_blkdev_put(rtdev);
449 fs_put_dax(dax_rtdev);
450 out_close_logdev:
451 if (logdev && logdev != ddev) {
452 xfs_blkdev_put(logdev);
453 fs_put_dax(dax_logdev);
454 }
455 out:
456 fs_put_dax(dax_ddev);
457 return error;
458 }
459
460 /*
461 * Setup xfs_mount buffer target pointers based on superblock
462 */
463 STATIC int
464 xfs_setup_devices(
465 struct xfs_mount *mp)
466 {
467 int error;
468
469 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
470 if (error)
471 return error;
472
473 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
474 unsigned int log_sector_size = BBSIZE;
475
476 if (xfs_sb_version_hassector(&mp->m_sb))
477 log_sector_size = mp->m_sb.sb_logsectsize;
478 error = xfs_setsize_buftarg(mp->m_logdev_targp,
479 log_sector_size);
480 if (error)
481 return error;
482 }
483 if (mp->m_rtdev_targp) {
484 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
485 mp->m_sb.sb_sectsize);
486 if (error)
487 return error;
488 }
489
490 return 0;
491 }
492
493 STATIC int
494 xfs_init_mount_workqueues(
495 struct xfs_mount *mp)
496 {
497 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
498 WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_super->s_id);
499 if (!mp->m_buf_workqueue)
500 goto out;
501
502 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
503 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_super->s_id);
504 if (!mp->m_unwritten_workqueue)
505 goto out_destroy_buf;
506
507 mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
508 WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND,
509 0, mp->m_super->s_id);
510 if (!mp->m_cil_workqueue)
511 goto out_destroy_unwritten;
512
513 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
514 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_super->s_id);
515 if (!mp->m_reclaim_workqueue)
516 goto out_destroy_cil;
517
518 mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
519 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_super->s_id);
520 if (!mp->m_eofblocks_workqueue)
521 goto out_destroy_reclaim;
522
523 mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s", WQ_FREEZABLE, 0,
524 mp->m_super->s_id);
525 if (!mp->m_sync_workqueue)
526 goto out_destroy_eofb;
527
528 return 0;
529
530 out_destroy_eofb:
531 destroy_workqueue(mp->m_eofblocks_workqueue);
532 out_destroy_reclaim:
533 destroy_workqueue(mp->m_reclaim_workqueue);
534 out_destroy_cil:
535 destroy_workqueue(mp->m_cil_workqueue);
536 out_destroy_unwritten:
537 destroy_workqueue(mp->m_unwritten_workqueue);
538 out_destroy_buf:
539 destroy_workqueue(mp->m_buf_workqueue);
540 out:
541 return -ENOMEM;
542 }
543
544 STATIC void
545 xfs_destroy_mount_workqueues(
546 struct xfs_mount *mp)
547 {
548 destroy_workqueue(mp->m_sync_workqueue);
549 destroy_workqueue(mp->m_eofblocks_workqueue);
550 destroy_workqueue(mp->m_reclaim_workqueue);
551 destroy_workqueue(mp->m_cil_workqueue);
552 destroy_workqueue(mp->m_unwritten_workqueue);
553 destroy_workqueue(mp->m_buf_workqueue);
554 }
555
556 static void
557 xfs_flush_inodes_worker(
558 struct work_struct *work)
559 {
560 struct xfs_mount *mp = container_of(work, struct xfs_mount,
561 m_flush_inodes_work);
562 struct super_block *sb = mp->m_super;
563
564 if (down_read_trylock(&sb->s_umount)) {
565 sync_inodes_sb(sb);
566 up_read(&sb->s_umount);
567 }
568 }
569
570 /*
571 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
572 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
573 * for IO to complete so that we effectively throttle multiple callers to the
574 * rate at which IO is completing.
575 */
576 void
577 xfs_flush_inodes(
578 struct xfs_mount *mp)
579 {
580 /*
581 * If flush_work() returns true then that means we waited for a flush
582 * which was already in progress. Don't bother running another scan.
583 */
584 if (flush_work(&mp->m_flush_inodes_work))
585 return;
586
587 queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
588 flush_work(&mp->m_flush_inodes_work);
589 }
590
591 /* Catch misguided souls that try to use this interface on XFS */
592 STATIC struct inode *
593 xfs_fs_alloc_inode(
594 struct super_block *sb)
595 {
596 BUG();
597 return NULL;
598 }
599
600 #ifdef DEBUG
601 static void
602 xfs_check_delalloc(
603 struct xfs_inode *ip,
604 int whichfork)
605 {
606 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
607 struct xfs_bmbt_irec got;
608 struct xfs_iext_cursor icur;
609
610 if (!ifp || !xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got))
611 return;
612 do {
613 if (isnullstartblock(got.br_startblock)) {
614 xfs_warn(ip->i_mount,
615 "ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]",
616 ip->i_ino,
617 whichfork == XFS_DATA_FORK ? "data" : "cow",
618 got.br_startoff, got.br_blockcount);
619 }
620 } while (xfs_iext_next_extent(ifp, &icur, &got));
621 }
622 #else
623 #define xfs_check_delalloc(ip, whichfork) do { } while (0)
624 #endif
625
626 /*
627 * Now that the generic code is guaranteed not to be accessing
628 * the linux inode, we can inactivate and reclaim the inode.
629 */
630 STATIC void
631 xfs_fs_destroy_inode(
632 struct inode *inode)
633 {
634 struct xfs_inode *ip = XFS_I(inode);
635
636 trace_xfs_destroy_inode(ip);
637
638 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
639 XFS_STATS_INC(ip->i_mount, vn_rele);
640 XFS_STATS_INC(ip->i_mount, vn_remove);
641
642 xfs_inactive(ip);
643
644 if (!XFS_FORCED_SHUTDOWN(ip->i_mount) && ip->i_delayed_blks) {
645 xfs_check_delalloc(ip, XFS_DATA_FORK);
646 xfs_check_delalloc(ip, XFS_COW_FORK);
647 ASSERT(0);
648 }
649
650 XFS_STATS_INC(ip->i_mount, vn_reclaim);
651
652 /*
653 * We should never get here with one of the reclaim flags already set.
654 */
655 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
656 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
657
658 /*
659 * We always use background reclaim here because even if the inode is
660 * clean, it still may be under IO and hence we have wait for IO
661 * completion to occur before we can reclaim the inode. The background
662 * reclaim path handles this more efficiently than we can here, so
663 * simply let background reclaim tear down all inodes.
664 */
665 xfs_inode_set_reclaim_tag(ip);
666 }
667
668 static void
669 xfs_fs_dirty_inode(
670 struct inode *inode,
671 int flag)
672 {
673 struct xfs_inode *ip = XFS_I(inode);
674 struct xfs_mount *mp = ip->i_mount;
675 struct xfs_trans *tp;
676
677 if (!(inode->i_sb->s_flags & SB_LAZYTIME))
678 return;
679 if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
680 return;
681
682 if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
683 return;
684 xfs_ilock(ip, XFS_ILOCK_EXCL);
685 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
686 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
687 xfs_trans_commit(tp);
688 }
689
690 /*
691 * Slab object creation initialisation for the XFS inode.
692 * This covers only the idempotent fields in the XFS inode;
693 * all other fields need to be initialised on allocation
694 * from the slab. This avoids the need to repeatedly initialise
695 * fields in the xfs inode that left in the initialise state
696 * when freeing the inode.
697 */
698 STATIC void
699 xfs_fs_inode_init_once(
700 void *inode)
701 {
702 struct xfs_inode *ip = inode;
703
704 memset(ip, 0, sizeof(struct xfs_inode));
705
706 /* vfs inode */
707 inode_init_once(VFS_I(ip));
708
709 /* xfs inode */
710 atomic_set(&ip->i_pincount, 0);
711 spin_lock_init(&ip->i_flags_lock);
712
713 mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
714 "xfsino", ip->i_ino);
715 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
716 "xfsino", ip->i_ino);
717 }
718
719 /*
720 * We do an unlocked check for XFS_IDONTCACHE here because we are already
721 * serialised against cache hits here via the inode->i_lock and igrab() in
722 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
723 * racing with us, and it avoids needing to grab a spinlock here for every inode
724 * we drop the final reference on.
725 */
726 STATIC int
727 xfs_fs_drop_inode(
728 struct inode *inode)
729 {
730 struct xfs_inode *ip = XFS_I(inode);
731
732 /*
733 * If this unlinked inode is in the middle of recovery, don't
734 * drop the inode just yet; log recovery will take care of
735 * that. See the comment for this inode flag.
736 */
737 if (ip->i_flags & XFS_IRECOVERY) {
738 ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
739 return 0;
740 }
741
742 return generic_drop_inode(inode);
743 }
744
745 static void
746 xfs_mount_free(
747 struct xfs_mount *mp)
748 {
749 kfree(mp->m_rtname);
750 kfree(mp->m_logname);
751 kmem_free(mp);
752 }
753
754 STATIC int
755 xfs_fs_sync_fs(
756 struct super_block *sb,
757 int wait)
758 {
759 struct xfs_mount *mp = XFS_M(sb);
760
761 /*
762 * Doing anything during the async pass would be counterproductive.
763 */
764 if (!wait)
765 return 0;
766
767 xfs_log_force(mp, XFS_LOG_SYNC);
768 if (laptop_mode) {
769 /*
770 * The disk must be active because we're syncing.
771 * We schedule log work now (now that the disk is
772 * active) instead of later (when it might not be).
773 */
774 flush_delayed_work(&mp->m_log->l_work);
775 }
776
777 return 0;
778 }
779
780 STATIC int
781 xfs_fs_statfs(
782 struct dentry *dentry,
783 struct kstatfs *statp)
784 {
785 struct xfs_mount *mp = XFS_M(dentry->d_sb);
786 xfs_sb_t *sbp = &mp->m_sb;
787 struct xfs_inode *ip = XFS_I(d_inode(dentry));
788 uint64_t fakeinos, id;
789 uint64_t icount;
790 uint64_t ifree;
791 uint64_t fdblocks;
792 xfs_extlen_t lsize;
793 int64_t ffree;
794
795 statp->f_type = XFS_SUPER_MAGIC;
796 statp->f_namelen = MAXNAMELEN - 1;
797
798 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
799 statp->f_fsid = u64_to_fsid(id);
800
801 icount = percpu_counter_sum(&mp->m_icount);
802 ifree = percpu_counter_sum(&mp->m_ifree);
803 fdblocks = percpu_counter_sum(&mp->m_fdblocks);
804
805 spin_lock(&mp->m_sb_lock);
806 statp->f_bsize = sbp->sb_blocksize;
807 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
808 statp->f_blocks = sbp->sb_dblocks - lsize;
809 spin_unlock(&mp->m_sb_lock);
810
811 /* make sure statp->f_bfree does not underflow */
812 statp->f_bfree = max_t(int64_t, fdblocks - mp->m_alloc_set_aside, 0);
813 statp->f_bavail = statp->f_bfree;
814
815 fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
816 statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
817 if (M_IGEO(mp)->maxicount)
818 statp->f_files = min_t(typeof(statp->f_files),
819 statp->f_files,
820 M_IGEO(mp)->maxicount);
821
822 /* If sb_icount overshot maxicount, report actual allocation */
823 statp->f_files = max_t(typeof(statp->f_files),
824 statp->f_files,
825 sbp->sb_icount);
826
827 /* make sure statp->f_ffree does not underflow */
828 ffree = statp->f_files - (icount - ifree);
829 statp->f_ffree = max_t(int64_t, ffree, 0);
830
831
832 if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
833 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
834 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
835 xfs_qm_statvfs(ip, statp);
836
837 if (XFS_IS_REALTIME_MOUNT(mp) &&
838 (ip->i_d.di_flags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
839 statp->f_blocks = sbp->sb_rblocks;
840 statp->f_bavail = statp->f_bfree =
841 sbp->sb_frextents * sbp->sb_rextsize;
842 }
843
844 return 0;
845 }
846
847 STATIC void
848 xfs_save_resvblks(struct xfs_mount *mp)
849 {
850 uint64_t resblks = 0;
851
852 mp->m_resblks_save = mp->m_resblks;
853 xfs_reserve_blocks(mp, &resblks, NULL);
854 }
855
856 STATIC void
857 xfs_restore_resvblks(struct xfs_mount *mp)
858 {
859 uint64_t resblks;
860
861 if (mp->m_resblks_save) {
862 resblks = mp->m_resblks_save;
863 mp->m_resblks_save = 0;
864 } else
865 resblks = xfs_default_resblks(mp);
866
867 xfs_reserve_blocks(mp, &resblks, NULL);
868 }
869
870 /*
871 * Trigger writeback of all the dirty metadata in the file system.
872 *
873 * This ensures that the metadata is written to their location on disk rather
874 * than just existing in transactions in the log. This means after a quiesce
875 * there is no log replay required to write the inodes to disk - this is the
876 * primary difference between a sync and a quiesce.
877 *
878 * We cancel log work early here to ensure all transactions the log worker may
879 * run have finished before we clean up and log the superblock and write an
880 * unmount record. The unfreeze process is responsible for restarting the log
881 * worker correctly.
882 */
883 void
884 xfs_quiesce_attr(
885 struct xfs_mount *mp)
886 {
887 int error = 0;
888
889 cancel_delayed_work_sync(&mp->m_log->l_work);
890
891 /* force the log to unpin objects from the now complete transactions */
892 xfs_log_force(mp, XFS_LOG_SYNC);
893
894
895 /* Push the superblock and write an unmount record */
896 error = xfs_log_sbcount(mp);
897 if (error)
898 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
899 "Frozen image may not be consistent.");
900 xfs_log_quiesce(mp);
901 }
902
903 /*
904 * Second stage of a freeze. The data is already frozen so we only
905 * need to take care of the metadata. Once that's done sync the superblock
906 * to the log to dirty it in case of a crash while frozen. This ensures that we
907 * will recover the unlinked inode lists on the next mount.
908 */
909 STATIC int
910 xfs_fs_freeze(
911 struct super_block *sb)
912 {
913 struct xfs_mount *mp = XFS_M(sb);
914 unsigned int flags;
915 int ret;
916
917 /*
918 * The filesystem is now frozen far enough that memory reclaim
919 * cannot safely operate on the filesystem. Hence we need to
920 * set a GFP_NOFS context here to avoid recursion deadlocks.
921 */
922 flags = memalloc_nofs_save();
923 xfs_stop_block_reaping(mp);
924 xfs_save_resvblks(mp);
925 xfs_quiesce_attr(mp);
926 ret = xfs_sync_sb(mp, true);
927 memalloc_nofs_restore(flags);
928 return ret;
929 }
930
931 STATIC int
932 xfs_fs_unfreeze(
933 struct super_block *sb)
934 {
935 struct xfs_mount *mp = XFS_M(sb);
936
937 xfs_restore_resvblks(mp);
938 xfs_log_work_queue(mp);
939 xfs_start_block_reaping(mp);
940 return 0;
941 }
942
943 /*
944 * This function fills in xfs_mount_t fields based on mount args.
945 * Note: the superblock _has_ now been read in.
946 */
947 STATIC int
948 xfs_finish_flags(
949 struct xfs_mount *mp)
950 {
951 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
952
953 /* Fail a mount where the logbuf is smaller than the log stripe */
954 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
955 if (mp->m_logbsize <= 0 &&
956 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
957 mp->m_logbsize = mp->m_sb.sb_logsunit;
958 } else if (mp->m_logbsize > 0 &&
959 mp->m_logbsize < mp->m_sb.sb_logsunit) {
960 xfs_warn(mp,
961 "logbuf size must be greater than or equal to log stripe size");
962 return -EINVAL;
963 }
964 } else {
965 /* Fail a mount if the logbuf is larger than 32K */
966 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
967 xfs_warn(mp,
968 "logbuf size for version 1 logs must be 16K or 32K");
969 return -EINVAL;
970 }
971 }
972
973 /*
974 * V5 filesystems always use attr2 format for attributes.
975 */
976 if (xfs_sb_version_hascrc(&mp->m_sb) &&
977 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
978 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
979 "attr2 is always enabled for V5 filesystems.");
980 return -EINVAL;
981 }
982
983 /*
984 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
985 * told by noattr2 to turn it off
986 */
987 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
988 !(mp->m_flags & XFS_MOUNT_NOATTR2))
989 mp->m_flags |= XFS_MOUNT_ATTR2;
990
991 /*
992 * prohibit r/w mounts of read-only filesystems
993 */
994 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
995 xfs_warn(mp,
996 "cannot mount a read-only filesystem as read-write");
997 return -EROFS;
998 }
999
1000 if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
1001 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
1002 !xfs_sb_version_has_pquotino(&mp->m_sb)) {
1003 xfs_warn(mp,
1004 "Super block does not support project and group quota together");
1005 return -EINVAL;
1006 }
1007
1008 return 0;
1009 }
1010
1011 static int
1012 xfs_init_percpu_counters(
1013 struct xfs_mount *mp)
1014 {
1015 int error;
1016
1017 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1018 if (error)
1019 return -ENOMEM;
1020
1021 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1022 if (error)
1023 goto free_icount;
1024
1025 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1026 if (error)
1027 goto free_ifree;
1028
1029 error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1030 if (error)
1031 goto free_fdblocks;
1032
1033 return 0;
1034
1035 free_fdblocks:
1036 percpu_counter_destroy(&mp->m_fdblocks);
1037 free_ifree:
1038 percpu_counter_destroy(&mp->m_ifree);
1039 free_icount:
1040 percpu_counter_destroy(&mp->m_icount);
1041 return -ENOMEM;
1042 }
1043
1044 void
1045 xfs_reinit_percpu_counters(
1046 struct xfs_mount *mp)
1047 {
1048 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1049 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1050 percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1051 }
1052
1053 static void
1054 xfs_destroy_percpu_counters(
1055 struct xfs_mount *mp)
1056 {
1057 percpu_counter_destroy(&mp->m_icount);
1058 percpu_counter_destroy(&mp->m_ifree);
1059 percpu_counter_destroy(&mp->m_fdblocks);
1060 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1061 percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1062 percpu_counter_destroy(&mp->m_delalloc_blks);
1063 }
1064
1065 static void
1066 xfs_fs_put_super(
1067 struct super_block *sb)
1068 {
1069 struct xfs_mount *mp = XFS_M(sb);
1070
1071 /* if ->fill_super failed, we have no mount to tear down */
1072 if (!sb->s_fs_info)
1073 return;
1074
1075 xfs_notice(mp, "Unmounting Filesystem");
1076 xfs_filestream_unmount(mp);
1077 xfs_unmountfs(mp);
1078
1079 xfs_freesb(mp);
1080 free_percpu(mp->m_stats.xs_stats);
1081 xfs_destroy_percpu_counters(mp);
1082 xfs_destroy_mount_workqueues(mp);
1083 xfs_close_devices(mp);
1084
1085 sb->s_fs_info = NULL;
1086 xfs_mount_free(mp);
1087 }
1088
1089 static long
1090 xfs_fs_nr_cached_objects(
1091 struct super_block *sb,
1092 struct shrink_control *sc)
1093 {
1094 /* Paranoia: catch incorrect calls during mount setup or teardown */
1095 if (WARN_ON_ONCE(!sb->s_fs_info))
1096 return 0;
1097 return xfs_reclaim_inodes_count(XFS_M(sb));
1098 }
1099
1100 static long
1101 xfs_fs_free_cached_objects(
1102 struct super_block *sb,
1103 struct shrink_control *sc)
1104 {
1105 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1106 }
1107
1108 static const struct super_operations xfs_super_operations = {
1109 .alloc_inode = xfs_fs_alloc_inode,
1110 .destroy_inode = xfs_fs_destroy_inode,
1111 .dirty_inode = xfs_fs_dirty_inode,
1112 .drop_inode = xfs_fs_drop_inode,
1113 .put_super = xfs_fs_put_super,
1114 .sync_fs = xfs_fs_sync_fs,
1115 .freeze_fs = xfs_fs_freeze,
1116 .unfreeze_fs = xfs_fs_unfreeze,
1117 .statfs = xfs_fs_statfs,
1118 .show_options = xfs_fs_show_options,
1119 .nr_cached_objects = xfs_fs_nr_cached_objects,
1120 .free_cached_objects = xfs_fs_free_cached_objects,
1121 };
1122
1123 static int
1124 suffix_kstrtoint(
1125 const char *s,
1126 unsigned int base,
1127 int *res)
1128 {
1129 int last, shift_left_factor = 0, _res;
1130 char *value;
1131 int ret = 0;
1132
1133 value = kstrdup(s, GFP_KERNEL);
1134 if (!value)
1135 return -ENOMEM;
1136
1137 last = strlen(value) - 1;
1138 if (value[last] == 'K' || value[last] == 'k') {
1139 shift_left_factor = 10;
1140 value[last] = '\0';
1141 }
1142 if (value[last] == 'M' || value[last] == 'm') {
1143 shift_left_factor = 20;
1144 value[last] = '\0';
1145 }
1146 if (value[last] == 'G' || value[last] == 'g') {
1147 shift_left_factor = 30;
1148 value[last] = '\0';
1149 }
1150
1151 if (kstrtoint(value, base, &_res))
1152 ret = -EINVAL;
1153 kfree(value);
1154 *res = _res << shift_left_factor;
1155 return ret;
1156 }
1157
1158 /*
1159 * Set mount state from a mount option.
1160 *
1161 * NOTE: mp->m_super is NULL here!
1162 */
1163 static int
1164 xfs_fs_parse_param(
1165 struct fs_context *fc,
1166 struct fs_parameter *param)
1167 {
1168 struct xfs_mount *mp = fc->s_fs_info;
1169 struct fs_parse_result result;
1170 int size = 0;
1171 int opt;
1172
1173 opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1174 if (opt < 0)
1175 return opt;
1176
1177 switch (opt) {
1178 case Opt_logbufs:
1179 mp->m_logbufs = result.uint_32;
1180 return 0;
1181 case Opt_logbsize:
1182 if (suffix_kstrtoint(param->string, 10, &mp->m_logbsize))
1183 return -EINVAL;
1184 return 0;
1185 case Opt_logdev:
1186 kfree(mp->m_logname);
1187 mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1188 if (!mp->m_logname)
1189 return -ENOMEM;
1190 return 0;
1191 case Opt_rtdev:
1192 kfree(mp->m_rtname);
1193 mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1194 if (!mp->m_rtname)
1195 return -ENOMEM;
1196 return 0;
1197 case Opt_allocsize:
1198 if (suffix_kstrtoint(param->string, 10, &size))
1199 return -EINVAL;
1200 mp->m_allocsize_log = ffs(size) - 1;
1201 mp->m_flags |= XFS_MOUNT_ALLOCSIZE;
1202 return 0;
1203 case Opt_grpid:
1204 case Opt_bsdgroups:
1205 mp->m_flags |= XFS_MOUNT_GRPID;
1206 return 0;
1207 case Opt_nogrpid:
1208 case Opt_sysvgroups:
1209 mp->m_flags &= ~XFS_MOUNT_GRPID;
1210 return 0;
1211 case Opt_wsync:
1212 mp->m_flags |= XFS_MOUNT_WSYNC;
1213 return 0;
1214 case Opt_norecovery:
1215 mp->m_flags |= XFS_MOUNT_NORECOVERY;
1216 return 0;
1217 case Opt_noalign:
1218 mp->m_flags |= XFS_MOUNT_NOALIGN;
1219 return 0;
1220 case Opt_swalloc:
1221 mp->m_flags |= XFS_MOUNT_SWALLOC;
1222 return 0;
1223 case Opt_sunit:
1224 mp->m_dalign = result.uint_32;
1225 return 0;
1226 case Opt_swidth:
1227 mp->m_swidth = result.uint_32;
1228 return 0;
1229 case Opt_inode32:
1230 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1231 return 0;
1232 case Opt_inode64:
1233 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1234 return 0;
1235 case Opt_nouuid:
1236 mp->m_flags |= XFS_MOUNT_NOUUID;
1237 return 0;
1238 case Opt_largeio:
1239 mp->m_flags |= XFS_MOUNT_LARGEIO;
1240 return 0;
1241 case Opt_nolargeio:
1242 mp->m_flags &= ~XFS_MOUNT_LARGEIO;
1243 return 0;
1244 case Opt_filestreams:
1245 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
1246 return 0;
1247 case Opt_noquota:
1248 mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1249 mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1250 mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
1251 return 0;
1252 case Opt_quota:
1253 case Opt_uquota:
1254 case Opt_usrquota:
1255 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
1256 XFS_UQUOTA_ENFD);
1257 return 0;
1258 case Opt_qnoenforce:
1259 case Opt_uqnoenforce:
1260 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
1261 mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1262 return 0;
1263 case Opt_pquota:
1264 case Opt_prjquota:
1265 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
1266 XFS_PQUOTA_ENFD);
1267 return 0;
1268 case Opt_pqnoenforce:
1269 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
1270 mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1271 return 0;
1272 case Opt_gquota:
1273 case Opt_grpquota:
1274 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
1275 XFS_GQUOTA_ENFD);
1276 return 0;
1277 case Opt_gqnoenforce:
1278 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
1279 mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1280 return 0;
1281 case Opt_discard:
1282 mp->m_flags |= XFS_MOUNT_DISCARD;
1283 return 0;
1284 case Opt_nodiscard:
1285 mp->m_flags &= ~XFS_MOUNT_DISCARD;
1286 return 0;
1287 #ifdef CONFIG_FS_DAX
1288 case Opt_dax:
1289 xfs_mount_set_dax_mode(mp, XFS_DAX_ALWAYS);
1290 return 0;
1291 case Opt_dax_enum:
1292 xfs_mount_set_dax_mode(mp, result.uint_32);
1293 return 0;
1294 #endif
1295 /* Following mount options will be removed in September 2025 */
1296 case Opt_ikeep:
1297 xfs_warn(mp, "%s mount option is deprecated.", param->key);
1298 mp->m_flags |= XFS_MOUNT_IKEEP;
1299 return 0;
1300 case Opt_noikeep:
1301 xfs_warn(mp, "%s mount option is deprecated.", param->key);
1302 mp->m_flags &= ~XFS_MOUNT_IKEEP;
1303 return 0;
1304 case Opt_attr2:
1305 xfs_warn(mp, "%s mount option is deprecated.", param->key);
1306 mp->m_flags |= XFS_MOUNT_ATTR2;
1307 return 0;
1308 case Opt_noattr2:
1309 xfs_warn(mp, "%s mount option is deprecated.", param->key);
1310 mp->m_flags &= ~XFS_MOUNT_ATTR2;
1311 mp->m_flags |= XFS_MOUNT_NOATTR2;
1312 return 0;
1313 default:
1314 xfs_warn(mp, "unknown mount option [%s].", param->key);
1315 return -EINVAL;
1316 }
1317
1318 return 0;
1319 }
1320
1321 static int
1322 xfs_fs_validate_params(
1323 struct xfs_mount *mp)
1324 {
1325 /*
1326 * no recovery flag requires a read-only mount
1327 */
1328 if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
1329 !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1330 xfs_warn(mp, "no-recovery mounts must be read-only.");
1331 return -EINVAL;
1332 }
1333
1334 if ((mp->m_flags & XFS_MOUNT_NOALIGN) &&
1335 (mp->m_dalign || mp->m_swidth)) {
1336 xfs_warn(mp,
1337 "sunit and swidth options incompatible with the noalign option");
1338 return -EINVAL;
1339 }
1340
1341 if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1342 xfs_warn(mp, "quota support not available in this kernel.");
1343 return -EINVAL;
1344 }
1345
1346 if ((mp->m_dalign && !mp->m_swidth) ||
1347 (!mp->m_dalign && mp->m_swidth)) {
1348 xfs_warn(mp, "sunit and swidth must be specified together");
1349 return -EINVAL;
1350 }
1351
1352 if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1353 xfs_warn(mp,
1354 "stripe width (%d) must be a multiple of the stripe unit (%d)",
1355 mp->m_swidth, mp->m_dalign);
1356 return -EINVAL;
1357 }
1358
1359 if (mp->m_logbufs != -1 &&
1360 mp->m_logbufs != 0 &&
1361 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1362 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1363 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1364 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1365 return -EINVAL;
1366 }
1367
1368 if (mp->m_logbsize != -1 &&
1369 mp->m_logbsize != 0 &&
1370 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1371 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1372 !is_power_of_2(mp->m_logbsize))) {
1373 xfs_warn(mp,
1374 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1375 mp->m_logbsize);
1376 return -EINVAL;
1377 }
1378
1379 if ((mp->m_flags & XFS_MOUNT_ALLOCSIZE) &&
1380 (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1381 mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1382 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1383 mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1384 return -EINVAL;
1385 }
1386
1387 return 0;
1388 }
1389
1390 static int
1391 xfs_fs_fill_super(
1392 struct super_block *sb,
1393 struct fs_context *fc)
1394 {
1395 struct xfs_mount *mp = sb->s_fs_info;
1396 struct inode *root;
1397 int flags = 0, error;
1398
1399 mp->m_super = sb;
1400
1401 error = xfs_fs_validate_params(mp);
1402 if (error)
1403 goto out_free_names;
1404
1405 sb_min_blocksize(sb, BBSIZE);
1406 sb->s_xattr = xfs_xattr_handlers;
1407 sb->s_export_op = &xfs_export_operations;
1408 #ifdef CONFIG_XFS_QUOTA
1409 sb->s_qcop = &xfs_quotactl_operations;
1410 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1411 #endif
1412 sb->s_op = &xfs_super_operations;
1413
1414 /*
1415 * Delay mount work if the debug hook is set. This is debug
1416 * instrumention to coordinate simulation of xfs mount failures with
1417 * VFS superblock operations
1418 */
1419 if (xfs_globals.mount_delay) {
1420 xfs_notice(mp, "Delaying mount for %d seconds.",
1421 xfs_globals.mount_delay);
1422 msleep(xfs_globals.mount_delay * 1000);
1423 }
1424
1425 if (fc->sb_flags & SB_SILENT)
1426 flags |= XFS_MFSI_QUIET;
1427
1428 error = xfs_open_devices(mp);
1429 if (error)
1430 goto out_free_names;
1431
1432 error = xfs_init_mount_workqueues(mp);
1433 if (error)
1434 goto out_close_devices;
1435
1436 error = xfs_init_percpu_counters(mp);
1437 if (error)
1438 goto out_destroy_workqueues;
1439
1440 /* Allocate stats memory before we do operations that might use it */
1441 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1442 if (!mp->m_stats.xs_stats) {
1443 error = -ENOMEM;
1444 goto out_destroy_counters;
1445 }
1446
1447 error = xfs_readsb(mp, flags);
1448 if (error)
1449 goto out_free_stats;
1450
1451 error = xfs_finish_flags(mp);
1452 if (error)
1453 goto out_free_sb;
1454
1455 error = xfs_setup_devices(mp);
1456 if (error)
1457 goto out_free_sb;
1458
1459 /* V4 support is undergoing deprecation. */
1460 if (!xfs_sb_version_hascrc(&mp->m_sb)) {
1461 #ifdef CONFIG_XFS_SUPPORT_V4
1462 xfs_warn_once(mp,
1463 "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1464 #else
1465 xfs_warn(mp,
1466 "Deprecated V4 format (crc=0) not supported by kernel.");
1467 error = -EINVAL;
1468 goto out_free_sb;
1469 #endif
1470 }
1471
1472 /* Filesystem claims it needs repair, so refuse the mount. */
1473 if (xfs_sb_version_needsrepair(&mp->m_sb)) {
1474 xfs_warn(mp, "Filesystem needs repair. Please run xfs_repair.");
1475 error = -EFSCORRUPTED;
1476 goto out_free_sb;
1477 }
1478
1479 /*
1480 * Don't touch the filesystem if a user tool thinks it owns the primary
1481 * superblock. mkfs doesn't clear the flag from secondary supers, so
1482 * we don't check them at all.
1483 */
1484 if (mp->m_sb.sb_inprogress) {
1485 xfs_warn(mp, "Offline file system operation in progress!");
1486 error = -EFSCORRUPTED;
1487 goto out_free_sb;
1488 }
1489
1490 /*
1491 * Until this is fixed only page-sized or smaller data blocks work.
1492 */
1493 if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1494 xfs_warn(mp,
1495 "File system with blocksize %d bytes. "
1496 "Only pagesize (%ld) or less will currently work.",
1497 mp->m_sb.sb_blocksize, PAGE_SIZE);
1498 error = -ENOSYS;
1499 goto out_free_sb;
1500 }
1501
1502 /* Ensure this filesystem fits in the page cache limits */
1503 if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1504 xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1505 xfs_warn(mp,
1506 "file system too large to be mounted on this system.");
1507 error = -EFBIG;
1508 goto out_free_sb;
1509 }
1510
1511 /*
1512 * XFS block mappings use 54 bits to store the logical block offset.
1513 * This should suffice to handle the maximum file size that the VFS
1514 * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1515 * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1516 * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1517 * to check this assertion.
1518 *
1519 * Avoid integer overflow by comparing the maximum bmbt offset to the
1520 * maximum pagecache offset in units of fs blocks.
1521 */
1522 if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1523 xfs_warn(mp,
1524 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1525 XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1526 XFS_MAX_FILEOFF);
1527 error = -EINVAL;
1528 goto out_free_sb;
1529 }
1530
1531 error = xfs_filestream_mount(mp);
1532 if (error)
1533 goto out_free_sb;
1534
1535 /*
1536 * we must configure the block size in the superblock before we run the
1537 * full mount process as the mount process can lookup and cache inodes.
1538 */
1539 sb->s_magic = XFS_SUPER_MAGIC;
1540 sb->s_blocksize = mp->m_sb.sb_blocksize;
1541 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1542 sb->s_maxbytes = MAX_LFS_FILESIZE;
1543 sb->s_max_links = XFS_MAXLINK;
1544 sb->s_time_gran = 1;
1545 if (xfs_sb_version_hasbigtime(&mp->m_sb)) {
1546 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1547 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1548 } else {
1549 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1550 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1551 }
1552 trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1553 sb->s_iflags |= SB_I_CGROUPWB;
1554
1555 set_posix_acl_flag(sb);
1556
1557 /* version 5 superblocks support inode version counters. */
1558 if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1559 sb->s_flags |= SB_I_VERSION;
1560
1561 if (xfs_sb_version_hasbigtime(&mp->m_sb))
1562 xfs_warn(mp,
1563 "EXPERIMENTAL big timestamp feature in use. Use at your own risk!");
1564
1565 if (mp->m_flags & XFS_MOUNT_DAX_ALWAYS) {
1566 bool rtdev_is_dax = false, datadev_is_dax;
1567
1568 xfs_warn(mp,
1569 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1570
1571 datadev_is_dax = bdev_dax_supported(mp->m_ddev_targp->bt_bdev,
1572 sb->s_blocksize);
1573 if (mp->m_rtdev_targp)
1574 rtdev_is_dax = bdev_dax_supported(
1575 mp->m_rtdev_targp->bt_bdev, sb->s_blocksize);
1576 if (!rtdev_is_dax && !datadev_is_dax) {
1577 xfs_alert(mp,
1578 "DAX unsupported by block device. Turning off DAX.");
1579 xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
1580 }
1581 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1582 xfs_alert(mp,
1583 "DAX and reflink cannot be used together!");
1584 error = -EINVAL;
1585 goto out_filestream_unmount;
1586 }
1587 }
1588
1589 if (mp->m_flags & XFS_MOUNT_DISCARD) {
1590 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1591
1592 if (!blk_queue_discard(q)) {
1593 xfs_warn(mp, "mounting with \"discard\" option, but "
1594 "the device does not support discard");
1595 mp->m_flags &= ~XFS_MOUNT_DISCARD;
1596 }
1597 }
1598
1599 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1600 if (mp->m_sb.sb_rblocks) {
1601 xfs_alert(mp,
1602 "reflink not compatible with realtime device!");
1603 error = -EINVAL;
1604 goto out_filestream_unmount;
1605 }
1606
1607 if (xfs_globals.always_cow) {
1608 xfs_info(mp, "using DEBUG-only always_cow mode.");
1609 mp->m_always_cow = true;
1610 }
1611 }
1612
1613 if (xfs_sb_version_hasrmapbt(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1614 xfs_alert(mp,
1615 "reverse mapping btree not compatible with realtime device!");
1616 error = -EINVAL;
1617 goto out_filestream_unmount;
1618 }
1619
1620 if (xfs_sb_version_hasinobtcounts(&mp->m_sb))
1621 xfs_warn(mp,
1622 "EXPERIMENTAL inode btree counters feature in use. Use at your own risk!");
1623
1624 error = xfs_mountfs(mp);
1625 if (error)
1626 goto out_filestream_unmount;
1627
1628 root = igrab(VFS_I(mp->m_rootip));
1629 if (!root) {
1630 error = -ENOENT;
1631 goto out_unmount;
1632 }
1633 sb->s_root = d_make_root(root);
1634 if (!sb->s_root) {
1635 error = -ENOMEM;
1636 goto out_unmount;
1637 }
1638
1639 return 0;
1640
1641 out_filestream_unmount:
1642 xfs_filestream_unmount(mp);
1643 out_free_sb:
1644 xfs_freesb(mp);
1645 out_free_stats:
1646 free_percpu(mp->m_stats.xs_stats);
1647 out_destroy_counters:
1648 xfs_destroy_percpu_counters(mp);
1649 out_destroy_workqueues:
1650 xfs_destroy_mount_workqueues(mp);
1651 out_close_devices:
1652 xfs_close_devices(mp);
1653 out_free_names:
1654 sb->s_fs_info = NULL;
1655 xfs_mount_free(mp);
1656 return error;
1657
1658 out_unmount:
1659 xfs_filestream_unmount(mp);
1660 xfs_unmountfs(mp);
1661 goto out_free_sb;
1662 }
1663
1664 static int
1665 xfs_fs_get_tree(
1666 struct fs_context *fc)
1667 {
1668 return get_tree_bdev(fc, xfs_fs_fill_super);
1669 }
1670
1671 static int
1672 xfs_remount_rw(
1673 struct xfs_mount *mp)
1674 {
1675 struct xfs_sb *sbp = &mp->m_sb;
1676 int error;
1677
1678 if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1679 xfs_warn(mp,
1680 "ro->rw transition prohibited on norecovery mount");
1681 return -EINVAL;
1682 }
1683
1684 if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1685 xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1686 xfs_warn(mp,
1687 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1688 (sbp->sb_features_ro_compat &
1689 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1690 return -EINVAL;
1691 }
1692
1693 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1694
1695 /*
1696 * If this is the first remount to writeable state we might have some
1697 * superblock changes to update.
1698 */
1699 if (mp->m_update_sb) {
1700 error = xfs_sync_sb(mp, false);
1701 if (error) {
1702 xfs_warn(mp, "failed to write sb changes");
1703 return error;
1704 }
1705 mp->m_update_sb = false;
1706 }
1707
1708 /*
1709 * Fill out the reserve pool if it is empty. Use the stashed value if
1710 * it is non-zero, otherwise go with the default.
1711 */
1712 xfs_restore_resvblks(mp);
1713 xfs_log_work_queue(mp);
1714
1715 /* Recover any CoW blocks that never got remapped. */
1716 error = xfs_reflink_recover_cow(mp);
1717 if (error) {
1718 xfs_err(mp,
1719 "Error %d recovering leftover CoW allocations.", error);
1720 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1721 return error;
1722 }
1723 xfs_start_block_reaping(mp);
1724
1725 /* Create the per-AG metadata reservation pool .*/
1726 error = xfs_fs_reserve_ag_blocks(mp);
1727 if (error && error != -ENOSPC)
1728 return error;
1729
1730 return 0;
1731 }
1732
1733 static int
1734 xfs_remount_ro(
1735 struct xfs_mount *mp)
1736 {
1737 int error;
1738
1739 /*
1740 * Cancel background eofb scanning so it cannot race with the final
1741 * log force+buftarg wait and deadlock the remount.
1742 */
1743 xfs_stop_block_reaping(mp);
1744
1745 /* Get rid of any leftover CoW reservations... */
1746 error = xfs_icache_free_cowblocks(mp, NULL);
1747 if (error) {
1748 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1749 return error;
1750 }
1751
1752 /* Free the per-AG metadata reservation pool. */
1753 error = xfs_fs_unreserve_ag_blocks(mp);
1754 if (error) {
1755 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1756 return error;
1757 }
1758
1759 /*
1760 * Before we sync the metadata, we need to free up the reserve block
1761 * pool so that the used block count in the superblock on disk is
1762 * correct at the end of the remount. Stash the current* reserve pool
1763 * size so that if we get remounted rw, we can return it to the same
1764 * size.
1765 */
1766 xfs_save_resvblks(mp);
1767
1768 xfs_quiesce_attr(mp);
1769 mp->m_flags |= XFS_MOUNT_RDONLY;
1770
1771 return 0;
1772 }
1773
1774 /*
1775 * Logically we would return an error here to prevent users from believing
1776 * they might have changed mount options using remount which can't be changed.
1777 *
1778 * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1779 * arguments in some cases so we can't blindly reject options, but have to
1780 * check for each specified option if it actually differs from the currently
1781 * set option and only reject it if that's the case.
1782 *
1783 * Until that is implemented we return success for every remount request, and
1784 * silently ignore all options that we can't actually change.
1785 */
1786 static int
1787 xfs_fs_reconfigure(
1788 struct fs_context *fc)
1789 {
1790 struct xfs_mount *mp = XFS_M(fc->root->d_sb);
1791 struct xfs_mount *new_mp = fc->s_fs_info;
1792 xfs_sb_t *sbp = &mp->m_sb;
1793 int flags = fc->sb_flags;
1794 int error;
1795
1796 /* version 5 superblocks always support version counters. */
1797 if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1798 fc->sb_flags |= SB_I_VERSION;
1799
1800 error = xfs_fs_validate_params(new_mp);
1801 if (error)
1802 return error;
1803
1804 sync_filesystem(mp->m_super);
1805
1806 /* inode32 -> inode64 */
1807 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1808 !(new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1809 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1810 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1811 }
1812
1813 /* inode64 -> inode32 */
1814 if (!(mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1815 (new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1816 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1817 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1818 }
1819
1820 /* ro -> rw */
1821 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(flags & SB_RDONLY)) {
1822 error = xfs_remount_rw(mp);
1823 if (error)
1824 return error;
1825 }
1826
1827 /* rw -> ro */
1828 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (flags & SB_RDONLY)) {
1829 error = xfs_remount_ro(mp);
1830 if (error)
1831 return error;
1832 }
1833
1834 return 0;
1835 }
1836
1837 static void xfs_fs_free(
1838 struct fs_context *fc)
1839 {
1840 struct xfs_mount *mp = fc->s_fs_info;
1841
1842 /*
1843 * mp is stored in the fs_context when it is initialized.
1844 * mp is transferred to the superblock on a successful mount,
1845 * but if an error occurs before the transfer we have to free
1846 * it here.
1847 */
1848 if (mp)
1849 xfs_mount_free(mp);
1850 }
1851
1852 static const struct fs_context_operations xfs_context_ops = {
1853 .parse_param = xfs_fs_parse_param,
1854 .get_tree = xfs_fs_get_tree,
1855 .reconfigure = xfs_fs_reconfigure,
1856 .free = xfs_fs_free,
1857 };
1858
1859 static int xfs_init_fs_context(
1860 struct fs_context *fc)
1861 {
1862 struct xfs_mount *mp;
1863
1864 mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1865 if (!mp)
1866 return -ENOMEM;
1867
1868 spin_lock_init(&mp->m_sb_lock);
1869 spin_lock_init(&mp->m_agirotor_lock);
1870 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1871 spin_lock_init(&mp->m_perag_lock);
1872 mutex_init(&mp->m_growlock);
1873 INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1874 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1875 INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1876 INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
1877 mp->m_kobj.kobject.kset = xfs_kset;
1878 /*
1879 * We don't create the finobt per-ag space reservation until after log
1880 * recovery, so we must set this to true so that an ifree transaction
1881 * started during log recovery will not depend on space reservations
1882 * for finobt expansion.
1883 */
1884 mp->m_finobt_nores = true;
1885
1886 /*
1887 * These can be overridden by the mount option parsing.
1888 */
1889 mp->m_logbufs = -1;
1890 mp->m_logbsize = -1;
1891 mp->m_allocsize_log = 16; /* 64k */
1892
1893 /*
1894 * Copy binary VFS mount flags we are interested in.
1895 */
1896 if (fc->sb_flags & SB_RDONLY)
1897 mp->m_flags |= XFS_MOUNT_RDONLY;
1898 if (fc->sb_flags & SB_DIRSYNC)
1899 mp->m_flags |= XFS_MOUNT_DIRSYNC;
1900 if (fc->sb_flags & SB_SYNCHRONOUS)
1901 mp->m_flags |= XFS_MOUNT_WSYNC;
1902
1903 fc->s_fs_info = mp;
1904 fc->ops = &xfs_context_ops;
1905
1906 return 0;
1907 }
1908
1909 static struct file_system_type xfs_fs_type = {
1910 .owner = THIS_MODULE,
1911 .name = "xfs",
1912 .init_fs_context = xfs_init_fs_context,
1913 .parameters = xfs_fs_parameters,
1914 .kill_sb = kill_block_super,
1915 .fs_flags = FS_REQUIRES_DEV,
1916 };
1917 MODULE_ALIAS_FS("xfs");
1918
1919 STATIC int __init
1920 xfs_init_zones(void)
1921 {
1922 xfs_log_ticket_zone = kmem_cache_create("xfs_log_ticket",
1923 sizeof(struct xlog_ticket),
1924 0, 0, NULL);
1925 if (!xfs_log_ticket_zone)
1926 goto out;
1927
1928 xfs_bmap_free_item_zone = kmem_cache_create("xfs_bmap_free_item",
1929 sizeof(struct xfs_extent_free_item),
1930 0, 0, NULL);
1931 if (!xfs_bmap_free_item_zone)
1932 goto out_destroy_log_ticket_zone;
1933
1934 xfs_btree_cur_zone = kmem_cache_create("xfs_btree_cur",
1935 sizeof(struct xfs_btree_cur),
1936 0, 0, NULL);
1937 if (!xfs_btree_cur_zone)
1938 goto out_destroy_bmap_free_item_zone;
1939
1940 xfs_da_state_zone = kmem_cache_create("xfs_da_state",
1941 sizeof(struct xfs_da_state),
1942 0, 0, NULL);
1943 if (!xfs_da_state_zone)
1944 goto out_destroy_btree_cur_zone;
1945
1946 xfs_ifork_zone = kmem_cache_create("xfs_ifork",
1947 sizeof(struct xfs_ifork),
1948 0, 0, NULL);
1949 if (!xfs_ifork_zone)
1950 goto out_destroy_da_state_zone;
1951
1952 xfs_trans_zone = kmem_cache_create("xf_trans",
1953 sizeof(struct xfs_trans),
1954 0, 0, NULL);
1955 if (!xfs_trans_zone)
1956 goto out_destroy_ifork_zone;
1957
1958
1959 /*
1960 * The size of the zone allocated buf log item is the maximum
1961 * size possible under XFS. This wastes a little bit of memory,
1962 * but it is much faster.
1963 */
1964 xfs_buf_item_zone = kmem_cache_create("xfs_buf_item",
1965 sizeof(struct xfs_buf_log_item),
1966 0, 0, NULL);
1967 if (!xfs_buf_item_zone)
1968 goto out_destroy_trans_zone;
1969
1970 xfs_efd_zone = kmem_cache_create("xfs_efd_item",
1971 (sizeof(struct xfs_efd_log_item) +
1972 (XFS_EFD_MAX_FAST_EXTENTS - 1) *
1973 sizeof(struct xfs_extent)),
1974 0, 0, NULL);
1975 if (!xfs_efd_zone)
1976 goto out_destroy_buf_item_zone;
1977
1978 xfs_efi_zone = kmem_cache_create("xfs_efi_item",
1979 (sizeof(struct xfs_efi_log_item) +
1980 (XFS_EFI_MAX_FAST_EXTENTS - 1) *
1981 sizeof(struct xfs_extent)),
1982 0, 0, NULL);
1983 if (!xfs_efi_zone)
1984 goto out_destroy_efd_zone;
1985
1986 xfs_inode_zone = kmem_cache_create("xfs_inode",
1987 sizeof(struct xfs_inode), 0,
1988 (SLAB_HWCACHE_ALIGN |
1989 SLAB_RECLAIM_ACCOUNT |
1990 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
1991 xfs_fs_inode_init_once);
1992 if (!xfs_inode_zone)
1993 goto out_destroy_efi_zone;
1994
1995 xfs_ili_zone = kmem_cache_create("xfs_ili",
1996 sizeof(struct xfs_inode_log_item), 0,
1997 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1998 NULL);
1999 if (!xfs_ili_zone)
2000 goto out_destroy_inode_zone;
2001
2002 xfs_icreate_zone = kmem_cache_create("xfs_icr",
2003 sizeof(struct xfs_icreate_item),
2004 0, 0, NULL);
2005 if (!xfs_icreate_zone)
2006 goto out_destroy_ili_zone;
2007
2008 xfs_rud_zone = kmem_cache_create("xfs_rud_item",
2009 sizeof(struct xfs_rud_log_item),
2010 0, 0, NULL);
2011 if (!xfs_rud_zone)
2012 goto out_destroy_icreate_zone;
2013
2014 xfs_rui_zone = kmem_cache_create("xfs_rui_item",
2015 xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2016 0, 0, NULL);
2017 if (!xfs_rui_zone)
2018 goto out_destroy_rud_zone;
2019
2020 xfs_cud_zone = kmem_cache_create("xfs_cud_item",
2021 sizeof(struct xfs_cud_log_item),
2022 0, 0, NULL);
2023 if (!xfs_cud_zone)
2024 goto out_destroy_rui_zone;
2025
2026 xfs_cui_zone = kmem_cache_create("xfs_cui_item",
2027 xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2028 0, 0, NULL);
2029 if (!xfs_cui_zone)
2030 goto out_destroy_cud_zone;
2031
2032 xfs_bud_zone = kmem_cache_create("xfs_bud_item",
2033 sizeof(struct xfs_bud_log_item),
2034 0, 0, NULL);
2035 if (!xfs_bud_zone)
2036 goto out_destroy_cui_zone;
2037
2038 xfs_bui_zone = kmem_cache_create("xfs_bui_item",
2039 xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2040 0, 0, NULL);
2041 if (!xfs_bui_zone)
2042 goto out_destroy_bud_zone;
2043
2044 return 0;
2045
2046 out_destroy_bud_zone:
2047 kmem_cache_destroy(xfs_bud_zone);
2048 out_destroy_cui_zone:
2049 kmem_cache_destroy(xfs_cui_zone);
2050 out_destroy_cud_zone:
2051 kmem_cache_destroy(xfs_cud_zone);
2052 out_destroy_rui_zone:
2053 kmem_cache_destroy(xfs_rui_zone);
2054 out_destroy_rud_zone:
2055 kmem_cache_destroy(xfs_rud_zone);
2056 out_destroy_icreate_zone:
2057 kmem_cache_destroy(xfs_icreate_zone);
2058 out_destroy_ili_zone:
2059 kmem_cache_destroy(xfs_ili_zone);
2060 out_destroy_inode_zone:
2061 kmem_cache_destroy(xfs_inode_zone);
2062 out_destroy_efi_zone:
2063 kmem_cache_destroy(xfs_efi_zone);
2064 out_destroy_efd_zone:
2065 kmem_cache_destroy(xfs_efd_zone);
2066 out_destroy_buf_item_zone:
2067 kmem_cache_destroy(xfs_buf_item_zone);
2068 out_destroy_trans_zone:
2069 kmem_cache_destroy(xfs_trans_zone);
2070 out_destroy_ifork_zone:
2071 kmem_cache_destroy(xfs_ifork_zone);
2072 out_destroy_da_state_zone:
2073 kmem_cache_destroy(xfs_da_state_zone);
2074 out_destroy_btree_cur_zone:
2075 kmem_cache_destroy(xfs_btree_cur_zone);
2076 out_destroy_bmap_free_item_zone:
2077 kmem_cache_destroy(xfs_bmap_free_item_zone);
2078 out_destroy_log_ticket_zone:
2079 kmem_cache_destroy(xfs_log_ticket_zone);
2080 out:
2081 return -ENOMEM;
2082 }
2083
2084 STATIC void
2085 xfs_destroy_zones(void)
2086 {
2087 /*
2088 * Make sure all delayed rcu free are flushed before we
2089 * destroy caches.
2090 */
2091 rcu_barrier();
2092 kmem_cache_destroy(xfs_bui_zone);
2093 kmem_cache_destroy(xfs_bud_zone);
2094 kmem_cache_destroy(xfs_cui_zone);
2095 kmem_cache_destroy(xfs_cud_zone);
2096 kmem_cache_destroy(xfs_rui_zone);
2097 kmem_cache_destroy(xfs_rud_zone);
2098 kmem_cache_destroy(xfs_icreate_zone);
2099 kmem_cache_destroy(xfs_ili_zone);
2100 kmem_cache_destroy(xfs_inode_zone);
2101 kmem_cache_destroy(xfs_efi_zone);
2102 kmem_cache_destroy(xfs_efd_zone);
2103 kmem_cache_destroy(xfs_buf_item_zone);
2104 kmem_cache_destroy(xfs_trans_zone);
2105 kmem_cache_destroy(xfs_ifork_zone);
2106 kmem_cache_destroy(xfs_da_state_zone);
2107 kmem_cache_destroy(xfs_btree_cur_zone);
2108 kmem_cache_destroy(xfs_bmap_free_item_zone);
2109 kmem_cache_destroy(xfs_log_ticket_zone);
2110 }
2111
2112 STATIC int __init
2113 xfs_init_workqueues(void)
2114 {
2115 /*
2116 * The allocation workqueue can be used in memory reclaim situations
2117 * (writepage path), and parallelism is only limited by the number of
2118 * AGs in all the filesystems mounted. Hence use the default large
2119 * max_active value for this workqueue.
2120 */
2121 xfs_alloc_wq = alloc_workqueue("xfsalloc",
2122 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
2123 if (!xfs_alloc_wq)
2124 return -ENOMEM;
2125
2126 xfs_discard_wq = alloc_workqueue("xfsdiscard", WQ_UNBOUND, 0);
2127 if (!xfs_discard_wq)
2128 goto out_free_alloc_wq;
2129
2130 return 0;
2131 out_free_alloc_wq:
2132 destroy_workqueue(xfs_alloc_wq);
2133 return -ENOMEM;
2134 }
2135
2136 STATIC void
2137 xfs_destroy_workqueues(void)
2138 {
2139 destroy_workqueue(xfs_discard_wq);
2140 destroy_workqueue(xfs_alloc_wq);
2141 }
2142
2143 STATIC int __init
2144 init_xfs_fs(void)
2145 {
2146 int error;
2147
2148 xfs_check_ondisk_structs();
2149
2150 printk(KERN_INFO XFS_VERSION_STRING " with "
2151 XFS_BUILD_OPTIONS " enabled\n");
2152
2153 xfs_dir_startup();
2154
2155 error = xfs_init_zones();
2156 if (error)
2157 goto out;
2158
2159 error = xfs_init_workqueues();
2160 if (error)
2161 goto out_destroy_zones;
2162
2163 error = xfs_mru_cache_init();
2164 if (error)
2165 goto out_destroy_wq;
2166
2167 error = xfs_buf_init();
2168 if (error)
2169 goto out_mru_cache_uninit;
2170
2171 error = xfs_init_procfs();
2172 if (error)
2173 goto out_buf_terminate;
2174
2175 error = xfs_sysctl_register();
2176 if (error)
2177 goto out_cleanup_procfs;
2178
2179 xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2180 if (!xfs_kset) {
2181 error = -ENOMEM;
2182 goto out_sysctl_unregister;
2183 }
2184
2185 xfsstats.xs_kobj.kobject.kset = xfs_kset;
2186
2187 xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2188 if (!xfsstats.xs_stats) {
2189 error = -ENOMEM;
2190 goto out_kset_unregister;
2191 }
2192
2193 error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2194 "stats");
2195 if (error)
2196 goto out_free_stats;
2197
2198 #ifdef DEBUG
2199 xfs_dbg_kobj.kobject.kset = xfs_kset;
2200 error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2201 if (error)
2202 goto out_remove_stats_kobj;
2203 #endif
2204
2205 error = xfs_qm_init();
2206 if (error)
2207 goto out_remove_dbg_kobj;
2208
2209 error = register_filesystem(&xfs_fs_type);
2210 if (error)
2211 goto out_qm_exit;
2212 return 0;
2213
2214 out_qm_exit:
2215 xfs_qm_exit();
2216 out_remove_dbg_kobj:
2217 #ifdef DEBUG
2218 xfs_sysfs_del(&xfs_dbg_kobj);
2219 out_remove_stats_kobj:
2220 #endif
2221 xfs_sysfs_del(&xfsstats.xs_kobj);
2222 out_free_stats:
2223 free_percpu(xfsstats.xs_stats);
2224 out_kset_unregister:
2225 kset_unregister(xfs_kset);
2226 out_sysctl_unregister:
2227 xfs_sysctl_unregister();
2228 out_cleanup_procfs:
2229 xfs_cleanup_procfs();
2230 out_buf_terminate:
2231 xfs_buf_terminate();
2232 out_mru_cache_uninit:
2233 xfs_mru_cache_uninit();
2234 out_destroy_wq:
2235 xfs_destroy_workqueues();
2236 out_destroy_zones:
2237 xfs_destroy_zones();
2238 out:
2239 return error;
2240 }
2241
2242 STATIC void __exit
2243 exit_xfs_fs(void)
2244 {
2245 xfs_qm_exit();
2246 unregister_filesystem(&xfs_fs_type);
2247 #ifdef DEBUG
2248 xfs_sysfs_del(&xfs_dbg_kobj);
2249 #endif
2250 xfs_sysfs_del(&xfsstats.xs_kobj);
2251 free_percpu(xfsstats.xs_stats);
2252 kset_unregister(xfs_kset);
2253 xfs_sysctl_unregister();
2254 xfs_cleanup_procfs();
2255 xfs_buf_terminate();
2256 xfs_mru_cache_uninit();
2257 xfs_destroy_workqueues();
2258 xfs_destroy_zones();
2259 xfs_uuid_table_free();
2260 }
2261
2262 module_init(init_xfs_fs);
2263 module_exit(exit_xfs_fs);
2264
2265 MODULE_AUTHOR("Silicon Graphics, Inc.");
2266 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2267 MODULE_LICENSE("GPL");
Linux with added FPC-III support