blob: cbe729310eeea6cd566a74bfc9c4ce2022861ce1 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_dir2.h"
28#include "xfs_dmapi.h"
29#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110031#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_dir2_sf.h"
Nathan Scotta844f452005-11-02 14:38:42 +110034#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_dinode.h"
36#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110037#include "xfs_btree.h"
38#include "xfs_ialloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_alloc.h"
40#include "xfs_rtalloc.h"
41#include "xfs_bmap.h"
42#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include "xfs_rw.h"
44#include "xfs_quota.h"
45#include "xfs_fsops.h"
46
47STATIC void xfs_mount_log_sbunit(xfs_mount_t *, __int64_t);
48STATIC int xfs_uuid_mount(xfs_mount_t *);
49STATIC void xfs_uuid_unmount(xfs_mount_t *mp);
Christoph Hellwigba0f32d2005-06-21 15:36:52 +100050STATIC void xfs_unmountfs_wait(xfs_mount_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
David Chinner8d280b92006-03-14 13:13:09 +110052
53#ifdef HAVE_PERCPU_SB
54STATIC void xfs_icsb_destroy_counters(xfs_mount_t *);
David Chinner20b64282007-02-10 18:35:09 +110055STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t, int,
56int);
David Chinner8d280b92006-03-14 13:13:09 +110057STATIC void xfs_icsb_sync_counters(xfs_mount_t *);
58STATIC int xfs_icsb_modify_counters(xfs_mount_t *, xfs_sb_field_t,
59 int, int);
David Chinnere8234a62006-03-14 13:23:52 +110060STATIC int xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
David Chinner8d280b92006-03-14 13:13:09 +110061
62#else
63
64#define xfs_icsb_destroy_counters(mp) do { } while (0)
David Chinner20b64282007-02-10 18:35:09 +110065#define xfs_icsb_balance_counter(mp, a, b, c) do { } while (0)
David Chinner8d280b92006-03-14 13:13:09 +110066#define xfs_icsb_sync_counters(mp) do { } while (0)
67#define xfs_icsb_modify_counters(mp, a, b, c) do { } while (0)
David Chinner8d280b92006-03-14 13:13:09 +110068
69#endif
70
Christoph Hellwig1df84c92006-01-11 15:29:52 +110071static const struct {
David Chinner8d280b92006-03-14 13:13:09 +110072 short offset;
73 short type; /* 0 = integer
74 * 1 = binary / string (no translation)
75 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070076} xfs_sb_info[] = {
77 { offsetof(xfs_sb_t, sb_magicnum), 0 },
78 { offsetof(xfs_sb_t, sb_blocksize), 0 },
79 { offsetof(xfs_sb_t, sb_dblocks), 0 },
80 { offsetof(xfs_sb_t, sb_rblocks), 0 },
81 { offsetof(xfs_sb_t, sb_rextents), 0 },
82 { offsetof(xfs_sb_t, sb_uuid), 1 },
83 { offsetof(xfs_sb_t, sb_logstart), 0 },
84 { offsetof(xfs_sb_t, sb_rootino), 0 },
85 { offsetof(xfs_sb_t, sb_rbmino), 0 },
86 { offsetof(xfs_sb_t, sb_rsumino), 0 },
87 { offsetof(xfs_sb_t, sb_rextsize), 0 },
88 { offsetof(xfs_sb_t, sb_agblocks), 0 },
89 { offsetof(xfs_sb_t, sb_agcount), 0 },
90 { offsetof(xfs_sb_t, sb_rbmblocks), 0 },
91 { offsetof(xfs_sb_t, sb_logblocks), 0 },
92 { offsetof(xfs_sb_t, sb_versionnum), 0 },
93 { offsetof(xfs_sb_t, sb_sectsize), 0 },
94 { offsetof(xfs_sb_t, sb_inodesize), 0 },
95 { offsetof(xfs_sb_t, sb_inopblock), 0 },
96 { offsetof(xfs_sb_t, sb_fname[0]), 1 },
97 { offsetof(xfs_sb_t, sb_blocklog), 0 },
98 { offsetof(xfs_sb_t, sb_sectlog), 0 },
99 { offsetof(xfs_sb_t, sb_inodelog), 0 },
100 { offsetof(xfs_sb_t, sb_inopblog), 0 },
101 { offsetof(xfs_sb_t, sb_agblklog), 0 },
102 { offsetof(xfs_sb_t, sb_rextslog), 0 },
103 { offsetof(xfs_sb_t, sb_inprogress), 0 },
104 { offsetof(xfs_sb_t, sb_imax_pct), 0 },
105 { offsetof(xfs_sb_t, sb_icount), 0 },
106 { offsetof(xfs_sb_t, sb_ifree), 0 },
107 { offsetof(xfs_sb_t, sb_fdblocks), 0 },
108 { offsetof(xfs_sb_t, sb_frextents), 0 },
109 { offsetof(xfs_sb_t, sb_uquotino), 0 },
110 { offsetof(xfs_sb_t, sb_gquotino), 0 },
111 { offsetof(xfs_sb_t, sb_qflags), 0 },
112 { offsetof(xfs_sb_t, sb_flags), 0 },
113 { offsetof(xfs_sb_t, sb_shared_vn), 0 },
114 { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
115 { offsetof(xfs_sb_t, sb_unit), 0 },
116 { offsetof(xfs_sb_t, sb_width), 0 },
117 { offsetof(xfs_sb_t, sb_dirblklog), 0 },
118 { offsetof(xfs_sb_t, sb_logsectlog), 0 },
119 { offsetof(xfs_sb_t, sb_logsectsize),0 },
120 { offsetof(xfs_sb_t, sb_logsunit), 0 },
121 { offsetof(xfs_sb_t, sb_features2), 0 },
122 { sizeof(xfs_sb_t), 0 }
123};
124
125/*
126 * Return a pointer to an initialized xfs_mount structure.
127 */
128xfs_mount_t *
129xfs_mount_init(void)
130{
131 xfs_mount_t *mp;
132
David Chinner8d280b92006-03-14 13:13:09 +1100133 mp = kmem_zalloc(sizeof(xfs_mount_t), KM_SLEEP);
134
135 if (xfs_icsb_init_counters(mp)) {
136 mp->m_flags |= XFS_MOUNT_NO_PERCPU_SB;
137 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
139 AIL_LOCKINIT(&mp->m_ail_lock, "xfs_ail");
140 spinlock_init(&mp->m_sb_lock, "xfs_sb");
Jes Sorensen794ee1b2006-01-09 15:59:21 -0800141 mutex_init(&mp->m_ilock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 initnsema(&mp->m_growlock, 1, "xfs_grow");
143 /*
144 * Initialize the AIL.
145 */
146 xfs_trans_ail_init(mp);
147
148 atomic_set(&mp->m_active_trans, 0);
149
150 return mp;
151}
152
153/*
154 * Free up the resources associated with a mount structure. Assume that
155 * the structure was initially zeroed, so we can tell which fields got
156 * initialized.
157 */
158void
159xfs_mount_free(
David Chinner8d280b92006-03-14 13:13:09 +1100160 xfs_mount_t *mp,
161 int remove_bhv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162{
163 if (mp->m_ihash)
164 xfs_ihash_free(mp);
165 if (mp->m_chash)
166 xfs_chash_free(mp);
167
168 if (mp->m_perag) {
169 int agno;
170
171 for (agno = 0; agno < mp->m_maxagi; agno++)
172 if (mp->m_perag[agno].pagb_list)
173 kmem_free(mp->m_perag[agno].pagb_list,
174 sizeof(xfs_perag_busy_t) *
175 XFS_PAGB_NUM_SLOTS);
176 kmem_free(mp->m_perag,
177 sizeof(xfs_perag_t) * mp->m_sb.sb_agcount);
178 }
179
180 AIL_LOCK_DESTROY(&mp->m_ail_lock);
181 spinlock_destroy(&mp->m_sb_lock);
182 mutex_destroy(&mp->m_ilock);
183 freesema(&mp->m_growlock);
184 if (mp->m_quotainfo)
185 XFS_QM_DONE(mp);
186
187 if (mp->m_fsname != NULL)
188 kmem_free(mp->m_fsname, mp->m_fsname_len);
Nathan Scottfc1f8c12005-11-02 11:44:33 +1100189 if (mp->m_rtname != NULL)
190 kmem_free(mp->m_rtname, strlen(mp->m_rtname) + 1);
191 if (mp->m_logname != NULL)
192 kmem_free(mp->m_logname, strlen(mp->m_logname) + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
194 if (remove_bhv) {
Nathan Scottb83bd132006-06-09 16:48:30 +1000195 struct bhv_vfs *vfsp = XFS_MTOVFS(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
197 bhv_remove_all_vfsops(vfsp, 0);
198 VFS_REMOVEBHV(vfsp, &mp->m_bhv);
199 }
200
David Chinner8d280b92006-03-14 13:13:09 +1100201 xfs_icsb_destroy_counters(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 kmem_free(mp, sizeof(xfs_mount_t));
203}
204
205
206/*
207 * Check the validity of the SB found.
208 */
209STATIC int
210xfs_mount_validate_sb(
211 xfs_mount_t *mp,
Nathan Scott764d1f82006-03-31 13:04:17 +1000212 xfs_sb_t *sbp,
213 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
215 /*
216 * If the log device and data device have the
217 * same device number, the log is internal.
218 * Consequently, the sb_logstart should be non-zero. If
219 * we have a zero sb_logstart in this case, we may be trying to mount
220 * a volume filesystem in a non-volume manner.
221 */
222 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000223 xfs_fs_mount_cmn_err(flags, "bad magic number");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 return XFS_ERROR(EWRONGFS);
225 }
226
227 if (!XFS_SB_GOOD_VERSION(sbp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000228 xfs_fs_mount_cmn_err(flags, "bad version");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 return XFS_ERROR(EWRONGFS);
230 }
231
232 if (unlikely(
233 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000234 xfs_fs_mount_cmn_err(flags,
235 "filesystem is marked as having an external log; "
236 "specify logdev on the\nmount command line.");
237 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 }
239
240 if (unlikely(
241 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000242 xfs_fs_mount_cmn_err(flags,
243 "filesystem is marked as having an internal log; "
244 "do not specify logdev on\nthe mount command line.");
245 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 }
247
248 /*
249 * More sanity checking. These were stolen directly from
250 * xfs_repair.
251 */
252 if (unlikely(
253 sbp->sb_agcount <= 0 ||
254 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
255 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
256 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
257 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
258 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
259 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
260 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
261 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
262 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
263 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
Nathan Scott9f989c92006-03-14 13:29:32 +1100264 sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
265 sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
266 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
268 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
Nathan Scotte50bd162006-04-11 15:10:45 +1000269 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000270 xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 return XFS_ERROR(EFSCORRUPTED);
272 }
273
274 /*
275 * Sanity check AG count, size fields against data size field
276 */
277 if (unlikely(
278 sbp->sb_dblocks == 0 ||
279 sbp->sb_dblocks >
280 (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
281 sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
282 sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000283 xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 return XFS_ERROR(EFSCORRUPTED);
285 }
286
287 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
288 ASSERT(sbp->sb_blocklog >= BBSHIFT);
289
290#if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
291 if (unlikely(
292 (sbp->sb_dblocks >> (PAGE_SHIFT - sbp->sb_blocklog)) > ULONG_MAX ||
293 (sbp->sb_rblocks >> (PAGE_SHIFT - sbp->sb_blocklog)) > ULONG_MAX)) {
294#else /* Limited by UINT_MAX of sectors */
295 if (unlikely(
296 (sbp->sb_dblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX ||
297 (sbp->sb_rblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX)) {
298#endif
Nathan Scott764d1f82006-03-31 13:04:17 +1000299 xfs_fs_mount_cmn_err(flags,
300 "file system too large to be mounted on this system.");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 return XFS_ERROR(E2BIG);
302 }
303
304 if (unlikely(sbp->sb_inprogress)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000305 xfs_fs_mount_cmn_err(flags, "file system busy");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 return XFS_ERROR(EFSCORRUPTED);
307 }
308
309 /*
Nathan Scottde206142005-05-05 13:24:13 -0700310 * Version 1 directory format has never worked on Linux.
311 */
312 if (unlikely(!XFS_SB_VERSION_HASDIRV2(sbp))) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000313 xfs_fs_mount_cmn_err(flags,
314 "file system using version 1 directory format");
Nathan Scottde206142005-05-05 13:24:13 -0700315 return XFS_ERROR(ENOSYS);
316 }
317
318 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 * Until this is fixed only page-sized or smaller data blocks work.
320 */
321 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000322 xfs_fs_mount_cmn_err(flags,
323 "file system with blocksize %d bytes",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 sbp->sb_blocksize);
Nathan Scott764d1f82006-03-31 13:04:17 +1000325 xfs_fs_mount_cmn_err(flags,
326 "only pagesize (%ld) or less will currently work.",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 PAGE_SIZE);
328 return XFS_ERROR(ENOSYS);
329 }
330
331 return 0;
332}
333
334xfs_agnumber_t
Nathan Scottc11e2c32005-11-02 15:11:45 +1100335xfs_initialize_perag(
Nathan Scottb83bd132006-06-09 16:48:30 +1000336 bhv_vfs_t *vfs,
Nathan Scottc11e2c32005-11-02 15:11:45 +1100337 xfs_mount_t *mp,
338 xfs_agnumber_t agcount)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339{
340 xfs_agnumber_t index, max_metadata;
341 xfs_perag_t *pag;
342 xfs_agino_t agino;
343 xfs_ino_t ino;
344 xfs_sb_t *sbp = &mp->m_sb;
345 xfs_ino_t max_inum = XFS_MAXINUMBER_32;
346
347 /* Check to see if the filesystem can overflow 32 bit inodes */
348 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
349 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
350
351 /* Clear the mount flag if no inode can overflow 32 bits
352 * on this filesystem, or if specifically requested..
353 */
Nathan Scottc11e2c32005-11-02 15:11:45 +1100354 if ((vfs->vfs_flag & VFS_32BITINODES) && ino > max_inum) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 mp->m_flags |= XFS_MOUNT_32BITINODES;
356 } else {
357 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
358 }
359
360 /* If we can overflow then setup the ag headers accordingly */
361 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
362 /* Calculate how much should be reserved for inodes to
363 * meet the max inode percentage.
364 */
365 if (mp->m_maxicount) {
366 __uint64_t icount;
367
368 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
369 do_div(icount, 100);
370 icount += sbp->sb_agblocks - 1;
Eric Sandeena749ee82005-11-02 15:13:42 +1100371 do_div(icount, sbp->sb_agblocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 max_metadata = icount;
373 } else {
374 max_metadata = agcount;
375 }
376 for (index = 0; index < agcount; index++) {
377 ino = XFS_AGINO_TO_INO(mp, index, agino);
378 if (ino > max_inum) {
379 index++;
380 break;
381 }
382
Nathan Scottc41564b2006-03-29 08:55:14 +1000383 /* This ag is preferred for inodes */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 pag = &mp->m_perag[index];
385 pag->pagi_inodeok = 1;
386 if (index < max_metadata)
387 pag->pagf_metadata = 1;
388 }
389 } else {
390 /* Setup default behavior for smaller filesystems */
391 for (index = 0; index < agcount; index++) {
392 pag = &mp->m_perag[index];
393 pag->pagi_inodeok = 1;
394 }
395 }
396 return index;
397}
398
399/*
400 * xfs_xlatesb
401 *
402 * data - on disk version of sb
403 * sb - a superblock
404 * dir - conversion direction: <0 - convert sb to buf
405 * >0 - convert buf to sb
406 * fields - which fields to copy (bitmask)
407 */
408void
409xfs_xlatesb(
410 void *data,
411 xfs_sb_t *sb,
412 int dir,
413 __int64_t fields)
414{
415 xfs_caddr_t buf_ptr;
416 xfs_caddr_t mem_ptr;
417 xfs_sb_field_t f;
418 int first;
419 int size;
420
421 ASSERT(dir);
422 ASSERT(fields);
423
424 if (!fields)
425 return;
426
427 buf_ptr = (xfs_caddr_t)data;
428 mem_ptr = (xfs_caddr_t)sb;
429
430 while (fields) {
431 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
432 first = xfs_sb_info[f].offset;
433 size = xfs_sb_info[f + 1].offset - first;
434
435 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
436
437 if (size == 1 || xfs_sb_info[f].type == 1) {
438 if (dir > 0) {
439 memcpy(mem_ptr + first, buf_ptr + first, size);
440 } else {
441 memcpy(buf_ptr + first, mem_ptr + first, size);
442 }
443 } else {
444 switch (size) {
445 case 2:
446 INT_XLATE(*(__uint16_t*)(buf_ptr+first),
447 *(__uint16_t*)(mem_ptr+first),
448 dir, ARCH_CONVERT);
449 break;
450 case 4:
451 INT_XLATE(*(__uint32_t*)(buf_ptr+first),
452 *(__uint32_t*)(mem_ptr+first),
453 dir, ARCH_CONVERT);
454 break;
455 case 8:
456 INT_XLATE(*(__uint64_t*)(buf_ptr+first),
457 *(__uint64_t*)(mem_ptr+first), dir, ARCH_CONVERT);
458 break;
459 default:
460 ASSERT(0);
461 }
462 }
463
464 fields &= ~(1LL << f);
465 }
466}
467
468/*
469 * xfs_readsb
470 *
471 * Does the initial read of the superblock.
472 */
473int
Nathan Scott764d1f82006-03-31 13:04:17 +1000474xfs_readsb(xfs_mount_t *mp, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475{
476 unsigned int sector_size;
477 unsigned int extra_flags;
478 xfs_buf_t *bp;
479 xfs_sb_t *sbp;
480 int error;
481
482 ASSERT(mp->m_sb_bp == NULL);
483 ASSERT(mp->m_ddev_targp != NULL);
484
485 /*
486 * Allocate a (locked) buffer to hold the superblock.
487 * This will be kept around at all times to optimize
488 * access to the superblock.
489 */
490 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
491 extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED;
492
493 bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
494 BTOBB(sector_size), extra_flags);
495 if (!bp || XFS_BUF_ISERROR(bp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000496 xfs_fs_mount_cmn_err(flags, "SB read failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
498 goto fail;
499 }
500 ASSERT(XFS_BUF_ISBUSY(bp));
501 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
502
503 /*
504 * Initialize the mount structure from the superblock.
505 * But first do some basic consistency checking.
506 */
507 sbp = XFS_BUF_TO_SBP(bp);
508 xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), 1, XFS_SB_ALL_BITS);
509
Nathan Scott764d1f82006-03-31 13:04:17 +1000510 error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 if (error) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000512 xfs_fs_mount_cmn_err(flags, "SB validate failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 goto fail;
514 }
515
516 /*
517 * We must be able to do sector-sized and sector-aligned IO.
518 */
519 if (sector_size > mp->m_sb.sb_sectsize) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000520 xfs_fs_mount_cmn_err(flags,
521 "device supports only %u byte sectors (not %u)",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 sector_size, mp->m_sb.sb_sectsize);
523 error = ENOSYS;
524 goto fail;
525 }
526
527 /*
528 * If device sector size is smaller than the superblock size,
529 * re-read the superblock so the buffer is correctly sized.
530 */
531 if (sector_size < mp->m_sb.sb_sectsize) {
532 XFS_BUF_UNMANAGE(bp);
533 xfs_buf_relse(bp);
534 sector_size = mp->m_sb.sb_sectsize;
535 bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
536 BTOBB(sector_size), extra_flags);
537 if (!bp || XFS_BUF_ISERROR(bp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000538 xfs_fs_mount_cmn_err(flags, "SB re-read failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
540 goto fail;
541 }
542 ASSERT(XFS_BUF_ISBUSY(bp));
543 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
544 }
545
David Chinner03135cf2007-02-10 18:35:15 +1100546 xfs_icsb_lock(mp);
David Chinner20b64282007-02-10 18:35:09 +1100547 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0, 0);
548 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0, 0);
549 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0, 0);
David Chinner03135cf2007-02-10 18:35:15 +1100550 xfs_icsb_unlock(mp);
David Chinner8d280b92006-03-14 13:13:09 +1100551
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 mp->m_sb_bp = bp;
553 xfs_buf_relse(bp);
554 ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
555 return 0;
556
557 fail:
558 if (bp) {
559 XFS_BUF_UNMANAGE(bp);
560 xfs_buf_relse(bp);
561 }
562 return error;
563}
564
565
566/*
567 * xfs_mount_common
568 *
569 * Mount initialization code establishing various mount
570 * fields from the superblock associated with the given
571 * mount structure
572 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000573STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
575{
576 int i;
577
578 mp->m_agfrotor = mp->m_agirotor = 0;
579 spinlock_init(&mp->m_agirotor_lock, "m_agirotor_lock");
580 mp->m_maxagi = mp->m_sb.sb_agcount;
581 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
582 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
583 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
584 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
585 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
586 mp->m_litino = sbp->sb_inodesize -
587 ((uint)sizeof(xfs_dinode_core_t) + (uint)sizeof(xfs_agino_t));
588 mp->m_blockmask = sbp->sb_blocksize - 1;
589 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
590 mp->m_blockwmask = mp->m_blockwsize - 1;
591 INIT_LIST_HEAD(&mp->m_del_inodes);
592
593 /*
594 * Setup for attributes, in case they get created.
595 * This value is for inodes getting attributes for the first time,
596 * the per-inode value is for old attribute values.
597 */
598 ASSERT(sbp->sb_inodesize >= 256 && sbp->sb_inodesize <= 2048);
599 switch (sbp->sb_inodesize) {
600 case 256:
Nathan Scottd8cc8902005-11-02 10:34:53 +1100601 mp->m_attroffset = XFS_LITINO(mp) -
602 XFS_BMDR_SPACE_CALC(MINABTPTRS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 break;
604 case 512:
605 case 1024:
606 case 2048:
Nathan Scottd8cc8902005-11-02 10:34:53 +1100607 mp->m_attroffset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 break;
609 default:
610 ASSERT(0);
611 }
612 ASSERT(mp->m_attroffset < XFS_LITINO(mp));
613
614 for (i = 0; i < 2; i++) {
615 mp->m_alloc_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
616 xfs_alloc, i == 0);
617 mp->m_alloc_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
618 xfs_alloc, i == 0);
619 }
620 for (i = 0; i < 2; i++) {
621 mp->m_bmap_dmxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
622 xfs_bmbt, i == 0);
623 mp->m_bmap_dmnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
624 xfs_bmbt, i == 0);
625 }
626 for (i = 0; i < 2; i++) {
627 mp->m_inobt_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
628 xfs_inobt, i == 0);
629 mp->m_inobt_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
630 xfs_inobt, i == 0);
631 }
632
633 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
634 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
635 sbp->sb_inopblock);
636 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
637}
638/*
639 * xfs_mountfs
640 *
641 * This function does the following on an initial mount of a file system:
642 * - reads the superblock from disk and init the mount struct
643 * - if we're a 32-bit kernel, do a size check on the superblock
644 * so we don't mount terabyte filesystems
645 * - init mount struct realtime fields
646 * - allocate inode hash table for fs
647 * - init directory manager
648 * - perform recovery and init the log manager
649 */
650int
651xfs_mountfs(
Nathan Scottb83bd132006-06-09 16:48:30 +1000652 bhv_vfs_t *vfsp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 xfs_mount_t *mp,
654 int mfsi_flags)
655{
656 xfs_buf_t *bp;
657 xfs_sb_t *sbp = &(mp->m_sb);
658 xfs_inode_t *rip;
Nathan Scott67fcaa72006-06-09 17:00:52 +1000659 bhv_vnode_t *rvp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 int readio_log, writeio_log;
661 xfs_daddr_t d;
662 __uint64_t ret64;
663 __int64_t update_flags;
664 uint quotamount, quotaflags;
665 int agno;
666 int uuid_mounted = 0;
667 int error = 0;
668
669 if (mp->m_sb_bp == NULL) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000670 if ((error = xfs_readsb(mp, mfsi_flags))) {
Jesper Juhl014c2542006-01-15 02:37:08 +0100671 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 }
673 }
674 xfs_mount_common(mp, sbp);
675
676 /*
677 * Check if sb_agblocks is aligned at stripe boundary
678 * If sb_agblocks is NOT aligned turn off m_dalign since
679 * allocator alignment is within an ag, therefore ag has
680 * to be aligned at stripe boundary.
681 */
682 update_flags = 0LL;
683 if (mp->m_dalign && !(mfsi_flags & XFS_MFSI_SECOND)) {
684 /*
685 * If stripe unit and stripe width are not multiples
686 * of the fs blocksize turn off alignment.
687 */
688 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
689 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
690 if (mp->m_flags & XFS_MOUNT_RETERR) {
691 cmn_err(CE_WARN,
692 "XFS: alignment check 1 failed");
693 error = XFS_ERROR(EINVAL);
694 goto error1;
695 }
696 mp->m_dalign = mp->m_swidth = 0;
697 } else {
698 /*
699 * Convert the stripe unit and width to FSBs.
700 */
701 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
702 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
703 if (mp->m_flags & XFS_MOUNT_RETERR) {
704 error = XFS_ERROR(EINVAL);
705 goto error1;
706 }
707 xfs_fs_cmn_err(CE_WARN, mp,
708"stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
709 mp->m_dalign, mp->m_swidth,
710 sbp->sb_agblocks);
711
712 mp->m_dalign = 0;
713 mp->m_swidth = 0;
714 } else if (mp->m_dalign) {
715 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
716 } else {
717 if (mp->m_flags & XFS_MOUNT_RETERR) {
718 xfs_fs_cmn_err(CE_WARN, mp,
719"stripe alignment turned off: sunit(%d) less than bsize(%d)",
720 mp->m_dalign,
721 mp->m_blockmask +1);
722 error = XFS_ERROR(EINVAL);
723 goto error1;
724 }
725 mp->m_swidth = 0;
726 }
727 }
728
729 /*
730 * Update superblock with new values
731 * and log changes
732 */
733 if (XFS_SB_VERSION_HASDALIGN(sbp)) {
734 if (sbp->sb_unit != mp->m_dalign) {
735 sbp->sb_unit = mp->m_dalign;
736 update_flags |= XFS_SB_UNIT;
737 }
738 if (sbp->sb_width != mp->m_swidth) {
739 sbp->sb_width = mp->m_swidth;
740 update_flags |= XFS_SB_WIDTH;
741 }
742 }
743 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
744 XFS_SB_VERSION_HASDALIGN(&mp->m_sb)) {
745 mp->m_dalign = sbp->sb_unit;
746 mp->m_swidth = sbp->sb_width;
747 }
748
749 xfs_alloc_compute_maxlevels(mp);
750 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
751 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
752 xfs_ialloc_compute_maxlevels(mp);
753
754 if (sbp->sb_imax_pct) {
755 __uint64_t icount;
756
757 /* Make sure the maximum inode count is a multiple of the
758 * units we allocate inodes in.
759 */
760
761 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
762 do_div(icount, 100);
763 do_div(icount, mp->m_ialloc_blks);
764 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
765 sbp->sb_inopblog;
766 } else
767 mp->m_maxicount = 0;
768
769 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
770
771 /*
772 * XFS uses the uuid from the superblock as the unique
773 * identifier for fsid. We can not use the uuid from the volume
774 * since a single partition filesystem is identical to a single
775 * partition volume/filesystem.
776 */
777 if ((mfsi_flags & XFS_MFSI_SECOND) == 0 &&
778 (mp->m_flags & XFS_MOUNT_NOUUID) == 0) {
779 if (xfs_uuid_mount(mp)) {
780 error = XFS_ERROR(EINVAL);
781 goto error1;
782 }
783 uuid_mounted=1;
784 ret64 = uuid_hash64(&sbp->sb_uuid);
785 memcpy(&vfsp->vfs_fsid, &ret64, sizeof(ret64));
786 }
787
788 /*
789 * Set the default minimum read and write sizes unless
790 * already specified in a mount option.
791 * We use smaller I/O sizes when the file system
792 * is being used for NFS service (wsync mount option).
793 */
794 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
795 if (mp->m_flags & XFS_MOUNT_WSYNC) {
796 readio_log = XFS_WSYNC_READIO_LOG;
797 writeio_log = XFS_WSYNC_WRITEIO_LOG;
798 } else {
799 readio_log = XFS_READIO_LOG_LARGE;
800 writeio_log = XFS_WRITEIO_LOG_LARGE;
801 }
802 } else {
803 readio_log = mp->m_readio_log;
804 writeio_log = mp->m_writeio_log;
805 }
806
807 /*
808 * Set the number of readahead buffers to use based on
809 * physical memory size.
810 */
811 if (xfs_physmem <= 4096) /* <= 16MB */
812 mp->m_nreadaheads = XFS_RW_NREADAHEAD_16MB;
813 else if (xfs_physmem <= 8192) /* <= 32MB */
814 mp->m_nreadaheads = XFS_RW_NREADAHEAD_32MB;
815 else
816 mp->m_nreadaheads = XFS_RW_NREADAHEAD_K32;
817 if (sbp->sb_blocklog > readio_log) {
818 mp->m_readio_log = sbp->sb_blocklog;
819 } else {
820 mp->m_readio_log = readio_log;
821 }
822 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
823 if (sbp->sb_blocklog > writeio_log) {
824 mp->m_writeio_log = sbp->sb_blocklog;
825 } else {
826 mp->m_writeio_log = writeio_log;
827 }
828 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
829
830 /*
831 * Set the inode cluster size based on the physical memory
832 * size. This may still be overridden by the file system
833 * block size if it is larger than the chosen cluster size.
834 */
835 if (xfs_physmem <= btoc(32 * 1024 * 1024)) { /* <= 32 MB */
836 mp->m_inode_cluster_size = XFS_INODE_SMALL_CLUSTER_SIZE;
837 } else {
838 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
839 }
840 /*
841 * Set whether we're using inode alignment.
842 */
843 if (XFS_SB_VERSION_HASALIGN(&mp->m_sb) &&
844 mp->m_sb.sb_inoalignmt >=
845 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
846 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
847 else
848 mp->m_inoalign_mask = 0;
849 /*
850 * If we are using stripe alignment, check whether
851 * the stripe unit is a multiple of the inode alignment
852 */
853 if (mp->m_dalign && mp->m_inoalign_mask &&
854 !(mp->m_dalign & mp->m_inoalign_mask))
855 mp->m_sinoalign = mp->m_dalign;
856 else
857 mp->m_sinoalign = 0;
858 /*
859 * Check that the data (and log if separate) are an ok size.
860 */
861 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
862 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
863 cmn_err(CE_WARN, "XFS: size check 1 failed");
864 error = XFS_ERROR(E2BIG);
865 goto error1;
866 }
867 error = xfs_read_buf(mp, mp->m_ddev_targp,
868 d - XFS_FSS_TO_BB(mp, 1),
869 XFS_FSS_TO_BB(mp, 1), 0, &bp);
870 if (!error) {
871 xfs_buf_relse(bp);
872 } else {
873 cmn_err(CE_WARN, "XFS: size check 2 failed");
874 if (error == ENOSPC) {
875 error = XFS_ERROR(E2BIG);
876 }
877 goto error1;
878 }
879
880 if (((mfsi_flags & XFS_MFSI_CLIENT) == 0) &&
881 mp->m_logdev_targp != mp->m_ddev_targp) {
882 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
883 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
884 cmn_err(CE_WARN, "XFS: size check 3 failed");
885 error = XFS_ERROR(E2BIG);
886 goto error1;
887 }
888 error = xfs_read_buf(mp, mp->m_logdev_targp,
889 d - XFS_FSB_TO_BB(mp, 1),
890 XFS_FSB_TO_BB(mp, 1), 0, &bp);
891 if (!error) {
892 xfs_buf_relse(bp);
893 } else {
894 cmn_err(CE_WARN, "XFS: size check 3 failed");
895 if (error == ENOSPC) {
896 error = XFS_ERROR(E2BIG);
897 }
898 goto error1;
899 }
900 }
901
902 /*
903 * Initialize realtime fields in the mount structure
904 */
905 if ((error = xfs_rtmount_init(mp))) {
906 cmn_err(CE_WARN, "XFS: RT mount failed");
907 goto error1;
908 }
909
910 /*
911 * For client case we are done now
912 */
913 if (mfsi_flags & XFS_MFSI_CLIENT) {
Jesper Juhl014c2542006-01-15 02:37:08 +0100914 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 }
916
917 /*
918 * Copies the low order bits of the timestamp and the randomly
919 * set "sequence" number out of a UUID.
920 */
921 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
922
923 /*
924 * The vfs structure needs to have a file system independent
925 * way of checking for the invariant file system ID. Since it
926 * can't look at mount structures it has a pointer to the data
927 * in the mount structure.
928 *
929 * File systems that don't support user level file handles (i.e.
930 * all of them except for XFS) will leave vfs_altfsid as NULL.
931 */
932 vfsp->vfs_altfsid = (xfs_fsid_t *)mp->m_fixedfsid;
933 mp->m_dmevmask = 0; /* not persistent; set after each mount */
934
Nathan Scottf6c2d1f2006-06-20 13:04:51 +1000935 xfs_dir_mount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936
937 /*
938 * Initialize the attribute manager's entries.
939 */
940 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
941
942 /*
943 * Initialize the precomputed transaction reservations values.
944 */
945 xfs_trans_init(mp);
946
947 /*
948 * Allocate and initialize the inode hash table for this
949 * file system.
950 */
951 xfs_ihash_init(mp);
952 xfs_chash_init(mp);
953
954 /*
955 * Allocate and initialize the per-ag data.
956 */
957 init_rwsem(&mp->m_peraglock);
958 mp->m_perag =
959 kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t), KM_SLEEP);
960
Nathan Scottc11e2c32005-11-02 15:11:45 +1100961 mp->m_maxagi = xfs_initialize_perag(vfsp, mp, sbp->sb_agcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962
963 /*
964 * log's mount-time initialization. Perform 1st part recovery if needed
965 */
966 if (likely(sbp->sb_logblocks > 0)) { /* check for volume case */
967 error = xfs_log_mount(mp, mp->m_logdev_targp,
968 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
969 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
970 if (error) {
971 cmn_err(CE_WARN, "XFS: log mount failed");
972 goto error2;
973 }
974 } else { /* No log has been defined */
975 cmn_err(CE_WARN, "XFS: no log defined");
976 XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW, mp);
977 error = XFS_ERROR(EFSCORRUPTED);
978 goto error2;
979 }
980
981 /*
982 * Get and sanity-check the root inode.
983 * Save the pointer to it in the mount structure.
984 */
985 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0);
986 if (error) {
987 cmn_err(CE_WARN, "XFS: failed to read root inode");
988 goto error3;
989 }
990
991 ASSERT(rip != NULL);
992 rvp = XFS_ITOV(rip);
993
994 if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
995 cmn_err(CE_WARN, "XFS: corrupted root inode");
Nathan Scottb6574522006-06-09 15:29:40 +1000996 cmn_err(CE_WARN, "Device %s - root %llu is not a directory",
997 XFS_BUFTARG_NAME(mp->m_ddev_targp),
998 (unsigned long long)rip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1000 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1001 mp);
1002 error = XFS_ERROR(EFSCORRUPTED);
1003 goto error4;
1004 }
1005 mp->m_rootip = rip; /* save it */
1006
1007 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1008
1009 /*
1010 * Initialize realtime inode pointers in the mount structure
1011 */
1012 if ((error = xfs_rtmount_inodes(mp))) {
1013 /*
1014 * Free up the root inode.
1015 */
1016 cmn_err(CE_WARN, "XFS: failed to read RT inodes");
1017 goto error4;
1018 }
1019
1020 /*
1021 * If fs is not mounted readonly, then update the superblock
1022 * unit and width changes.
1023 */
1024 if (update_flags && !(vfsp->vfs_flag & VFS_RDONLY))
1025 xfs_mount_log_sbunit(mp, update_flags);
1026
1027 /*
1028 * Initialise the XFS quota management subsystem for this mount
1029 */
1030 if ((error = XFS_QM_INIT(mp, &quotamount, &quotaflags)))
1031 goto error4;
1032
1033 /*
1034 * Finish recovering the file system. This part needed to be
1035 * delayed until after the root and real-time bitmap inodes
1036 * were consistently read in.
1037 */
1038 error = xfs_log_mount_finish(mp, mfsi_flags);
1039 if (error) {
1040 cmn_err(CE_WARN, "XFS: log mount finish failed");
1041 goto error4;
1042 }
1043
1044 /*
1045 * Complete the quota initialisation, post-log-replay component.
1046 */
1047 if ((error = XFS_QM_MOUNT(mp, quotamount, quotaflags, mfsi_flags)))
1048 goto error4;
1049
1050 return 0;
1051
1052 error4:
1053 /*
1054 * Free up the root inode.
1055 */
1056 VN_RELE(rvp);
1057 error3:
1058 xfs_log_unmount_dealloc(mp);
1059 error2:
1060 xfs_ihash_free(mp);
1061 xfs_chash_free(mp);
1062 for (agno = 0; agno < sbp->sb_agcount; agno++)
1063 if (mp->m_perag[agno].pagb_list)
1064 kmem_free(mp->m_perag[agno].pagb_list,
1065 sizeof(xfs_perag_busy_t) * XFS_PAGB_NUM_SLOTS);
1066 kmem_free(mp->m_perag, sbp->sb_agcount * sizeof(xfs_perag_t));
1067 mp->m_perag = NULL;
1068 /* FALLTHROUGH */
1069 error1:
1070 if (uuid_mounted)
1071 xfs_uuid_unmount(mp);
1072 xfs_freesb(mp);
1073 return error;
1074}
1075
1076/*
1077 * xfs_unmountfs
1078 *
1079 * This flushes out the inodes,dquots and the superblock, unmounts the
1080 * log and makes sure that incore structures are freed.
1081 */
1082int
1083xfs_unmountfs(xfs_mount_t *mp, struct cred *cr)
1084{
Nathan Scottb83bd132006-06-09 16:48:30 +10001085 struct bhv_vfs *vfsp = XFS_MTOVFS(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086#if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1087 int64_t fsid;
1088#endif
1089
Christoph Hellwigefa80272005-06-21 15:37:17 +10001090 xfs_iflush_all(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091
Nathan Scottee2a4f72006-01-11 15:33:36 +11001092 XFS_QM_DQPURGEALL(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_UMOUNTING);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093
1094 /*
1095 * Flush out the log synchronously so that we know for sure
1096 * that nothing is pinned. This is important because bflush()
1097 * will skip pinned buffers.
1098 */
1099 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1100
1101 xfs_binval(mp->m_ddev_targp);
1102 if (mp->m_rtdev_targp) {
1103 xfs_binval(mp->m_rtdev_targp);
1104 }
1105
1106 xfs_unmountfs_writesb(mp);
1107
1108 xfs_unmountfs_wait(mp); /* wait for async bufs */
1109
1110 xfs_log_unmount(mp); /* Done! No more fs ops. */
1111
1112 xfs_freesb(mp);
1113
1114 /*
1115 * All inodes from this mount point should be freed.
1116 */
1117 ASSERT(mp->m_inodes == NULL);
1118
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 xfs_unmountfs_close(mp, cr);
1120 if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0)
1121 xfs_uuid_unmount(mp);
1122
1123#if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1124 /*
1125 * clear all error tags on this filesystem
1126 */
1127 memcpy(&fsid, &vfsp->vfs_fsid, sizeof(int64_t));
1128 xfs_errortag_clearall_umount(fsid, mp->m_fsname, 0);
1129#endif
1130 XFS_IODONE(vfsp);
1131 xfs_mount_free(mp, 1);
1132 return 0;
1133}
1134
1135void
1136xfs_unmountfs_close(xfs_mount_t *mp, struct cred *cr)
1137{
1138 if (mp->m_logdev_targp != mp->m_ddev_targp)
1139 xfs_free_buftarg(mp->m_logdev_targp, 1);
1140 if (mp->m_rtdev_targp)
1141 xfs_free_buftarg(mp->m_rtdev_targp, 1);
1142 xfs_free_buftarg(mp->m_ddev_targp, 0);
1143}
1144
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10001145STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146xfs_unmountfs_wait(xfs_mount_t *mp)
1147{
1148 if (mp->m_logdev_targp != mp->m_ddev_targp)
1149 xfs_wait_buftarg(mp->m_logdev_targp);
1150 if (mp->m_rtdev_targp)
1151 xfs_wait_buftarg(mp->m_rtdev_targp);
1152 xfs_wait_buftarg(mp->m_ddev_targp);
1153}
1154
1155int
1156xfs_unmountfs_writesb(xfs_mount_t *mp)
1157{
1158 xfs_buf_t *sbp;
1159 xfs_sb_t *sb;
1160 int error = 0;
1161
1162 /*
1163 * skip superblock write if fs is read-only, or
1164 * if we are doing a forced umount.
1165 */
1166 sbp = xfs_getsb(mp, 0);
1167 if (!(XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY ||
1168 XFS_FORCED_SHUTDOWN(mp))) {
David Chinner8d280b92006-03-14 13:13:09 +11001169
1170 xfs_icsb_sync_counters(mp);
1171
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 /*
1173 * mark shared-readonly if desired
1174 */
1175 sb = XFS_BUF_TO_SBP(sbp);
1176 if (mp->m_mk_sharedro) {
1177 if (!(sb->sb_flags & XFS_SBF_READONLY))
1178 sb->sb_flags |= XFS_SBF_READONLY;
1179 if (!XFS_SB_VERSION_HASSHARED(sb))
1180 XFS_SB_VERSION_ADDSHARED(sb);
1181 xfs_fs_cmn_err(CE_NOTE, mp,
1182 "Unmounting, marking shared read-only");
1183 }
1184 XFS_BUF_UNDONE(sbp);
1185 XFS_BUF_UNREAD(sbp);
1186 XFS_BUF_UNDELAYWRITE(sbp);
1187 XFS_BUF_WRITE(sbp);
1188 XFS_BUF_UNASYNC(sbp);
1189 ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1190 xfsbdstrat(mp, sbp);
1191 /* Nevermind errors we might get here. */
1192 error = xfs_iowait(sbp);
1193 if (error)
1194 xfs_ioerror_alert("xfs_unmountfs_writesb",
1195 mp, sbp, XFS_BUF_ADDR(sbp));
1196 if (error && mp->m_mk_sharedro)
1197 xfs_fs_cmn_err(CE_ALERT, mp, "Superblock write error detected while unmounting. Filesystem may not be marked shared readonly");
1198 }
1199 xfs_buf_relse(sbp);
Jesper Juhl014c2542006-01-15 02:37:08 +01001200 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201}
1202
1203/*
1204 * xfs_mod_sb() can be used to copy arbitrary changes to the
1205 * in-core superblock into the superblock buffer to be logged.
1206 * It does not provide the higher level of locking that is
1207 * needed to protect the in-core superblock from concurrent
1208 * access.
1209 */
1210void
1211xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1212{
1213 xfs_buf_t *bp;
1214 int first;
1215 int last;
1216 xfs_mount_t *mp;
1217 xfs_sb_t *sbp;
1218 xfs_sb_field_t f;
1219
1220 ASSERT(fields);
1221 if (!fields)
1222 return;
1223 mp = tp->t_mountp;
1224 bp = xfs_trans_getsb(tp, mp, 0);
1225 sbp = XFS_BUF_TO_SBP(bp);
1226 first = sizeof(xfs_sb_t);
1227 last = 0;
1228
1229 /* translate/copy */
1230
1231 xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), -1, fields);
1232
1233 /* find modified range */
1234
1235 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1236 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1237 first = xfs_sb_info[f].offset;
1238
1239 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1240 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1241 last = xfs_sb_info[f + 1].offset - 1;
1242
1243 xfs_trans_log_buf(tp, bp, first, last);
1244}
Yingping Lud210a282006-06-09 14:55:18 +10001245
Yingping Lud210a282006-06-09 14:55:18 +10001246
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247/*
1248 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1249 * a delta to a specified field in the in-core superblock. Simply
1250 * switch on the field indicated and apply the delta to that field.
1251 * Fields are not allowed to dip below zero, so if the delta would
1252 * do this do not apply it and return EINVAL.
1253 *
1254 * The SB_LOCK must be held when this routine is called.
1255 */
David Chinner8d280b92006-03-14 13:13:09 +11001256int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257xfs_mod_incore_sb_unlocked(xfs_mount_t *mp, xfs_sb_field_t field,
1258 int delta, int rsvd)
1259{
1260 int scounter; /* short counter for 32 bit fields */
1261 long long lcounter; /* long counter for 64 bit fields */
1262 long long res_used, rem;
1263
1264 /*
1265 * With the in-core superblock spin lock held, switch
1266 * on the indicated field. Apply the delta to the
1267 * proper field. If the fields value would dip below
1268 * 0, then do not apply the delta and return EINVAL.
1269 */
1270 switch (field) {
1271 case XFS_SBS_ICOUNT:
1272 lcounter = (long long)mp->m_sb.sb_icount;
1273 lcounter += delta;
1274 if (lcounter < 0) {
1275 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001276 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 }
1278 mp->m_sb.sb_icount = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001279 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 case XFS_SBS_IFREE:
1281 lcounter = (long long)mp->m_sb.sb_ifree;
1282 lcounter += delta;
1283 if (lcounter < 0) {
1284 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001285 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 }
1287 mp->m_sb.sb_ifree = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001288 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 case XFS_SBS_FDBLOCKS:
1290
David Chinner4be536d2006-09-07 14:26:50 +10001291 lcounter = (long long)
1292 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1294
1295 if (delta > 0) { /* Putting blocks back */
1296 if (res_used > delta) {
1297 mp->m_resblks_avail += delta;
1298 } else {
1299 rem = delta - res_used;
1300 mp->m_resblks_avail = mp->m_resblks;
1301 lcounter += rem;
1302 }
1303 } else { /* Taking blocks away */
1304
1305 lcounter += delta;
1306
1307 /*
1308 * If were out of blocks, use any available reserved blocks if
1309 * were allowed to.
1310 */
1311
1312 if (lcounter < 0) {
1313 if (rsvd) {
1314 lcounter = (long long)mp->m_resblks_avail + delta;
1315 if (lcounter < 0) {
Jesper Juhl014c2542006-01-15 02:37:08 +01001316 return XFS_ERROR(ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 }
1318 mp->m_resblks_avail = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001319 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 } else { /* not reserved */
Jesper Juhl014c2542006-01-15 02:37:08 +01001321 return XFS_ERROR(ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 }
1323 }
1324 }
1325
David Chinner4be536d2006-09-07 14:26:50 +10001326 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
Jesper Juhl014c2542006-01-15 02:37:08 +01001327 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 case XFS_SBS_FREXTENTS:
1329 lcounter = (long long)mp->m_sb.sb_frextents;
1330 lcounter += delta;
1331 if (lcounter < 0) {
Jesper Juhl014c2542006-01-15 02:37:08 +01001332 return XFS_ERROR(ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 }
1334 mp->m_sb.sb_frextents = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001335 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 case XFS_SBS_DBLOCKS:
1337 lcounter = (long long)mp->m_sb.sb_dblocks;
1338 lcounter += delta;
1339 if (lcounter < 0) {
1340 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001341 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 }
1343 mp->m_sb.sb_dblocks = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001344 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 case XFS_SBS_AGCOUNT:
1346 scounter = mp->m_sb.sb_agcount;
1347 scounter += delta;
1348 if (scounter < 0) {
1349 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001350 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 }
1352 mp->m_sb.sb_agcount = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001353 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 case XFS_SBS_IMAX_PCT:
1355 scounter = mp->m_sb.sb_imax_pct;
1356 scounter += delta;
1357 if (scounter < 0) {
1358 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001359 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 }
1361 mp->m_sb.sb_imax_pct = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001362 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 case XFS_SBS_REXTSIZE:
1364 scounter = mp->m_sb.sb_rextsize;
1365 scounter += delta;
1366 if (scounter < 0) {
1367 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001368 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 }
1370 mp->m_sb.sb_rextsize = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001371 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 case XFS_SBS_RBMBLOCKS:
1373 scounter = mp->m_sb.sb_rbmblocks;
1374 scounter += delta;
1375 if (scounter < 0) {
1376 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001377 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 }
1379 mp->m_sb.sb_rbmblocks = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001380 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 case XFS_SBS_RBLOCKS:
1382 lcounter = (long long)mp->m_sb.sb_rblocks;
1383 lcounter += delta;
1384 if (lcounter < 0) {
1385 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001386 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 }
1388 mp->m_sb.sb_rblocks = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001389 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 case XFS_SBS_REXTENTS:
1391 lcounter = (long long)mp->m_sb.sb_rextents;
1392 lcounter += delta;
1393 if (lcounter < 0) {
1394 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001395 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 }
1397 mp->m_sb.sb_rextents = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001398 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 case XFS_SBS_REXTSLOG:
1400 scounter = mp->m_sb.sb_rextslog;
1401 scounter += delta;
1402 if (scounter < 0) {
1403 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001404 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 }
1406 mp->m_sb.sb_rextslog = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001407 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 default:
1409 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001410 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 }
1412}
1413
1414/*
1415 * xfs_mod_incore_sb() is used to change a field in the in-core
1416 * superblock structure by the specified delta. This modification
1417 * is protected by the SB_LOCK. Just use the xfs_mod_incore_sb_unlocked()
1418 * routine to do the work.
1419 */
1420int
1421xfs_mod_incore_sb(xfs_mount_t *mp, xfs_sb_field_t field, int delta, int rsvd)
1422{
1423 unsigned long s;
1424 int status;
1425
David Chinner8d280b92006-03-14 13:13:09 +11001426 /* check for per-cpu counters */
1427 switch (field) {
1428#ifdef HAVE_PERCPU_SB
1429 case XFS_SBS_ICOUNT:
1430 case XFS_SBS_IFREE:
1431 case XFS_SBS_FDBLOCKS:
1432 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1433 status = xfs_icsb_modify_counters(mp, field,
1434 delta, rsvd);
1435 break;
1436 }
1437 /* FALLTHROUGH */
1438#endif
1439 default:
1440 s = XFS_SB_LOCK(mp);
1441 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1442 XFS_SB_UNLOCK(mp, s);
1443 break;
1444 }
1445
Jesper Juhl014c2542006-01-15 02:37:08 +01001446 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447}
1448
1449/*
1450 * xfs_mod_incore_sb_batch() is used to change more than one field
1451 * in the in-core superblock structure at a time. This modification
1452 * is protected by a lock internal to this module. The fields and
1453 * changes to those fields are specified in the array of xfs_mod_sb
1454 * structures passed in.
1455 *
1456 * Either all of the specified deltas will be applied or none of
1457 * them will. If any modified field dips below 0, then all modifications
1458 * will be backed out and EINVAL will be returned.
1459 */
1460int
1461xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1462{
1463 unsigned long s;
1464 int status=0;
1465 xfs_mod_sb_t *msbp;
1466
1467 /*
1468 * Loop through the array of mod structures and apply each
1469 * individually. If any fail, then back out all those
1470 * which have already been applied. Do all of this within
1471 * the scope of the SB_LOCK so that all of the changes will
1472 * be atomic.
1473 */
1474 s = XFS_SB_LOCK(mp);
1475 msbp = &msb[0];
1476 for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1477 /*
1478 * Apply the delta at index n. If it fails, break
1479 * from the loop so we'll fall into the undo loop
1480 * below.
1481 */
David Chinner8d280b92006-03-14 13:13:09 +11001482 switch (msbp->msb_field) {
1483#ifdef HAVE_PERCPU_SB
1484 case XFS_SBS_ICOUNT:
1485 case XFS_SBS_IFREE:
1486 case XFS_SBS_FDBLOCKS:
1487 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
David Chinner20b64282007-02-10 18:35:09 +11001488 XFS_SB_UNLOCK(mp, s);
1489 status = xfs_icsb_modify_counters(mp,
David Chinner8d280b92006-03-14 13:13:09 +11001490 msbp->msb_field,
1491 msbp->msb_delta, rsvd);
David Chinner20b64282007-02-10 18:35:09 +11001492 s = XFS_SB_LOCK(mp);
David Chinner8d280b92006-03-14 13:13:09 +11001493 break;
1494 }
1495 /* FALLTHROUGH */
1496#endif
1497 default:
1498 status = xfs_mod_incore_sb_unlocked(mp,
1499 msbp->msb_field,
1500 msbp->msb_delta, rsvd);
1501 break;
1502 }
1503
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 if (status != 0) {
1505 break;
1506 }
1507 }
1508
1509 /*
1510 * If we didn't complete the loop above, then back out
1511 * any changes made to the superblock. If you add code
1512 * between the loop above and here, make sure that you
1513 * preserve the value of status. Loop back until
1514 * we step below the beginning of the array. Make sure
1515 * we don't touch anything back there.
1516 */
1517 if (status != 0) {
1518 msbp--;
1519 while (msbp >= msb) {
David Chinner8d280b92006-03-14 13:13:09 +11001520 switch (msbp->msb_field) {
1521#ifdef HAVE_PERCPU_SB
1522 case XFS_SBS_ICOUNT:
1523 case XFS_SBS_IFREE:
1524 case XFS_SBS_FDBLOCKS:
1525 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
David Chinner20b64282007-02-10 18:35:09 +11001526 XFS_SB_UNLOCK(mp, s);
1527 status = xfs_icsb_modify_counters(mp,
David Chinner8d280b92006-03-14 13:13:09 +11001528 msbp->msb_field,
1529 -(msbp->msb_delta),
1530 rsvd);
David Chinner20b64282007-02-10 18:35:09 +11001531 s = XFS_SB_LOCK(mp);
David Chinner8d280b92006-03-14 13:13:09 +11001532 break;
1533 }
1534 /* FALLTHROUGH */
1535#endif
1536 default:
1537 status = xfs_mod_incore_sb_unlocked(mp,
1538 msbp->msb_field,
1539 -(msbp->msb_delta),
1540 rsvd);
1541 break;
1542 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 ASSERT(status == 0);
1544 msbp--;
1545 }
1546 }
1547 XFS_SB_UNLOCK(mp, s);
Jesper Juhl014c2542006-01-15 02:37:08 +01001548 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549}
1550
1551/*
1552 * xfs_getsb() is called to obtain the buffer for the superblock.
1553 * The buffer is returned locked and read in from disk.
1554 * The buffer should be released with a call to xfs_brelse().
1555 *
1556 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1557 * the superblock buffer if it can be locked without sleeping.
1558 * If it can't then we'll return NULL.
1559 */
1560xfs_buf_t *
1561xfs_getsb(
1562 xfs_mount_t *mp,
1563 int flags)
1564{
1565 xfs_buf_t *bp;
1566
1567 ASSERT(mp->m_sb_bp != NULL);
1568 bp = mp->m_sb_bp;
1569 if (flags & XFS_BUF_TRYLOCK) {
1570 if (!XFS_BUF_CPSEMA(bp)) {
1571 return NULL;
1572 }
1573 } else {
1574 XFS_BUF_PSEMA(bp, PRIBIO);
1575 }
1576 XFS_BUF_HOLD(bp);
1577 ASSERT(XFS_BUF_ISDONE(bp));
Jesper Juhl014c2542006-01-15 02:37:08 +01001578 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579}
1580
1581/*
1582 * Used to free the superblock along various error paths.
1583 */
1584void
1585xfs_freesb(
1586 xfs_mount_t *mp)
1587{
1588 xfs_buf_t *bp;
1589
1590 /*
1591 * Use xfs_getsb() so that the buffer will be locked
1592 * when we call xfs_buf_relse().
1593 */
1594 bp = xfs_getsb(mp, 0);
1595 XFS_BUF_UNMANAGE(bp);
1596 xfs_buf_relse(bp);
1597 mp->m_sb_bp = NULL;
1598}
1599
1600/*
1601 * See if the UUID is unique among mounted XFS filesystems.
1602 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1603 */
1604STATIC int
1605xfs_uuid_mount(
1606 xfs_mount_t *mp)
1607{
1608 if (uuid_is_nil(&mp->m_sb.sb_uuid)) {
1609 cmn_err(CE_WARN,
1610 "XFS: Filesystem %s has nil UUID - can't mount",
1611 mp->m_fsname);
1612 return -1;
1613 }
1614 if (!uuid_table_insert(&mp->m_sb.sb_uuid)) {
1615 cmn_err(CE_WARN,
1616 "XFS: Filesystem %s has duplicate UUID - can't mount",
1617 mp->m_fsname);
1618 return -1;
1619 }
1620 return 0;
1621}
1622
1623/*
1624 * Remove filesystem from the UUID table.
1625 */
1626STATIC void
1627xfs_uuid_unmount(
1628 xfs_mount_t *mp)
1629{
1630 uuid_table_remove(&mp->m_sb.sb_uuid);
1631}
1632
1633/*
1634 * Used to log changes to the superblock unit and width fields which could
1635 * be altered by the mount options. Only the first superblock is updated.
1636 */
1637STATIC void
1638xfs_mount_log_sbunit(
1639 xfs_mount_t *mp,
1640 __int64_t fields)
1641{
1642 xfs_trans_t *tp;
1643
1644 ASSERT(fields & (XFS_SB_UNIT|XFS_SB_WIDTH|XFS_SB_UUID));
1645
1646 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1647 if (xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1648 XFS_DEFAULT_LOG_COUNT)) {
1649 xfs_trans_cancel(tp, 0);
1650 return;
1651 }
1652 xfs_mod_sb(tp, fields);
1653 xfs_trans_commit(tp, 0, NULL);
1654}
David Chinner8d280b92006-03-14 13:13:09 +11001655
1656
1657#ifdef HAVE_PERCPU_SB
1658/*
1659 * Per-cpu incore superblock counters
1660 *
1661 * Simple concept, difficult implementation
1662 *
1663 * Basically, replace the incore superblock counters with a distributed per cpu
1664 * counter for contended fields (e.g. free block count).
1665 *
1666 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1667 * hence needs to be accurately read when we are running low on space. Hence
1668 * there is a method to enable and disable the per-cpu counters based on how
1669 * much "stuff" is available in them.
1670 *
1671 * Basically, a counter is enabled if there is enough free resource to justify
1672 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1673 * ENOSPC), then we disable the counters to synchronise all callers and
1674 * re-distribute the available resources.
1675 *
1676 * If, once we redistributed the available resources, we still get a failure,
1677 * we disable the per-cpu counter and go through the slow path.
1678 *
1679 * The slow path is the current xfs_mod_incore_sb() function. This means that
1680 * when we disable a per-cpu counter, we need to drain it's resources back to
1681 * the global superblock. We do this after disabling the counter to prevent
1682 * more threads from queueing up on the counter.
1683 *
1684 * Essentially, this means that we still need a lock in the fast path to enable
1685 * synchronisation between the global counters and the per-cpu counters. This
1686 * is not a problem because the lock will be local to a CPU almost all the time
1687 * and have little contention except when we get to ENOSPC conditions.
1688 *
1689 * Basically, this lock becomes a barrier that enables us to lock out the fast
1690 * path while we do things like enabling and disabling counters and
1691 * synchronising the counters.
1692 *
1693 * Locking rules:
1694 *
1695 * 1. XFS_SB_LOCK() before picking up per-cpu locks
1696 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1697 * 3. accurate counter sync requires XFS_SB_LOCK + per cpu locks
1698 * 4. modifying per-cpu counters requires holding per-cpu lock
1699 * 5. modifying global counters requires holding XFS_SB_LOCK
1700 * 6. enabling or disabling a counter requires holding the XFS_SB_LOCK
1701 * and _none_ of the per-cpu locks.
1702 *
1703 * Disabled counters are only ever re-enabled by a balance operation
1704 * that results in more free resources per CPU than a given threshold.
1705 * To ensure counters don't remain disabled, they are rebalanced when
1706 * the global resource goes above a higher threshold (i.e. some hysteresis
1707 * is present to prevent thrashing).
David Chinner8d280b92006-03-14 13:13:09 +11001708 */
David Chinnere8234a62006-03-14 13:23:52 +11001709
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001710#ifdef CONFIG_HOTPLUG_CPU
David Chinnere8234a62006-03-14 13:23:52 +11001711/*
1712 * hot-plug CPU notifier support.
1713 *
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001714 * We need a notifier per filesystem as we need to be able to identify
1715 * the filesystem to balance the counters out. This is achieved by
1716 * having a notifier block embedded in the xfs_mount_t and doing pointer
1717 * magic to get the mount pointer from the notifier block address.
David Chinnere8234a62006-03-14 13:23:52 +11001718 */
1719STATIC int
1720xfs_icsb_cpu_notify(
1721 struct notifier_block *nfb,
1722 unsigned long action,
1723 void *hcpu)
1724{
1725 xfs_icsb_cnts_t *cntp;
1726 xfs_mount_t *mp;
1727 int s;
1728
1729 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
1730 cntp = (xfs_icsb_cnts_t *)
1731 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
1732 switch (action) {
1733 case CPU_UP_PREPARE:
1734 /* Easy Case - initialize the area and locks, and
1735 * then rebalance when online does everything else for us. */
David Chinner01e1b692006-03-14 13:29:16 +11001736 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinnere8234a62006-03-14 13:23:52 +11001737 break;
1738 case CPU_ONLINE:
David Chinner03135cf2007-02-10 18:35:15 +11001739 xfs_icsb_lock(mp);
David Chinner20b64282007-02-10 18:35:09 +11001740 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0, 0);
1741 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0, 0);
1742 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0, 0);
David Chinner03135cf2007-02-10 18:35:15 +11001743 xfs_icsb_unlock(mp);
David Chinnere8234a62006-03-14 13:23:52 +11001744 break;
1745 case CPU_DEAD:
1746 /* Disable all the counters, then fold the dead cpu's
1747 * count into the total on the global superblock and
1748 * re-enable the counters. */
David Chinner03135cf2007-02-10 18:35:15 +11001749 xfs_icsb_lock(mp);
David Chinnere8234a62006-03-14 13:23:52 +11001750 s = XFS_SB_LOCK(mp);
1751 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
1752 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
1753 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
1754
1755 mp->m_sb.sb_icount += cntp->icsb_icount;
1756 mp->m_sb.sb_ifree += cntp->icsb_ifree;
1757 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
1758
David Chinner01e1b692006-03-14 13:29:16 +11001759 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinnere8234a62006-03-14 13:23:52 +11001760
David Chinner20b64282007-02-10 18:35:09 +11001761 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT,
1762 XFS_ICSB_SB_LOCKED, 0);
1763 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE,
1764 XFS_ICSB_SB_LOCKED, 0);
1765 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS,
1766 XFS_ICSB_SB_LOCKED, 0);
David Chinnere8234a62006-03-14 13:23:52 +11001767 XFS_SB_UNLOCK(mp, s);
David Chinner03135cf2007-02-10 18:35:15 +11001768 xfs_icsb_unlock(mp);
David Chinnere8234a62006-03-14 13:23:52 +11001769 break;
1770 }
1771
1772 return NOTIFY_OK;
1773}
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001774#endif /* CONFIG_HOTPLUG_CPU */
David Chinnere8234a62006-03-14 13:23:52 +11001775
David Chinner8d280b92006-03-14 13:13:09 +11001776int
1777xfs_icsb_init_counters(
1778 xfs_mount_t *mp)
1779{
1780 xfs_icsb_cnts_t *cntp;
1781 int i;
1782
1783 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
1784 if (mp->m_sb_cnts == NULL)
1785 return -ENOMEM;
1786
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001787#ifdef CONFIG_HOTPLUG_CPU
David Chinnere8234a62006-03-14 13:23:52 +11001788 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
1789 mp->m_icsb_notifier.priority = 0;
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001790 register_hotcpu_notifier(&mp->m_icsb_notifier);
1791#endif /* CONFIG_HOTPLUG_CPU */
David Chinnere8234a62006-03-14 13:23:52 +11001792
David Chinner8d280b92006-03-14 13:13:09 +11001793 for_each_online_cpu(i) {
1794 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11001795 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinner8d280b92006-03-14 13:13:09 +11001796 }
David Chinner20b64282007-02-10 18:35:09 +11001797
1798 mutex_init(&mp->m_icsb_mutex);
1799
David Chinner8d280b92006-03-14 13:13:09 +11001800 /*
1801 * start with all counters disabled so that the
1802 * initial balance kicks us off correctly
1803 */
1804 mp->m_icsb_counters = -1;
1805 return 0;
1806}
1807
1808STATIC void
1809xfs_icsb_destroy_counters(
1810 xfs_mount_t *mp)
1811{
David Chinnere8234a62006-03-14 13:23:52 +11001812 if (mp->m_sb_cnts) {
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001813 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
David Chinner8d280b92006-03-14 13:13:09 +11001814 free_percpu(mp->m_sb_cnts);
David Chinnere8234a62006-03-14 13:23:52 +11001815 }
David Chinner03135cf2007-02-10 18:35:15 +11001816 mutex_destroy(&mp->m_icsb_mutex);
David Chinner8d280b92006-03-14 13:13:09 +11001817}
1818
David Chinner7989cb82007-02-10 18:34:56 +11001819STATIC_INLINE void
David Chinner01e1b692006-03-14 13:29:16 +11001820xfs_icsb_lock_cntr(
1821 xfs_icsb_cnts_t *icsbp)
1822{
1823 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
1824 ndelay(1000);
1825 }
1826}
1827
David Chinner7989cb82007-02-10 18:34:56 +11001828STATIC_INLINE void
David Chinner01e1b692006-03-14 13:29:16 +11001829xfs_icsb_unlock_cntr(
1830 xfs_icsb_cnts_t *icsbp)
1831{
1832 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
1833}
1834
David Chinner8d280b92006-03-14 13:13:09 +11001835
David Chinner7989cb82007-02-10 18:34:56 +11001836STATIC_INLINE void
David Chinner8d280b92006-03-14 13:13:09 +11001837xfs_icsb_lock_all_counters(
1838 xfs_mount_t *mp)
1839{
1840 xfs_icsb_cnts_t *cntp;
1841 int i;
1842
1843 for_each_online_cpu(i) {
1844 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11001845 xfs_icsb_lock_cntr(cntp);
David Chinner8d280b92006-03-14 13:13:09 +11001846 }
1847}
1848
David Chinner7989cb82007-02-10 18:34:56 +11001849STATIC_INLINE void
David Chinner8d280b92006-03-14 13:13:09 +11001850xfs_icsb_unlock_all_counters(
1851 xfs_mount_t *mp)
1852{
1853 xfs_icsb_cnts_t *cntp;
1854 int i;
1855
1856 for_each_online_cpu(i) {
1857 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11001858 xfs_icsb_unlock_cntr(cntp);
David Chinner8d280b92006-03-14 13:13:09 +11001859 }
1860}
1861
1862STATIC void
1863xfs_icsb_count(
1864 xfs_mount_t *mp,
1865 xfs_icsb_cnts_t *cnt,
1866 int flags)
1867{
1868 xfs_icsb_cnts_t *cntp;
1869 int i;
1870
1871 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
1872
1873 if (!(flags & XFS_ICSB_LAZY_COUNT))
1874 xfs_icsb_lock_all_counters(mp);
1875
1876 for_each_online_cpu(i) {
1877 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1878 cnt->icsb_icount += cntp->icsb_icount;
1879 cnt->icsb_ifree += cntp->icsb_ifree;
1880 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
1881 }
1882
1883 if (!(flags & XFS_ICSB_LAZY_COUNT))
1884 xfs_icsb_unlock_all_counters(mp);
1885}
1886
1887STATIC int
1888xfs_icsb_counter_disabled(
1889 xfs_mount_t *mp,
1890 xfs_sb_field_t field)
1891{
1892 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1893 return test_bit(field, &mp->m_icsb_counters);
1894}
1895
1896STATIC int
1897xfs_icsb_disable_counter(
1898 xfs_mount_t *mp,
1899 xfs_sb_field_t field)
1900{
1901 xfs_icsb_cnts_t cnt;
1902
1903 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1904
David Chinner20b64282007-02-10 18:35:09 +11001905 /*
1906 * If we are already disabled, then there is nothing to do
1907 * here. We check before locking all the counters to avoid
1908 * the expensive lock operation when being called in the
1909 * slow path and the counter is already disabled. This is
1910 * safe because the only time we set or clear this state is under
1911 * the m_icsb_mutex.
1912 */
1913 if (xfs_icsb_counter_disabled(mp, field))
1914 return 0;
1915
David Chinner8d280b92006-03-14 13:13:09 +11001916 xfs_icsb_lock_all_counters(mp);
1917 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
1918 /* drain back to superblock */
1919
1920 xfs_icsb_count(mp, &cnt, XFS_ICSB_SB_LOCKED|XFS_ICSB_LAZY_COUNT);
1921 switch(field) {
1922 case XFS_SBS_ICOUNT:
1923 mp->m_sb.sb_icount = cnt.icsb_icount;
1924 break;
1925 case XFS_SBS_IFREE:
1926 mp->m_sb.sb_ifree = cnt.icsb_ifree;
1927 break;
1928 case XFS_SBS_FDBLOCKS:
1929 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
1930 break;
1931 default:
1932 BUG();
1933 }
1934 }
1935
1936 xfs_icsb_unlock_all_counters(mp);
1937
1938 return 0;
1939}
1940
1941STATIC void
1942xfs_icsb_enable_counter(
1943 xfs_mount_t *mp,
1944 xfs_sb_field_t field,
1945 uint64_t count,
1946 uint64_t resid)
1947{
1948 xfs_icsb_cnts_t *cntp;
1949 int i;
1950
1951 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1952
1953 xfs_icsb_lock_all_counters(mp);
1954 for_each_online_cpu(i) {
1955 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
1956 switch (field) {
1957 case XFS_SBS_ICOUNT:
1958 cntp->icsb_icount = count + resid;
1959 break;
1960 case XFS_SBS_IFREE:
1961 cntp->icsb_ifree = count + resid;
1962 break;
1963 case XFS_SBS_FDBLOCKS:
1964 cntp->icsb_fdblocks = count + resid;
1965 break;
1966 default:
1967 BUG();
1968 break;
1969 }
1970 resid = 0;
1971 }
1972 clear_bit(field, &mp->m_icsb_counters);
1973 xfs_icsb_unlock_all_counters(mp);
1974}
1975
1976STATIC void
1977xfs_icsb_sync_counters_int(
1978 xfs_mount_t *mp,
1979 int flags)
1980{
1981 xfs_icsb_cnts_t cnt;
1982 int s;
1983
1984 /* Pass 1: lock all counters */
1985 if ((flags & XFS_ICSB_SB_LOCKED) == 0)
1986 s = XFS_SB_LOCK(mp);
1987
1988 xfs_icsb_count(mp, &cnt, flags);
1989
1990 /* Step 3: update mp->m_sb fields */
1991 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
1992 mp->m_sb.sb_icount = cnt.icsb_icount;
1993 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
1994 mp->m_sb.sb_ifree = cnt.icsb_ifree;
1995 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
1996 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
1997
1998 if ((flags & XFS_ICSB_SB_LOCKED) == 0)
1999 XFS_SB_UNLOCK(mp, s);
2000}
2001
2002/*
2003 * Accurate update of per-cpu counters to incore superblock
2004 */
2005STATIC void
2006xfs_icsb_sync_counters(
2007 xfs_mount_t *mp)
2008{
2009 xfs_icsb_sync_counters_int(mp, 0);
2010}
2011
2012/*
2013 * lazy addition used for things like df, background sb syncs, etc
2014 */
2015void
2016xfs_icsb_sync_counters_lazy(
2017 xfs_mount_t *mp)
2018{
2019 xfs_icsb_sync_counters_int(mp, XFS_ICSB_LAZY_COUNT);
2020}
2021
2022/*
2023 * Balance and enable/disable counters as necessary.
2024 *
David Chinner20b64282007-02-10 18:35:09 +11002025 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2026 * chosen to be the same number as single on disk allocation chunk per CPU, and
2027 * free blocks is something far enough zero that we aren't going thrash when we
2028 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2029 * prevent looping endlessly when xfs_alloc_space asks for more than will
2030 * be distributed to a single CPU but each CPU has enough blocks to be
2031 * reenabled.
2032 *
2033 * Note that we can be called when counters are already disabled.
2034 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2035 * prevent locking every per-cpu counter needlessly.
David Chinner8d280b92006-03-14 13:13:09 +11002036 */
David Chinner20b64282007-02-10 18:35:09 +11002037
2038#define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
David Chinner4be536d2006-09-07 14:26:50 +10002039#define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
David Chinner20b64282007-02-10 18:35:09 +11002040 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
David Chinner8d280b92006-03-14 13:13:09 +11002041STATIC void
2042xfs_icsb_balance_counter(
2043 xfs_mount_t *mp,
2044 xfs_sb_field_t field,
David Chinner20b64282007-02-10 18:35:09 +11002045 int flags,
2046 int min_per_cpu)
David Chinner8d280b92006-03-14 13:13:09 +11002047{
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002048 uint64_t count, resid;
David Chinner8d280b92006-03-14 13:13:09 +11002049 int weight = num_online_cpus();
2050 int s;
David Chinner20b64282007-02-10 18:35:09 +11002051 uint64_t min = (uint64_t)min_per_cpu;
David Chinner8d280b92006-03-14 13:13:09 +11002052
2053 if (!(flags & XFS_ICSB_SB_LOCKED))
2054 s = XFS_SB_LOCK(mp);
2055
2056 /* disable counter and sync counter */
2057 xfs_icsb_disable_counter(mp, field);
2058
2059 /* update counters - first CPU gets residual*/
2060 switch (field) {
2061 case XFS_SBS_ICOUNT:
2062 count = mp->m_sb.sb_icount;
2063 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002064 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
David Chinner8d280b92006-03-14 13:13:09 +11002065 goto out;
2066 break;
2067 case XFS_SBS_IFREE:
2068 count = mp->m_sb.sb_ifree;
2069 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002070 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
David Chinner8d280b92006-03-14 13:13:09 +11002071 goto out;
2072 break;
2073 case XFS_SBS_FDBLOCKS:
2074 count = mp->m_sb.sb_fdblocks;
2075 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002076 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
David Chinner8d280b92006-03-14 13:13:09 +11002077 goto out;
2078 break;
2079 default:
2080 BUG();
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002081 count = resid = 0; /* quiet, gcc */
David Chinner8d280b92006-03-14 13:13:09 +11002082 break;
2083 }
2084
2085 xfs_icsb_enable_counter(mp, field, count, resid);
2086out:
2087 if (!(flags & XFS_ICSB_SB_LOCKED))
2088 XFS_SB_UNLOCK(mp, s);
2089}
2090
David Chinner20b64282007-02-10 18:35:09 +11002091int
2092xfs_icsb_modify_counters(
David Chinner8d280b92006-03-14 13:13:09 +11002093 xfs_mount_t *mp,
2094 xfs_sb_field_t field,
2095 int delta,
David Chinner20b64282007-02-10 18:35:09 +11002096 int rsvd)
David Chinner8d280b92006-03-14 13:13:09 +11002097{
2098 xfs_icsb_cnts_t *icsbp;
2099 long long lcounter; /* long counter for 64 bit fields */
David Chinner20b64282007-02-10 18:35:09 +11002100 int cpu, ret = 0, s;
David Chinner8d280b92006-03-14 13:13:09 +11002101
David Chinner20b64282007-02-10 18:35:09 +11002102 might_sleep();
David Chinner8d280b92006-03-14 13:13:09 +11002103again:
2104 cpu = get_cpu();
David Chinner20b64282007-02-10 18:35:09 +11002105 icsbp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, cpu);
2106
2107 /*
2108 * if the counter is disabled, go to slow path
2109 */
David Chinner8d280b92006-03-14 13:13:09 +11002110 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2111 goto slow_path;
David Chinner20b64282007-02-10 18:35:09 +11002112 xfs_icsb_lock_cntr(icsbp);
2113 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2114 xfs_icsb_unlock_cntr(icsbp);
2115 goto slow_path;
2116 }
David Chinner8d280b92006-03-14 13:13:09 +11002117
2118 switch (field) {
2119 case XFS_SBS_ICOUNT:
2120 lcounter = icsbp->icsb_icount;
2121 lcounter += delta;
2122 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002123 goto balance_counter;
David Chinner8d280b92006-03-14 13:13:09 +11002124 icsbp->icsb_icount = lcounter;
2125 break;
2126
2127 case XFS_SBS_IFREE:
2128 lcounter = icsbp->icsb_ifree;
2129 lcounter += delta;
2130 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002131 goto balance_counter;
David Chinner8d280b92006-03-14 13:13:09 +11002132 icsbp->icsb_ifree = lcounter;
2133 break;
2134
2135 case XFS_SBS_FDBLOCKS:
2136 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2137
David Chinner4be536d2006-09-07 14:26:50 +10002138 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002139 lcounter += delta;
2140 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002141 goto balance_counter;
David Chinner4be536d2006-09-07 14:26:50 +10002142 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002143 break;
2144 default:
2145 BUG();
2146 break;
2147 }
David Chinner01e1b692006-03-14 13:29:16 +11002148 xfs_icsb_unlock_cntr(icsbp);
David Chinner8d280b92006-03-14 13:13:09 +11002149 put_cpu();
David Chinner8d280b92006-03-14 13:13:09 +11002150 return 0;
2151
David Chinner8d280b92006-03-14 13:13:09 +11002152slow_path:
David Chinner20b64282007-02-10 18:35:09 +11002153 put_cpu();
2154
2155 /*
2156 * serialise with a mutex so we don't burn lots of cpu on
2157 * the superblock lock. We still need to hold the superblock
2158 * lock, however, when we modify the global structures.
2159 */
David Chinner03135cf2007-02-10 18:35:15 +11002160 xfs_icsb_lock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002161
2162 /*
2163 * Now running atomically.
2164 *
2165 * If the counter is enabled, someone has beaten us to rebalancing.
2166 * Drop the lock and try again in the fast path....
2167 */
2168 if (!(xfs_icsb_counter_disabled(mp, field))) {
David Chinner03135cf2007-02-10 18:35:15 +11002169 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002170 goto again;
2171 }
2172
2173 /*
2174 * The counter is currently disabled. Because we are
2175 * running atomically here, we know a rebalance cannot
2176 * be in progress. Hence we can go straight to operating
2177 * on the global superblock. We do not call xfs_mod_incore_sb()
2178 * here even though we need to get the SB_LOCK. Doing so
2179 * will cause us to re-enter this function and deadlock.
2180 * Hence we get the SB_LOCK ourselves and then call
2181 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2182 * directly on the global counters.
2183 */
2184 s = XFS_SB_LOCK(mp);
2185 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2186 XFS_SB_UNLOCK(mp, s);
2187
2188 /*
2189 * Now that we've modified the global superblock, we
2190 * may be able to re-enable the distributed counters
2191 * (e.g. lots of space just got freed). After that
2192 * we are done.
2193 */
2194 if (ret != ENOSPC)
2195 xfs_icsb_balance_counter(mp, field, 0, 0);
David Chinner03135cf2007-02-10 18:35:15 +11002196 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002197 return ret;
2198
2199balance_counter:
David Chinner01e1b692006-03-14 13:29:16 +11002200 xfs_icsb_unlock_cntr(icsbp);
David Chinner8d280b92006-03-14 13:13:09 +11002201 put_cpu();
2202
David Chinner20b64282007-02-10 18:35:09 +11002203 /*
2204 * We may have multiple threads here if multiple per-cpu
2205 * counters run dry at the same time. This will mean we can
2206 * do more balances than strictly necessary but it is not
2207 * the common slowpath case.
2208 */
David Chinner03135cf2007-02-10 18:35:15 +11002209 xfs_icsb_lock(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002210
David Chinner20b64282007-02-10 18:35:09 +11002211 /*
2212 * running atomically.
2213 *
2214 * This will leave the counter in the correct state for future
2215 * accesses. After the rebalance, we simply try again and our retry
2216 * will either succeed through the fast path or slow path without
2217 * another balance operation being required.
2218 */
2219 xfs_icsb_balance_counter(mp, field, 0, delta);
David Chinner03135cf2007-02-10 18:35:15 +11002220 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002221 goto again;
David Chinner8d280b92006-03-14 13:13:09 +11002222}
2223
David Chinner8d280b92006-03-14 13:13:09 +11002224#endif