blob: 65a99725d0cc589d9e1b96690a2c12448849e249 [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"
Christoph Hellwig43355092008-03-27 18:01:08 +110046#include "xfs_utils.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
Christoph Hellwigba0f32d2005-06-21 15:36:52 +100048STATIC void xfs_unmountfs_wait(xfs_mount_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
David Chinner8d280b92006-03-14 13:13:09 +110050
51#ifdef HAVE_PERCPU_SB
David Chinner20f4ebf2007-02-10 18:36:10 +110052STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
Christoph Hellwig45af6c62008-04-22 17:34:44 +100053 int);
54STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t,
55 int);
David Chinner8d280b92006-03-14 13:13:09 +110056STATIC int xfs_icsb_modify_counters(xfs_mount_t *, xfs_sb_field_t,
David Chinner20f4ebf2007-02-10 18:36:10 +110057 int64_t, int);
David Chinner36fbe6e2008-04-10 12:19:56 +100058STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
David Chinner8d280b92006-03-14 13:13:09 +110059
60#else
61
Christoph Hellwig45af6c62008-04-22 17:34:44 +100062#define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
63#define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
David Chinner8d280b92006-03-14 13:13:09 +110064#define xfs_icsb_modify_counters(mp, a, b, c) do { } while (0)
David Chinner8d280b92006-03-14 13:13:09 +110065
66#endif
67
Christoph Hellwig1df84c92006-01-11 15:29:52 +110068static const struct {
David Chinner8d280b92006-03-14 13:13:09 +110069 short offset;
70 short type; /* 0 = integer
71 * 1 = binary / string (no translation)
72 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070073} xfs_sb_info[] = {
74 { offsetof(xfs_sb_t, sb_magicnum), 0 },
75 { offsetof(xfs_sb_t, sb_blocksize), 0 },
76 { offsetof(xfs_sb_t, sb_dblocks), 0 },
77 { offsetof(xfs_sb_t, sb_rblocks), 0 },
78 { offsetof(xfs_sb_t, sb_rextents), 0 },
79 { offsetof(xfs_sb_t, sb_uuid), 1 },
80 { offsetof(xfs_sb_t, sb_logstart), 0 },
81 { offsetof(xfs_sb_t, sb_rootino), 0 },
82 { offsetof(xfs_sb_t, sb_rbmino), 0 },
83 { offsetof(xfs_sb_t, sb_rsumino), 0 },
84 { offsetof(xfs_sb_t, sb_rextsize), 0 },
85 { offsetof(xfs_sb_t, sb_agblocks), 0 },
86 { offsetof(xfs_sb_t, sb_agcount), 0 },
87 { offsetof(xfs_sb_t, sb_rbmblocks), 0 },
88 { offsetof(xfs_sb_t, sb_logblocks), 0 },
89 { offsetof(xfs_sb_t, sb_versionnum), 0 },
90 { offsetof(xfs_sb_t, sb_sectsize), 0 },
91 { offsetof(xfs_sb_t, sb_inodesize), 0 },
92 { offsetof(xfs_sb_t, sb_inopblock), 0 },
93 { offsetof(xfs_sb_t, sb_fname[0]), 1 },
94 { offsetof(xfs_sb_t, sb_blocklog), 0 },
95 { offsetof(xfs_sb_t, sb_sectlog), 0 },
96 { offsetof(xfs_sb_t, sb_inodelog), 0 },
97 { offsetof(xfs_sb_t, sb_inopblog), 0 },
98 { offsetof(xfs_sb_t, sb_agblklog), 0 },
99 { offsetof(xfs_sb_t, sb_rextslog), 0 },
100 { offsetof(xfs_sb_t, sb_inprogress), 0 },
101 { offsetof(xfs_sb_t, sb_imax_pct), 0 },
102 { offsetof(xfs_sb_t, sb_icount), 0 },
103 { offsetof(xfs_sb_t, sb_ifree), 0 },
104 { offsetof(xfs_sb_t, sb_fdblocks), 0 },
105 { offsetof(xfs_sb_t, sb_frextents), 0 },
106 { offsetof(xfs_sb_t, sb_uquotino), 0 },
107 { offsetof(xfs_sb_t, sb_gquotino), 0 },
108 { offsetof(xfs_sb_t, sb_qflags), 0 },
109 { offsetof(xfs_sb_t, sb_flags), 0 },
110 { offsetof(xfs_sb_t, sb_shared_vn), 0 },
111 { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
112 { offsetof(xfs_sb_t, sb_unit), 0 },
113 { offsetof(xfs_sb_t, sb_width), 0 },
114 { offsetof(xfs_sb_t, sb_dirblklog), 0 },
115 { offsetof(xfs_sb_t, sb_logsectlog), 0 },
116 { offsetof(xfs_sb_t, sb_logsectsize),0 },
117 { offsetof(xfs_sb_t, sb_logsunit), 0 },
118 { offsetof(xfs_sb_t, sb_features2), 0 },
David Chinneree1c0902008-03-06 13:45:50 +1100119 { offsetof(xfs_sb_t, sb_bad_features2), 0 },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 { sizeof(xfs_sb_t), 0 }
121};
122
Christoph Hellwig27174202009-03-30 10:21:31 +0200123static DEFINE_MUTEX(xfs_uuid_table_mutex);
124static int xfs_uuid_table_size;
125static uuid_t *xfs_uuid_table;
126
127/*
128 * See if the UUID is unique among mounted XFS filesystems.
129 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
130 */
131STATIC int
132xfs_uuid_mount(
133 struct xfs_mount *mp)
134{
135 uuid_t *uuid = &mp->m_sb.sb_uuid;
136 int hole, i;
137
138 if (mp->m_flags & XFS_MOUNT_NOUUID)
139 return 0;
140
141 if (uuid_is_nil(uuid)) {
142 cmn_err(CE_WARN,
143 "XFS: Filesystem %s has nil UUID - can't mount",
144 mp->m_fsname);
145 return XFS_ERROR(EINVAL);
146 }
147
148 mutex_lock(&xfs_uuid_table_mutex);
149 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
150 if (uuid_is_nil(&xfs_uuid_table[i])) {
151 hole = i;
152 continue;
153 }
154 if (uuid_equal(uuid, &xfs_uuid_table[i]))
155 goto out_duplicate;
156 }
157
158 if (hole < 0) {
159 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
160 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
161 xfs_uuid_table_size * sizeof(*xfs_uuid_table),
162 KM_SLEEP);
163 hole = xfs_uuid_table_size++;
164 }
165 xfs_uuid_table[hole] = *uuid;
166 mutex_unlock(&xfs_uuid_table_mutex);
167
168 return 0;
169
170 out_duplicate:
171 mutex_unlock(&xfs_uuid_table_mutex);
172 cmn_err(CE_WARN, "XFS: Filesystem %s has duplicate UUID - can't mount",
173 mp->m_fsname);
174 return XFS_ERROR(EINVAL);
175}
176
177STATIC void
178xfs_uuid_unmount(
179 struct xfs_mount *mp)
180{
181 uuid_t *uuid = &mp->m_sb.sb_uuid;
182 int i;
183
184 if (mp->m_flags & XFS_MOUNT_NOUUID)
185 return;
186
187 mutex_lock(&xfs_uuid_table_mutex);
188 for (i = 0; i < xfs_uuid_table_size; i++) {
189 if (uuid_is_nil(&xfs_uuid_table[i]))
190 continue;
191 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
192 continue;
193 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
194 break;
195 }
196 ASSERT(i < xfs_uuid_table_size);
197 mutex_unlock(&xfs_uuid_table_mutex);
198}
199
200
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 * Free up the resources associated with a mount structure. Assume that
203 * the structure was initially zeroed, so we can tell which fields got
204 * initialized.
205 */
Christoph Hellwigc962fb72008-05-20 15:10:52 +1000206STATIC void
Christoph Hellwigff4f0382008-08-13 16:50:47 +1000207xfs_free_perag(
Christoph Hellwig745f6912007-08-30 17:20:39 +1000208 xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 if (mp->m_perag) {
211 int agno;
212
213 for (agno = 0; agno < mp->m_maxagi; agno++)
214 if (mp->m_perag[agno].pagb_list)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +1000215 kmem_free(mp->m_perag[agno].pagb_list);
216 kmem_free(mp->m_perag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218}
219
Nathan Scott4cc929e2007-05-14 18:24:02 +1000220/*
221 * Check size of device based on the (data/realtime) block count.
222 * Note: this check is used by the growfs code as well as mount.
223 */
224int
225xfs_sb_validate_fsb_count(
226 xfs_sb_t *sbp,
227 __uint64_t nblocks)
228{
229 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
230 ASSERT(sbp->sb_blocklog >= BBSHIFT);
231
232#if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
233 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
234 return E2BIG;
235#else /* Limited by UINT_MAX of sectors */
236 if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
237 return E2BIG;
238#endif
239 return 0;
240}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
242/*
243 * Check the validity of the SB found.
244 */
245STATIC int
246xfs_mount_validate_sb(
247 xfs_mount_t *mp,
Nathan Scott764d1f82006-03-31 13:04:17 +1000248 xfs_sb_t *sbp,
249 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250{
251 /*
252 * If the log device and data device have the
253 * same device number, the log is internal.
254 * Consequently, the sb_logstart should be non-zero. If
255 * we have a zero sb_logstart in this case, we may be trying to mount
256 * a volume filesystem in a non-volume manner.
257 */
258 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000259 xfs_fs_mount_cmn_err(flags, "bad magic number");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 return XFS_ERROR(EWRONGFS);
261 }
262
Eric Sandeen62118702008-03-06 13:44:28 +1100263 if (!xfs_sb_good_version(sbp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000264 xfs_fs_mount_cmn_err(flags, "bad version");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 return XFS_ERROR(EWRONGFS);
266 }
267
268 if (unlikely(
269 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000270 xfs_fs_mount_cmn_err(flags,
271 "filesystem is marked as having an external log; "
272 "specify logdev on the\nmount command line.");
273 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 }
275
276 if (unlikely(
277 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000278 xfs_fs_mount_cmn_err(flags,
279 "filesystem is marked as having an internal log; "
280 "do not specify logdev on\nthe mount command line.");
281 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 }
283
284 /*
285 * More sanity checking. These were stolen directly from
286 * xfs_repair.
287 */
288 if (unlikely(
289 sbp->sb_agcount <= 0 ||
290 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
291 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
292 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
293 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
Olaf Weberb9ec9062009-04-17 16:12:45 -0500294 sbp->sb_sectsize != (1 << sbp->sb_sectlog) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
296 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
297 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
298 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
Olaf Weberb9ec9062009-04-17 16:12:45 -0500299 sbp->sb_blocksize != (1 << sbp->sb_blocklog) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
301 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
Nathan Scott9f989c92006-03-14 13:29:32 +1100302 sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
303 sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
Olaf Weberb9ec9062009-04-17 16:12:45 -0500304 sbp->sb_inodesize != (1 << sbp->sb_inodelog) ||
Nathan Scott9f989c92006-03-14 13:29:32 +1100305 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
307 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
Nathan Scotte50bd162006-04-11 15:10:45 +1000308 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000309 xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 return XFS_ERROR(EFSCORRUPTED);
311 }
312
313 /*
314 * Sanity check AG count, size fields against data size field
315 */
316 if (unlikely(
317 sbp->sb_dblocks == 0 ||
318 sbp->sb_dblocks >
319 (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
320 sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
321 sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000322 xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 return XFS_ERROR(EFSCORRUPTED);
324 }
325
Lachlan McIlroy2edbddd2008-06-27 13:34:34 +1000326 /*
327 * Until this is fixed only page-sized or smaller data blocks work.
328 */
329 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
330 xfs_fs_mount_cmn_err(flags,
331 "file system with blocksize %d bytes",
332 sbp->sb_blocksize);
333 xfs_fs_mount_cmn_err(flags,
334 "only pagesize (%ld) or less will currently work.",
335 PAGE_SIZE);
336 return XFS_ERROR(ENOSYS);
337 }
338
Christoph Hellwig1a5902c2009-03-29 19:26:46 +0200339 /*
340 * Currently only very few inode sizes are supported.
341 */
342 switch (sbp->sb_inodesize) {
343 case 256:
344 case 512:
345 case 1024:
346 case 2048:
347 break;
348 default:
349 xfs_fs_mount_cmn_err(flags,
350 "inode size of %d bytes not supported",
351 sbp->sb_inodesize);
352 return XFS_ERROR(ENOSYS);
353 }
354
Nathan Scott4cc929e2007-05-14 18:24:02 +1000355 if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
356 xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000357 xfs_fs_mount_cmn_err(flags,
358 "file system too large to be mounted on this system.");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 return XFS_ERROR(E2BIG);
360 }
361
362 if (unlikely(sbp->sb_inprogress)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000363 xfs_fs_mount_cmn_err(flags, "file system busy");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 return XFS_ERROR(EFSCORRUPTED);
365 }
366
367 /*
Nathan Scottde206142005-05-05 13:24:13 -0700368 * Version 1 directory format has never worked on Linux.
369 */
Eric Sandeen62118702008-03-06 13:44:28 +1100370 if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000371 xfs_fs_mount_cmn_err(flags,
372 "file system using version 1 directory format");
Nathan Scottde206142005-05-05 13:24:13 -0700373 return XFS_ERROR(ENOSYS);
374 }
375
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 return 0;
377}
378
David Chinnerda353b02007-08-28 14:00:13 +1000379STATIC void
380xfs_initialize_perag_icache(
381 xfs_perag_t *pag)
382{
383 if (!pag->pag_ici_init) {
384 rwlock_init(&pag->pag_ici_lock);
385 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
386 pag->pag_ici_init = 1;
387 }
388}
389
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390xfs_agnumber_t
Nathan Scottc11e2c32005-11-02 15:11:45 +1100391xfs_initialize_perag(
Nathan Scottc11e2c32005-11-02 15:11:45 +1100392 xfs_mount_t *mp,
393 xfs_agnumber_t agcount)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394{
395 xfs_agnumber_t index, max_metadata;
396 xfs_perag_t *pag;
397 xfs_agino_t agino;
398 xfs_ino_t ino;
399 xfs_sb_t *sbp = &mp->m_sb;
400 xfs_ino_t max_inum = XFS_MAXINUMBER_32;
401
402 /* Check to see if the filesystem can overflow 32 bit inodes */
403 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
404 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
405
406 /* Clear the mount flag if no inode can overflow 32 bits
407 * on this filesystem, or if specifically requested..
408 */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000409 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > max_inum) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 mp->m_flags |= XFS_MOUNT_32BITINODES;
411 } else {
412 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
413 }
414
415 /* If we can overflow then setup the ag headers accordingly */
416 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
417 /* Calculate how much should be reserved for inodes to
418 * meet the max inode percentage.
419 */
420 if (mp->m_maxicount) {
421 __uint64_t icount;
422
423 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
424 do_div(icount, 100);
425 icount += sbp->sb_agblocks - 1;
Eric Sandeena749ee82005-11-02 15:13:42 +1100426 do_div(icount, sbp->sb_agblocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 max_metadata = icount;
428 } else {
429 max_metadata = agcount;
430 }
431 for (index = 0; index < agcount; index++) {
432 ino = XFS_AGINO_TO_INO(mp, index, agino);
433 if (ino > max_inum) {
434 index++;
435 break;
436 }
437
Nathan Scottc41564b2006-03-29 08:55:14 +1000438 /* This ag is preferred for inodes */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 pag = &mp->m_perag[index];
440 pag->pagi_inodeok = 1;
441 if (index < max_metadata)
442 pag->pagf_metadata = 1;
David Chinnerda353b02007-08-28 14:00:13 +1000443 xfs_initialize_perag_icache(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 }
445 } else {
446 /* Setup default behavior for smaller filesystems */
447 for (index = 0; index < agcount; index++) {
448 pag = &mp->m_perag[index];
449 pag->pagi_inodeok = 1;
David Chinnerda353b02007-08-28 14:00:13 +1000450 xfs_initialize_perag_icache(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 }
452 }
453 return index;
454}
455
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000456void
457xfs_sb_from_disk(
458 xfs_sb_t *to,
459 xfs_dsb_t *from)
460{
461 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
462 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
463 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
464 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
465 to->sb_rextents = be64_to_cpu(from->sb_rextents);
466 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
467 to->sb_logstart = be64_to_cpu(from->sb_logstart);
468 to->sb_rootino = be64_to_cpu(from->sb_rootino);
469 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
470 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
471 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
472 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
473 to->sb_agcount = be32_to_cpu(from->sb_agcount);
474 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
475 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
476 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
477 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
478 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
479 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
480 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
481 to->sb_blocklog = from->sb_blocklog;
482 to->sb_sectlog = from->sb_sectlog;
483 to->sb_inodelog = from->sb_inodelog;
484 to->sb_inopblog = from->sb_inopblog;
485 to->sb_agblklog = from->sb_agblklog;
486 to->sb_rextslog = from->sb_rextslog;
487 to->sb_inprogress = from->sb_inprogress;
488 to->sb_imax_pct = from->sb_imax_pct;
489 to->sb_icount = be64_to_cpu(from->sb_icount);
490 to->sb_ifree = be64_to_cpu(from->sb_ifree);
491 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
492 to->sb_frextents = be64_to_cpu(from->sb_frextents);
493 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
494 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
495 to->sb_qflags = be16_to_cpu(from->sb_qflags);
496 to->sb_flags = from->sb_flags;
497 to->sb_shared_vn = from->sb_shared_vn;
498 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
499 to->sb_unit = be32_to_cpu(from->sb_unit);
500 to->sb_width = be32_to_cpu(from->sb_width);
501 to->sb_dirblklog = from->sb_dirblklog;
502 to->sb_logsectlog = from->sb_logsectlog;
503 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
504 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
505 to->sb_features2 = be32_to_cpu(from->sb_features2);
David Chinneree1c0902008-03-06 13:45:50 +1100506 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000507}
508
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509/*
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000510 * Copy in core superblock to ondisk one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 *
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000512 * The fields argument is mask of superblock fields to copy.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 */
514void
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000515xfs_sb_to_disk(
516 xfs_dsb_t *to,
517 xfs_sb_t *from,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 __int64_t fields)
519{
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000520 xfs_caddr_t to_ptr = (xfs_caddr_t)to;
521 xfs_caddr_t from_ptr = (xfs_caddr_t)from;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 xfs_sb_field_t f;
523 int first;
524 int size;
525
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 ASSERT(fields);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 if (!fields)
528 return;
529
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 while (fields) {
531 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
532 first = xfs_sb_info[f].offset;
533 size = xfs_sb_info[f + 1].offset - first;
534
535 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
536
537 if (size == 1 || xfs_sb_info[f].type == 1) {
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000538 memcpy(to_ptr + first, from_ptr + first, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 } else {
540 switch (size) {
541 case 2:
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000542 *(__be16 *)(to_ptr + first) =
543 cpu_to_be16(*(__u16 *)(from_ptr + first));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 break;
545 case 4:
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000546 *(__be32 *)(to_ptr + first) =
547 cpu_to_be32(*(__u32 *)(from_ptr + first));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 break;
549 case 8:
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000550 *(__be64 *)(to_ptr + first) =
551 cpu_to_be64(*(__u64 *)(from_ptr + first));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 break;
553 default:
554 ASSERT(0);
555 }
556 }
557
558 fields &= ~(1LL << f);
559 }
560}
561
562/*
563 * xfs_readsb
564 *
565 * Does the initial read of the superblock.
566 */
567int
Nathan Scott764d1f82006-03-31 13:04:17 +1000568xfs_readsb(xfs_mount_t *mp, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569{
570 unsigned int sector_size;
571 unsigned int extra_flags;
572 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 int error;
574
575 ASSERT(mp->m_sb_bp == NULL);
576 ASSERT(mp->m_ddev_targp != NULL);
577
578 /*
579 * Allocate a (locked) buffer to hold the superblock.
580 * This will be kept around at all times to optimize
581 * access to the superblock.
582 */
583 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
584 extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED;
585
586 bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
587 BTOBB(sector_size), extra_flags);
588 if (!bp || XFS_BUF_ISERROR(bp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000589 xfs_fs_mount_cmn_err(flags, "SB read failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
591 goto fail;
592 }
593 ASSERT(XFS_BUF_ISBUSY(bp));
594 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
595
596 /*
597 * Initialize the mount structure from the superblock.
598 * But first do some basic consistency checking.
599 */
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000600 xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601
Nathan Scott764d1f82006-03-31 13:04:17 +1000602 error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 if (error) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000604 xfs_fs_mount_cmn_err(flags, "SB validate failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 goto fail;
606 }
607
608 /*
609 * We must be able to do sector-sized and sector-aligned IO.
610 */
611 if (sector_size > mp->m_sb.sb_sectsize) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000612 xfs_fs_mount_cmn_err(flags,
613 "device supports only %u byte sectors (not %u)",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 sector_size, mp->m_sb.sb_sectsize);
615 error = ENOSYS;
616 goto fail;
617 }
618
619 /*
620 * If device sector size is smaller than the superblock size,
621 * re-read the superblock so the buffer is correctly sized.
622 */
623 if (sector_size < mp->m_sb.sb_sectsize) {
624 XFS_BUF_UNMANAGE(bp);
625 xfs_buf_relse(bp);
626 sector_size = mp->m_sb.sb_sectsize;
627 bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
628 BTOBB(sector_size), extra_flags);
629 if (!bp || XFS_BUF_ISERROR(bp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000630 xfs_fs_mount_cmn_err(flags, "SB re-read failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
632 goto fail;
633 }
634 ASSERT(XFS_BUF_ISBUSY(bp));
635 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
636 }
637
Lachlan McIlroy5478eea2007-02-10 18:36:29 +1100638 /* Initialize per-cpu counters */
639 xfs_icsb_reinit_counters(mp);
David Chinner8d280b92006-03-14 13:13:09 +1100640
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 mp->m_sb_bp = bp;
642 xfs_buf_relse(bp);
643 ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
644 return 0;
645
646 fail:
647 if (bp) {
648 XFS_BUF_UNMANAGE(bp);
649 xfs_buf_relse(bp);
650 }
651 return error;
652}
653
654
655/*
656 * xfs_mount_common
657 *
658 * Mount initialization code establishing various mount
659 * fields from the superblock associated with the given
660 * mount structure
661 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000662STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
664{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 mp->m_agfrotor = mp->m_agirotor = 0;
Eric Sandeen007c61c2007-10-11 17:43:56 +1000666 spin_lock_init(&mp->m_agirotor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 mp->m_maxagi = mp->m_sb.sb_agcount;
668 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
669 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
670 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
671 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
672 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 mp->m_blockmask = sbp->sb_blocksize - 1;
674 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
675 mp->m_blockwmask = mp->m_blockwsize - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
Christoph Hellwig60197e82008-10-30 17:11:19 +1100677 mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
678 mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
679 mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
680 mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
681
682 mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
683 mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
684 mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
685 mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
686
687 mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
688 mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
689 mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
690 mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691
692 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
693 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
694 sbp->sb_inopblock);
695 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
696}
David Chinner92821e22007-05-24 15:26:31 +1000697
698/*
699 * xfs_initialize_perag_data
700 *
701 * Read in each per-ag structure so we can count up the number of
702 * allocated inodes, free inodes and used filesystem blocks as this
703 * information is no longer persistent in the superblock. Once we have
704 * this information, write it into the in-core superblock structure.
705 */
706STATIC int
707xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
708{
709 xfs_agnumber_t index;
710 xfs_perag_t *pag;
711 xfs_sb_t *sbp = &mp->m_sb;
712 uint64_t ifree = 0;
713 uint64_t ialloc = 0;
714 uint64_t bfree = 0;
715 uint64_t bfreelst = 0;
716 uint64_t btree = 0;
717 int error;
David Chinner92821e22007-05-24 15:26:31 +1000718
719 for (index = 0; index < agcount; index++) {
720 /*
721 * read the agf, then the agi. This gets us
Malcolm Parsons9da096f2009-03-29 09:55:42 +0200722 * all the information we need and populates the
David Chinner92821e22007-05-24 15:26:31 +1000723 * per-ag structures for us.
724 */
725 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
726 if (error)
727 return error;
728
729 error = xfs_ialloc_pagi_init(mp, NULL, index);
730 if (error)
731 return error;
732 pag = &mp->m_perag[index];
733 ifree += pag->pagi_freecount;
734 ialloc += pag->pagi_count;
735 bfree += pag->pagf_freeblks;
736 bfreelst += pag->pagf_flcount;
737 btree += pag->pagf_btreeblks;
738 }
739 /*
740 * Overwrite incore superblock counters with just-read data
741 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +1000742 spin_lock(&mp->m_sb_lock);
David Chinner92821e22007-05-24 15:26:31 +1000743 sbp->sb_ifree = ifree;
744 sbp->sb_icount = ialloc;
745 sbp->sb_fdblocks = bfree + bfreelst + btree;
Eric Sandeen3685c2a2007-10-11 17:42:32 +1000746 spin_unlock(&mp->m_sb_lock);
David Chinner92821e22007-05-24 15:26:31 +1000747
748 /* Fixup the per-cpu counters as well. */
749 xfs_icsb_reinit_counters(mp);
750
751 return 0;
752}
753
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754/*
Eric Sandeen0771fb42007-10-12 11:03:40 +1000755 * Update alignment values based on mount options and sb values
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 */
Eric Sandeen0771fb42007-10-12 11:03:40 +1000757STATIC int
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100758xfs_update_alignment(xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 xfs_sb_t *sbp = &(mp->m_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761
Christoph Hellwig42490232008-08-13 16:49:32 +1000762 if (mp->m_dalign) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 /*
764 * If stripe unit and stripe width are not multiples
765 * of the fs blocksize turn off alignment.
766 */
767 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
768 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
769 if (mp->m_flags & XFS_MOUNT_RETERR) {
770 cmn_err(CE_WARN,
771 "XFS: alignment check 1 failed");
Eric Sandeen0771fb42007-10-12 11:03:40 +1000772 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 }
774 mp->m_dalign = mp->m_swidth = 0;
775 } else {
776 /*
777 * Convert the stripe unit and width to FSBs.
778 */
779 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
780 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
781 if (mp->m_flags & XFS_MOUNT_RETERR) {
Eric Sandeen0771fb42007-10-12 11:03:40 +1000782 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 }
784 xfs_fs_cmn_err(CE_WARN, mp,
785"stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
786 mp->m_dalign, mp->m_swidth,
787 sbp->sb_agblocks);
788
789 mp->m_dalign = 0;
790 mp->m_swidth = 0;
791 } else if (mp->m_dalign) {
792 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
793 } else {
794 if (mp->m_flags & XFS_MOUNT_RETERR) {
795 xfs_fs_cmn_err(CE_WARN, mp,
796"stripe alignment turned off: sunit(%d) less than bsize(%d)",
797 mp->m_dalign,
798 mp->m_blockmask +1);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000799 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 }
801 mp->m_swidth = 0;
802 }
803 }
804
805 /*
806 * Update superblock with new values
807 * and log changes
808 */
Eric Sandeen62118702008-03-06 13:44:28 +1100809 if (xfs_sb_version_hasdalign(sbp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 if (sbp->sb_unit != mp->m_dalign) {
811 sbp->sb_unit = mp->m_dalign;
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100812 mp->m_update_flags |= XFS_SB_UNIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 }
814 if (sbp->sb_width != mp->m_swidth) {
815 sbp->sb_width = mp->m_swidth;
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100816 mp->m_update_flags |= XFS_SB_WIDTH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 }
818 }
819 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
Eric Sandeen62118702008-03-06 13:44:28 +1100820 xfs_sb_version_hasdalign(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 mp->m_dalign = sbp->sb_unit;
822 mp->m_swidth = sbp->sb_width;
823 }
824
Eric Sandeen0771fb42007-10-12 11:03:40 +1000825 return 0;
826}
827
828/*
829 * Set the maximum inode count for this filesystem
830 */
831STATIC void
832xfs_set_maxicount(xfs_mount_t *mp)
833{
834 xfs_sb_t *sbp = &(mp->m_sb);
835 __uint64_t icount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836
837 if (sbp->sb_imax_pct) {
Eric Sandeen0771fb42007-10-12 11:03:40 +1000838 /*
839 * Make sure the maximum inode count is a multiple
840 * of the units we allocate inodes in.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
843 do_div(icount, 100);
844 do_div(icount, mp->m_ialloc_blks);
845 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
846 sbp->sb_inopblog;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000847 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 mp->m_maxicount = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 }
Eric Sandeen0771fb42007-10-12 11:03:40 +1000850}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
Eric Sandeen0771fb42007-10-12 11:03:40 +1000852/*
853 * Set the default minimum read and write sizes unless
854 * already specified in a mount option.
855 * We use smaller I/O sizes when the file system
856 * is being used for NFS service (wsync mount option).
857 */
858STATIC void
859xfs_set_rw_sizes(xfs_mount_t *mp)
860{
861 xfs_sb_t *sbp = &(mp->m_sb);
862 int readio_log, writeio_log;
863
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
865 if (mp->m_flags & XFS_MOUNT_WSYNC) {
866 readio_log = XFS_WSYNC_READIO_LOG;
867 writeio_log = XFS_WSYNC_WRITEIO_LOG;
868 } else {
869 readio_log = XFS_READIO_LOG_LARGE;
870 writeio_log = XFS_WRITEIO_LOG_LARGE;
871 }
872 } else {
873 readio_log = mp->m_readio_log;
874 writeio_log = mp->m_writeio_log;
875 }
876
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 if (sbp->sb_blocklog > readio_log) {
878 mp->m_readio_log = sbp->sb_blocklog;
879 } else {
880 mp->m_readio_log = readio_log;
881 }
882 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
883 if (sbp->sb_blocklog > writeio_log) {
884 mp->m_writeio_log = sbp->sb_blocklog;
885 } else {
886 mp->m_writeio_log = writeio_log;
887 }
888 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000889}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890
Eric Sandeen0771fb42007-10-12 11:03:40 +1000891/*
892 * Set whether we're using inode alignment.
893 */
894STATIC void
895xfs_set_inoalignment(xfs_mount_t *mp)
896{
Eric Sandeen62118702008-03-06 13:44:28 +1100897 if (xfs_sb_version_hasalign(&mp->m_sb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 mp->m_sb.sb_inoalignmt >=
899 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
900 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
901 else
902 mp->m_inoalign_mask = 0;
903 /*
904 * If we are using stripe alignment, check whether
905 * the stripe unit is a multiple of the inode alignment
906 */
907 if (mp->m_dalign && mp->m_inoalign_mask &&
908 !(mp->m_dalign & mp->m_inoalign_mask))
909 mp->m_sinoalign = mp->m_dalign;
910 else
911 mp->m_sinoalign = 0;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000912}
913
914/*
915 * Check that the data (and log if separate) are an ok size.
916 */
917STATIC int
Christoph Hellwig42490232008-08-13 16:49:32 +1000918xfs_check_sizes(xfs_mount_t *mp)
Eric Sandeen0771fb42007-10-12 11:03:40 +1000919{
920 xfs_buf_t *bp;
921 xfs_daddr_t d;
922 int error;
923
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
925 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
926 cmn_err(CE_WARN, "XFS: size check 1 failed");
Eric Sandeen0771fb42007-10-12 11:03:40 +1000927 return XFS_ERROR(E2BIG);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 }
929 error = xfs_read_buf(mp, mp->m_ddev_targp,
930 d - XFS_FSS_TO_BB(mp, 1),
931 XFS_FSS_TO_BB(mp, 1), 0, &bp);
932 if (!error) {
933 xfs_buf_relse(bp);
934 } else {
935 cmn_err(CE_WARN, "XFS: size check 2 failed");
Eric Sandeen0771fb42007-10-12 11:03:40 +1000936 if (error == ENOSPC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 error = XFS_ERROR(E2BIG);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000938 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 }
940
Christoph Hellwig42490232008-08-13 16:49:32 +1000941 if (mp->m_logdev_targp != mp->m_ddev_targp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
943 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
944 cmn_err(CE_WARN, "XFS: size check 3 failed");
Eric Sandeen0771fb42007-10-12 11:03:40 +1000945 return XFS_ERROR(E2BIG);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 }
947 error = xfs_read_buf(mp, mp->m_logdev_targp,
948 d - XFS_FSB_TO_BB(mp, 1),
949 XFS_FSB_TO_BB(mp, 1), 0, &bp);
950 if (!error) {
951 xfs_buf_relse(bp);
952 } else {
953 cmn_err(CE_WARN, "XFS: size check 3 failed");
Eric Sandeen0771fb42007-10-12 11:03:40 +1000954 if (error == ENOSPC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 error = XFS_ERROR(E2BIG);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000956 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 }
958 }
Eric Sandeen0771fb42007-10-12 11:03:40 +1000959 return 0;
960}
961
962/*
Eric Sandeen0771fb42007-10-12 11:03:40 +1000963 * This function does the following on an initial mount of a file system:
964 * - reads the superblock from disk and init the mount struct
965 * - if we're a 32-bit kernel, do a size check on the superblock
966 * so we don't mount terabyte filesystems
967 * - init mount struct realtime fields
968 * - allocate inode hash table for fs
969 * - init directory manager
970 * - perform recovery and init the log manager
971 */
972int
973xfs_mountfs(
Christoph Hellwig42490232008-08-13 16:49:32 +1000974 xfs_mount_t *mp)
Eric Sandeen0771fb42007-10-12 11:03:40 +1000975{
976 xfs_sb_t *sbp = &(mp->m_sb);
977 xfs_inode_t *rip;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000978 __uint64_t resblks;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000979 uint quotamount, quotaflags;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000980 int error = 0;
981
Eric Sandeen0771fb42007-10-12 11:03:40 +1000982 xfs_mount_common(mp, sbp);
983
984 /*
Eric Sandeene6957ea2008-04-10 12:19:34 +1000985 * Check for a mismatched features2 values. Older kernels
986 * read & wrote into the wrong sb offset for sb_features2
987 * on some platforms due to xfs_sb_t not being 64bit size aligned
988 * when sb_features2 was added, which made older superblock
989 * reading/writing routines swap it as a 64-bit value.
David Chinneree1c0902008-03-06 13:45:50 +1100990 *
Eric Sandeene6957ea2008-04-10 12:19:34 +1000991 * For backwards compatibility, we make both slots equal.
992 *
993 * If we detect a mismatched field, we OR the set bits into the
994 * existing features2 field in case it has already been modified; we
995 * don't want to lose any features. We then update the bad location
996 * with the ORed value so that older kernels will see any features2
997 * flags, and mark the two fields as needing updates once the
998 * transaction subsystem is online.
David Chinneree1c0902008-03-06 13:45:50 +1100999 */
Eric Sandeene6957ea2008-04-10 12:19:34 +10001000 if (xfs_sb_has_mismatched_features2(sbp)) {
David Chinneree1c0902008-03-06 13:45:50 +11001001 cmn_err(CE_WARN,
1002 "XFS: correcting sb_features alignment problem");
1003 sbp->sb_features2 |= sbp->sb_bad_features2;
Eric Sandeene6957ea2008-04-10 12:19:34 +10001004 sbp->sb_bad_features2 = sbp->sb_features2;
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001005 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
Eric Sandeene6957ea2008-04-10 12:19:34 +10001006
1007 /*
1008 * Re-check for ATTR2 in case it was found in bad_features2
1009 * slot.
1010 */
Tim Shimmin7c12f292008-04-30 18:15:28 +10001011 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1012 !(mp->m_flags & XFS_MOUNT_NOATTR2))
Eric Sandeene6957ea2008-04-10 12:19:34 +10001013 mp->m_flags |= XFS_MOUNT_ATTR2;
Tim Shimmin7c12f292008-04-30 18:15:28 +10001014 }
Eric Sandeene6957ea2008-04-10 12:19:34 +10001015
Tim Shimmin7c12f292008-04-30 18:15:28 +10001016 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1017 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1018 xfs_sb_version_removeattr2(&mp->m_sb);
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001019 mp->m_update_flags |= XFS_SB_FEATURES2;
Tim Shimmin7c12f292008-04-30 18:15:28 +10001020
1021 /* update sb_versionnum for the clearing of the morebits */
1022 if (!sbp->sb_features2)
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001023 mp->m_update_flags |= XFS_SB_VERSIONNUM;
David Chinneree1c0902008-03-06 13:45:50 +11001024 }
1025
1026 /*
Eric Sandeen0771fb42007-10-12 11:03:40 +10001027 * Check if sb_agblocks is aligned at stripe boundary
1028 * If sb_agblocks is NOT aligned turn off m_dalign since
1029 * allocator alignment is within an ag, therefore ag has
1030 * to be aligned at stripe boundary.
1031 */
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001032 error = xfs_update_alignment(mp);
Eric Sandeen0771fb42007-10-12 11:03:40 +10001033 if (error)
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001034 goto out;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001035
1036 xfs_alloc_compute_maxlevels(mp);
1037 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
1038 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
1039 xfs_ialloc_compute_maxlevels(mp);
1040
1041 xfs_set_maxicount(mp);
1042
1043 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
1044
Christoph Hellwig27174202009-03-30 10:21:31 +02001045 error = xfs_uuid_mount(mp);
1046 if (error)
1047 goto out;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001048
1049 /*
1050 * Set the minimum read and write sizes
1051 */
1052 xfs_set_rw_sizes(mp);
1053
1054 /*
1055 * Set the inode cluster size.
1056 * This may still be overridden by the file system
1057 * block size if it is larger than the chosen cluster size.
1058 */
1059 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1060
1061 /*
1062 * Set inode alignment fields
1063 */
1064 xfs_set_inoalignment(mp);
1065
1066 /*
1067 * Check that the data (and log if separate) are an ok size.
1068 */
Christoph Hellwig42490232008-08-13 16:49:32 +10001069 error = xfs_check_sizes(mp);
Eric Sandeen0771fb42007-10-12 11:03:40 +10001070 if (error)
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001071 goto out_remove_uuid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072
1073 /*
1074 * Initialize realtime fields in the mount structure
1075 */
Eric Sandeen0771fb42007-10-12 11:03:40 +10001076 error = xfs_rtmount_init(mp);
1077 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 cmn_err(CE_WARN, "XFS: RT mount failed");
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001079 goto out_remove_uuid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 }
1081
1082 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 * Copies the low order bits of the timestamp and the randomly
1084 * set "sequence" number out of a UUID.
1085 */
1086 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1087
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 mp->m_dmevmask = 0; /* not persistent; set after each mount */
1089
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001090 xfs_dir_mount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091
1092 /*
1093 * Initialize the attribute manager's entries.
1094 */
1095 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1096
1097 /*
1098 * Initialize the precomputed transaction reservations values.
1099 */
1100 xfs_trans_init(mp);
1101
1102 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 * Allocate and initialize the per-ag data.
1104 */
1105 init_rwsem(&mp->m_peraglock);
Christoph Hellwigd62c2512008-08-13 16:51:29 +10001106 mp->m_perag = kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t),
1107 KM_MAYFAIL);
1108 if (!mp->m_perag)
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001109 goto out_remove_uuid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110
Christoph Hellwigb267ce92007-08-30 17:21:30 +10001111 mp->m_maxagi = xfs_initialize_perag(mp, sbp->sb_agcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001113 if (!sbp->sb_logblocks) {
1114 cmn_err(CE_WARN, "XFS: no log defined");
1115 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
1116 error = XFS_ERROR(EFSCORRUPTED);
1117 goto out_free_perag;
1118 }
1119
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 /*
1121 * log's mount-time initialization. Perform 1st part recovery if needed
1122 */
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001123 error = xfs_log_mount(mp, mp->m_logdev_targp,
1124 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1125 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1126 if (error) {
1127 cmn_err(CE_WARN, "XFS: log mount failed");
1128 goto out_free_perag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 }
1130
1131 /*
David Chinner92821e22007-05-24 15:26:31 +10001132 * Now the log is mounted, we know if it was an unclean shutdown or
1133 * not. If it was, with the first phase of recovery has completed, we
1134 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1135 * but they are recovered transactionally in the second recovery phase
1136 * later.
1137 *
1138 * Hence we can safely re-initialise incore superblock counters from
1139 * the per-ag data. These may not be correct if the filesystem was not
1140 * cleanly unmounted, so we need to wait for recovery to finish before
1141 * doing this.
1142 *
1143 * If the filesystem was cleanly unmounted, then we can trust the
1144 * values in the superblock to be correct and we don't need to do
1145 * anything here.
1146 *
1147 * If we are currently making the filesystem, the initialisation will
1148 * fail as the perag data is in an undefined state.
1149 */
David Chinner92821e22007-05-24 15:26:31 +10001150 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1151 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1152 !mp->m_sb.sb_inprogress) {
1153 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001154 if (error)
1155 goto out_free_perag;
David Chinner92821e22007-05-24 15:26:31 +10001156 }
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001157
David Chinner92821e22007-05-24 15:26:31 +10001158 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 * Get and sanity-check the root inode.
1160 * Save the pointer to it in the mount structure.
1161 */
1162 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0);
1163 if (error) {
1164 cmn_err(CE_WARN, "XFS: failed to read root inode");
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001165 goto out_log_dealloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 }
1167
1168 ASSERT(rip != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
1170 if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
1171 cmn_err(CE_WARN, "XFS: corrupted root inode");
Nathan Scottb6574522006-06-09 15:29:40 +10001172 cmn_err(CE_WARN, "Device %s - root %llu is not a directory",
1173 XFS_BUFTARG_NAME(mp->m_ddev_targp),
1174 (unsigned long long)rip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1176 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1177 mp);
1178 error = XFS_ERROR(EFSCORRUPTED);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001179 goto out_rele_rip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 }
1181 mp->m_rootip = rip; /* save it */
1182
1183 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1184
1185 /*
1186 * Initialize realtime inode pointers in the mount structure
1187 */
Eric Sandeen0771fb42007-10-12 11:03:40 +10001188 error = xfs_rtmount_inodes(mp);
1189 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 /*
1191 * Free up the root inode.
1192 */
1193 cmn_err(CE_WARN, "XFS: failed to read RT inodes");
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001194 goto out_rele_rip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 }
1196
1197 /*
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001198 * If this is a read-only mount defer the superblock updates until
1199 * the next remount into writeable mode. Otherwise we would never
1200 * perform the update e.g. for the root filesystem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 */
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001202 if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1203 error = xfs_mount_log_sb(mp, mp->m_update_flags);
David Chinnere5720ee2008-04-10 12:21:18 +10001204 if (error) {
1205 cmn_err(CE_WARN, "XFS: failed to write sb changes");
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001206 goto out_rtunmount;
David Chinnere5720ee2008-04-10 12:21:18 +10001207 }
1208 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209
1210 /*
1211 * Initialise the XFS quota management subsystem for this mount
1212 */
Eric Sandeen0771fb42007-10-12 11:03:40 +10001213 error = XFS_QM_INIT(mp, &quotamount, &quotaflags);
1214 if (error)
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001215 goto out_rtunmount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216
1217 /*
1218 * Finish recovering the file system. This part needed to be
1219 * delayed until after the root and real-time bitmap inodes
1220 * were consistently read in.
1221 */
Christoph Hellwig42490232008-08-13 16:49:32 +10001222 error = xfs_log_mount_finish(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 if (error) {
1224 cmn_err(CE_WARN, "XFS: log mount finish failed");
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001225 goto out_rtunmount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 }
1227
1228 /*
1229 * Complete the quota initialisation, post-log-replay component.
1230 */
Christoph Hellwig42490232008-08-13 16:49:32 +10001231 error = XFS_QM_MOUNT(mp, quotamount, quotaflags);
Eric Sandeen0771fb42007-10-12 11:03:40 +10001232 if (error)
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001233 goto out_rtunmount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234
David Chinner84e1e992007-06-18 16:50:27 +10001235 /*
1236 * Now we are mounted, reserve a small amount of unused space for
1237 * privileged transactions. This is needed so that transaction
1238 * space required for critical operations can dip into this pool
1239 * when at ENOSPC. This is needed for operations like create with
1240 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1241 * are not allowed to use this reserved space.
1242 *
1243 * We default to 5% or 1024 fsbs of space reserved, whichever is smaller.
1244 * This may drive us straight to ENOSPC on mount, but that implies
David Chinner714082b2008-04-10 12:20:03 +10001245 * we were already there on the last unmount. Warn if this occurs.
David Chinner84e1e992007-06-18 16:50:27 +10001246 */
Christoph Hellwig39726be2007-06-18 17:57:45 +10001247 resblks = mp->m_sb.sb_dblocks;
1248 do_div(resblks, 20);
1249 resblks = min_t(__uint64_t, resblks, 1024);
David Chinner714082b2008-04-10 12:20:03 +10001250 error = xfs_reserve_blocks(mp, &resblks, NULL);
1251 if (error)
1252 cmn_err(CE_WARN, "XFS: Unable to allocate reserve blocks. "
1253 "Continuing without a reserve pool.");
David Chinner84e1e992007-06-18 16:50:27 +10001254
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 return 0;
1256
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001257 out_rtunmount:
1258 xfs_rtunmount_inodes(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001259 out_rele_rip:
Christoph Hellwig43355092008-03-27 18:01:08 +11001260 IRELE(rip);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001261 out_log_dealloc:
Christoph Hellwig21b699c2009-03-16 08:19:29 +01001262 xfs_log_unmount(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001263 out_free_perag:
Christoph Hellwigff4f0382008-08-13 16:50:47 +10001264 xfs_free_perag(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001265 out_remove_uuid:
Christoph Hellwig27174202009-03-30 10:21:31 +02001266 xfs_uuid_unmount(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001267 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 return error;
1269}
1270
1271/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 * This flushes out the inodes,dquots and the superblock, unmounts the
1273 * log and makes sure that incore structures are freed.
1274 */
Christoph Hellwig41b5c2e2008-08-13 16:49:57 +10001275void
1276xfs_unmountfs(
1277 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278{
Christoph Hellwig41b5c2e2008-08-13 16:49:57 +10001279 __uint64_t resblks;
1280 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281
Christoph Hellwigf95099b2008-12-03 12:20:37 +01001282 /*
1283 * Release dquot that rootinode, rbmino and rsumino might be holding,
1284 * and release the quota inodes.
1285 */
1286 XFS_QM_UNMOUNT(mp);
1287
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001288 xfs_rtunmount_inodes(mp);
Christoph Hellwig77508ec2008-08-13 16:49:04 +10001289 IRELE(mp->m_rootip);
1290
David Chinner641c56f2007-06-18 16:50:17 +10001291 /*
1292 * We can potentially deadlock here if we have an inode cluster
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001293 * that has been freed has its buffer still pinned in memory because
David Chinner641c56f2007-06-18 16:50:17 +10001294 * the transaction is still sitting in a iclog. The stale inodes
1295 * on that buffer will have their flush locks held until the
1296 * transaction hits the disk and the callbacks run. the inode
1297 * flush takes the flush lock unconditionally and with nothing to
1298 * push out the iclog we will never get that unlocked. hence we
1299 * need to force the log first.
1300 */
1301 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
David Chinner1dc33182008-10-30 17:37:15 +11001302 xfs_reclaim_inodes(mp, 0, XFS_IFLUSH_ASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
Nathan Scottee2a4f72006-01-11 15:33:36 +11001304 XFS_QM_DQPURGEALL(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_UMOUNTING);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305
Lachlan McIlroya357a122008-10-30 16:53:25 +11001306 if (mp->m_quotainfo)
1307 XFS_QM_DONE(mp);
1308
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 /*
1310 * Flush out the log synchronously so that we know for sure
1311 * that nothing is pinned. This is important because bflush()
1312 * will skip pinned buffers.
1313 */
1314 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1315
1316 xfs_binval(mp->m_ddev_targp);
1317 if (mp->m_rtdev_targp) {
1318 xfs_binval(mp->m_rtdev_targp);
1319 }
1320
David Chinner84e1e992007-06-18 16:50:27 +10001321 /*
1322 * Unreserve any blocks we have so that when we unmount we don't account
1323 * the reserved free space as used. This is really only necessary for
1324 * lazy superblock counting because it trusts the incore superblock
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001325 * counters to be absolutely correct on clean unmount.
David Chinner84e1e992007-06-18 16:50:27 +10001326 *
1327 * We don't bother correcting this elsewhere for lazy superblock
1328 * counting because on mount of an unclean filesystem we reconstruct the
1329 * correct counter value and this is irrelevant.
1330 *
1331 * For non-lazy counter filesystems, this doesn't matter at all because
1332 * we only every apply deltas to the superblock and hence the incore
1333 * value does not matter....
1334 */
1335 resblks = 0;
David Chinner714082b2008-04-10 12:20:03 +10001336 error = xfs_reserve_blocks(mp, &resblks, NULL);
1337 if (error)
1338 cmn_err(CE_WARN, "XFS: Unable to free reserved block pool. "
1339 "Freespace may not be correct on next mount.");
1340
David Chinnere5720ee2008-04-10 12:21:18 +10001341 error = xfs_log_sbcount(mp, 1);
1342 if (error)
1343 cmn_err(CE_WARN, "XFS: Unable to update superblock counters. "
1344 "Freespace may not be correct on next mount.");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 xfs_unmountfs_writesb(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 xfs_unmountfs_wait(mp); /* wait for async bufs */
Christoph Hellwig21b699c2009-03-16 08:19:29 +01001347 xfs_log_unmount_write(mp);
1348 xfs_log_unmount(mp);
Christoph Hellwig27174202009-03-30 10:21:31 +02001349 xfs_uuid_unmount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
Christoph Hellwig1550d0b2008-08-13 16:17:37 +10001351#if defined(DEBUG)
Christoph Hellwig0ce4cfd2007-08-30 17:20:53 +10001352 xfs_errortag_clearall(mp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353#endif
Christoph Hellwigff4f0382008-08-13 16:50:47 +10001354 xfs_free_perag(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355}
1356
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10001357STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358xfs_unmountfs_wait(xfs_mount_t *mp)
1359{
1360 if (mp->m_logdev_targp != mp->m_ddev_targp)
1361 xfs_wait_buftarg(mp->m_logdev_targp);
1362 if (mp->m_rtdev_targp)
1363 xfs_wait_buftarg(mp->m_rtdev_targp);
1364 xfs_wait_buftarg(mp->m_ddev_targp);
1365}
1366
1367int
David Chinner92821e22007-05-24 15:26:31 +10001368xfs_fs_writable(xfs_mount_t *mp)
1369{
Christoph Hellwigb267ce92007-08-30 17:21:30 +10001370 return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001371 (mp->m_flags & XFS_MOUNT_RDONLY));
David Chinner92821e22007-05-24 15:26:31 +10001372}
1373
1374/*
1375 * xfs_log_sbcount
1376 *
1377 * Called either periodically to keep the on disk superblock values
1378 * roughly up to date or from unmount to make sure the values are
1379 * correct on a clean unmount.
1380 *
1381 * Note this code can be called during the process of freezing, so
1382 * we may need to use the transaction allocator which does not not
1383 * block when the transaction subsystem is in its frozen state.
1384 */
1385int
1386xfs_log_sbcount(
1387 xfs_mount_t *mp,
1388 uint sync)
1389{
1390 xfs_trans_t *tp;
1391 int error;
1392
1393 if (!xfs_fs_writable(mp))
1394 return 0;
1395
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10001396 xfs_icsb_sync_counters(mp, 0);
David Chinner92821e22007-05-24 15:26:31 +10001397
1398 /*
1399 * we don't need to do this if we are updating the superblock
1400 * counters on every modification.
1401 */
1402 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1403 return 0;
1404
1405 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT);
1406 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1407 XFS_DEFAULT_LOG_COUNT);
1408 if (error) {
1409 xfs_trans_cancel(tp, 0);
1410 return error;
1411 }
1412
1413 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1414 if (sync)
1415 xfs_trans_set_sync(tp);
David Chinnere5720ee2008-04-10 12:21:18 +10001416 error = xfs_trans_commit(tp, 0);
1417 return error;
David Chinner92821e22007-05-24 15:26:31 +10001418}
1419
1420int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421xfs_unmountfs_writesb(xfs_mount_t *mp)
1422{
1423 xfs_buf_t *sbp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 int error = 0;
1425
1426 /*
1427 * skip superblock write if fs is read-only, or
1428 * if we are doing a forced umount.
1429 */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001430 if (!((mp->m_flags & XFS_MOUNT_RDONLY) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 XFS_FORCED_SHUTDOWN(mp))) {
David Chinner8d280b92006-03-14 13:13:09 +11001432
David Chinner92821e22007-05-24 15:26:31 +10001433 sbp = xfs_getsb(mp, 0);
David Chinner8d280b92006-03-14 13:13:09 +11001434
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 XFS_BUF_UNDONE(sbp);
1436 XFS_BUF_UNREAD(sbp);
1437 XFS_BUF_UNDELAYWRITE(sbp);
1438 XFS_BUF_WRITE(sbp);
1439 XFS_BUF_UNASYNC(sbp);
1440 ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1441 xfsbdstrat(mp, sbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 error = xfs_iowait(sbp);
1443 if (error)
1444 xfs_ioerror_alert("xfs_unmountfs_writesb",
1445 mp, sbp, XFS_BUF_ADDR(sbp));
David Chinner92821e22007-05-24 15:26:31 +10001446 xfs_buf_relse(sbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001448 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449}
1450
1451/*
1452 * xfs_mod_sb() can be used to copy arbitrary changes to the
1453 * in-core superblock into the superblock buffer to be logged.
1454 * It does not provide the higher level of locking that is
1455 * needed to protect the in-core superblock from concurrent
1456 * access.
1457 */
1458void
1459xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1460{
1461 xfs_buf_t *bp;
1462 int first;
1463 int last;
1464 xfs_mount_t *mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 xfs_sb_field_t f;
1466
1467 ASSERT(fields);
1468 if (!fields)
1469 return;
1470 mp = tp->t_mountp;
1471 bp = xfs_trans_getsb(tp, mp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 first = sizeof(xfs_sb_t);
1473 last = 0;
1474
1475 /* translate/copy */
1476
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +10001477 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478
1479 /* find modified range */
1480
1481 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1482 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1483 first = xfs_sb_info[f].offset;
1484
1485 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1486 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1487 last = xfs_sb_info[f + 1].offset - 1;
1488
1489 xfs_trans_log_buf(tp, bp, first, last);
1490}
Yingping Lud210a282006-06-09 14:55:18 +10001491
Yingping Lud210a282006-06-09 14:55:18 +10001492
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493/*
1494 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1495 * a delta to a specified field in the in-core superblock. Simply
1496 * switch on the field indicated and apply the delta to that field.
1497 * Fields are not allowed to dip below zero, so if the delta would
1498 * do this do not apply it and return EINVAL.
1499 *
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001500 * The m_sb_lock must be held when this routine is called.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 */
David Chinner8d280b92006-03-14 13:13:09 +11001502int
David Chinner20f4ebf2007-02-10 18:36:10 +11001503xfs_mod_incore_sb_unlocked(
1504 xfs_mount_t *mp,
1505 xfs_sb_field_t field,
1506 int64_t delta,
1507 int rsvd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508{
1509 int scounter; /* short counter for 32 bit fields */
1510 long long lcounter; /* long counter for 64 bit fields */
1511 long long res_used, rem;
1512
1513 /*
1514 * With the in-core superblock spin lock held, switch
1515 * on the indicated field. Apply the delta to the
1516 * proper field. If the fields value would dip below
1517 * 0, then do not apply the delta and return EINVAL.
1518 */
1519 switch (field) {
1520 case XFS_SBS_ICOUNT:
1521 lcounter = (long long)mp->m_sb.sb_icount;
1522 lcounter += delta;
1523 if (lcounter < 0) {
1524 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001525 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 }
1527 mp->m_sb.sb_icount = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001528 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 case XFS_SBS_IFREE:
1530 lcounter = (long long)mp->m_sb.sb_ifree;
1531 lcounter += delta;
1532 if (lcounter < 0) {
1533 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001534 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 }
1536 mp->m_sb.sb_ifree = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001537 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 case XFS_SBS_FDBLOCKS:
David Chinner4be536d2006-09-07 14:26:50 +10001539 lcounter = (long long)
1540 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1542
1543 if (delta > 0) { /* Putting blocks back */
1544 if (res_used > delta) {
1545 mp->m_resblks_avail += delta;
1546 } else {
1547 rem = delta - res_used;
1548 mp->m_resblks_avail = mp->m_resblks;
1549 lcounter += rem;
1550 }
1551 } else { /* Taking blocks away */
1552
1553 lcounter += delta;
1554
1555 /*
1556 * If were out of blocks, use any available reserved blocks if
1557 * were allowed to.
1558 */
1559
1560 if (lcounter < 0) {
1561 if (rsvd) {
1562 lcounter = (long long)mp->m_resblks_avail + delta;
1563 if (lcounter < 0) {
Jesper Juhl014c2542006-01-15 02:37:08 +01001564 return XFS_ERROR(ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 }
1566 mp->m_resblks_avail = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001567 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 } else { /* not reserved */
Jesper Juhl014c2542006-01-15 02:37:08 +01001569 return XFS_ERROR(ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 }
1571 }
1572 }
1573
David Chinner4be536d2006-09-07 14:26:50 +10001574 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
Jesper Juhl014c2542006-01-15 02:37:08 +01001575 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 case XFS_SBS_FREXTENTS:
1577 lcounter = (long long)mp->m_sb.sb_frextents;
1578 lcounter += delta;
1579 if (lcounter < 0) {
Jesper Juhl014c2542006-01-15 02:37:08 +01001580 return XFS_ERROR(ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 }
1582 mp->m_sb.sb_frextents = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001583 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 case XFS_SBS_DBLOCKS:
1585 lcounter = (long long)mp->m_sb.sb_dblocks;
1586 lcounter += delta;
1587 if (lcounter < 0) {
1588 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001589 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 }
1591 mp->m_sb.sb_dblocks = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001592 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 case XFS_SBS_AGCOUNT:
1594 scounter = mp->m_sb.sb_agcount;
1595 scounter += delta;
1596 if (scounter < 0) {
1597 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001598 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 }
1600 mp->m_sb.sb_agcount = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001601 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 case XFS_SBS_IMAX_PCT:
1603 scounter = mp->m_sb.sb_imax_pct;
1604 scounter += delta;
1605 if (scounter < 0) {
1606 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001607 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 }
1609 mp->m_sb.sb_imax_pct = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001610 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 case XFS_SBS_REXTSIZE:
1612 scounter = mp->m_sb.sb_rextsize;
1613 scounter += delta;
1614 if (scounter < 0) {
1615 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001616 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 }
1618 mp->m_sb.sb_rextsize = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001619 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 case XFS_SBS_RBMBLOCKS:
1621 scounter = mp->m_sb.sb_rbmblocks;
1622 scounter += delta;
1623 if (scounter < 0) {
1624 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001625 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 }
1627 mp->m_sb.sb_rbmblocks = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001628 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 case XFS_SBS_RBLOCKS:
1630 lcounter = (long long)mp->m_sb.sb_rblocks;
1631 lcounter += delta;
1632 if (lcounter < 0) {
1633 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001634 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 }
1636 mp->m_sb.sb_rblocks = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001637 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 case XFS_SBS_REXTENTS:
1639 lcounter = (long long)mp->m_sb.sb_rextents;
1640 lcounter += delta;
1641 if (lcounter < 0) {
1642 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001643 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 }
1645 mp->m_sb.sb_rextents = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001646 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 case XFS_SBS_REXTSLOG:
1648 scounter = mp->m_sb.sb_rextslog;
1649 scounter += delta;
1650 if (scounter < 0) {
1651 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001652 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 }
1654 mp->m_sb.sb_rextslog = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001655 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 default:
1657 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001658 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 }
1660}
1661
1662/*
1663 * xfs_mod_incore_sb() is used to change a field in the in-core
1664 * superblock structure by the specified delta. This modification
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001665 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 * routine to do the work.
1667 */
1668int
David Chinner20f4ebf2007-02-10 18:36:10 +11001669xfs_mod_incore_sb(
1670 xfs_mount_t *mp,
1671 xfs_sb_field_t field,
1672 int64_t delta,
1673 int rsvd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 int status;
1676
David Chinner8d280b92006-03-14 13:13:09 +11001677 /* check for per-cpu counters */
1678 switch (field) {
1679#ifdef HAVE_PERCPU_SB
1680 case XFS_SBS_ICOUNT:
1681 case XFS_SBS_IFREE:
1682 case XFS_SBS_FDBLOCKS:
1683 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1684 status = xfs_icsb_modify_counters(mp, field,
1685 delta, rsvd);
1686 break;
1687 }
1688 /* FALLTHROUGH */
1689#endif
1690 default:
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001691 spin_lock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001692 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001693 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001694 break;
1695 }
1696
Jesper Juhl014c2542006-01-15 02:37:08 +01001697 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698}
1699
1700/*
1701 * xfs_mod_incore_sb_batch() is used to change more than one field
1702 * in the in-core superblock structure at a time. This modification
1703 * is protected by a lock internal to this module. The fields and
1704 * changes to those fields are specified in the array of xfs_mod_sb
1705 * structures passed in.
1706 *
1707 * Either all of the specified deltas will be applied or none of
1708 * them will. If any modified field dips below 0, then all modifications
1709 * will be backed out and EINVAL will be returned.
1710 */
1711int
1712xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1713{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 int status=0;
1715 xfs_mod_sb_t *msbp;
1716
1717 /*
1718 * Loop through the array of mod structures and apply each
1719 * individually. If any fail, then back out all those
1720 * which have already been applied. Do all of this within
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001721 * the scope of the m_sb_lock so that all of the changes will
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 * be atomic.
1723 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001724 spin_lock(&mp->m_sb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 msbp = &msb[0];
1726 for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1727 /*
1728 * Apply the delta at index n. If it fails, break
1729 * from the loop so we'll fall into the undo loop
1730 * below.
1731 */
David Chinner8d280b92006-03-14 13:13:09 +11001732 switch (msbp->msb_field) {
1733#ifdef HAVE_PERCPU_SB
1734 case XFS_SBS_ICOUNT:
1735 case XFS_SBS_IFREE:
1736 case XFS_SBS_FDBLOCKS:
1737 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001738 spin_unlock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11001739 status = xfs_icsb_modify_counters(mp,
David Chinner8d280b92006-03-14 13:13:09 +11001740 msbp->msb_field,
1741 msbp->msb_delta, rsvd);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001742 spin_lock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001743 break;
1744 }
1745 /* FALLTHROUGH */
1746#endif
1747 default:
1748 status = xfs_mod_incore_sb_unlocked(mp,
1749 msbp->msb_field,
1750 msbp->msb_delta, rsvd);
1751 break;
1752 }
1753
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 if (status != 0) {
1755 break;
1756 }
1757 }
1758
1759 /*
1760 * If we didn't complete the loop above, then back out
1761 * any changes made to the superblock. If you add code
1762 * between the loop above and here, make sure that you
1763 * preserve the value of status. Loop back until
1764 * we step below the beginning of the array. Make sure
1765 * we don't touch anything back there.
1766 */
1767 if (status != 0) {
1768 msbp--;
1769 while (msbp >= msb) {
David Chinner8d280b92006-03-14 13:13:09 +11001770 switch (msbp->msb_field) {
1771#ifdef HAVE_PERCPU_SB
1772 case XFS_SBS_ICOUNT:
1773 case XFS_SBS_IFREE:
1774 case XFS_SBS_FDBLOCKS:
1775 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001776 spin_unlock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11001777 status = xfs_icsb_modify_counters(mp,
David Chinner8d280b92006-03-14 13:13:09 +11001778 msbp->msb_field,
1779 -(msbp->msb_delta),
1780 rsvd);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001781 spin_lock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001782 break;
1783 }
1784 /* FALLTHROUGH */
1785#endif
1786 default:
1787 status = xfs_mod_incore_sb_unlocked(mp,
1788 msbp->msb_field,
1789 -(msbp->msb_delta),
1790 rsvd);
1791 break;
1792 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 ASSERT(status == 0);
1794 msbp--;
1795 }
1796 }
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001797 spin_unlock(&mp->m_sb_lock);
Jesper Juhl014c2542006-01-15 02:37:08 +01001798 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799}
1800
1801/*
1802 * xfs_getsb() is called to obtain the buffer for the superblock.
1803 * The buffer is returned locked and read in from disk.
1804 * The buffer should be released with a call to xfs_brelse().
1805 *
1806 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1807 * the superblock buffer if it can be locked without sleeping.
1808 * If it can't then we'll return NULL.
1809 */
1810xfs_buf_t *
1811xfs_getsb(
1812 xfs_mount_t *mp,
1813 int flags)
1814{
1815 xfs_buf_t *bp;
1816
1817 ASSERT(mp->m_sb_bp != NULL);
1818 bp = mp->m_sb_bp;
1819 if (flags & XFS_BUF_TRYLOCK) {
1820 if (!XFS_BUF_CPSEMA(bp)) {
1821 return NULL;
1822 }
1823 } else {
1824 XFS_BUF_PSEMA(bp, PRIBIO);
1825 }
1826 XFS_BUF_HOLD(bp);
1827 ASSERT(XFS_BUF_ISDONE(bp));
Jesper Juhl014c2542006-01-15 02:37:08 +01001828 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829}
1830
1831/*
1832 * Used to free the superblock along various error paths.
1833 */
1834void
1835xfs_freesb(
1836 xfs_mount_t *mp)
1837{
1838 xfs_buf_t *bp;
1839
1840 /*
1841 * Use xfs_getsb() so that the buffer will be locked
1842 * when we call xfs_buf_relse().
1843 */
1844 bp = xfs_getsb(mp, 0);
1845 XFS_BUF_UNMANAGE(bp);
1846 xfs_buf_relse(bp);
1847 mp->m_sb_bp = NULL;
1848}
1849
1850/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 * Used to log changes to the superblock unit and width fields which could
Eric Sandeene6957ea2008-04-10 12:19:34 +10001852 * be altered by the mount options, as well as any potential sb_features2
1853 * fixup. Only the first superblock is updated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 */
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001855int
David Chinneree1c0902008-03-06 13:45:50 +11001856xfs_mount_log_sb(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 xfs_mount_t *mp,
1858 __int64_t fields)
1859{
1860 xfs_trans_t *tp;
David Chinnere5720ee2008-04-10 12:21:18 +10001861 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862
David Chinneree1c0902008-03-06 13:45:50 +11001863 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
David Chinner4b166de2008-05-20 11:30:27 +10001864 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1865 XFS_SB_VERSIONNUM));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866
1867 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
David Chinnere5720ee2008-04-10 12:21:18 +10001868 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1869 XFS_DEFAULT_LOG_COUNT);
1870 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 xfs_trans_cancel(tp, 0);
David Chinnere5720ee2008-04-10 12:21:18 +10001872 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 }
1874 xfs_mod_sb(tp, fields);
David Chinnere5720ee2008-04-10 12:21:18 +10001875 error = xfs_trans_commit(tp, 0);
1876 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877}
David Chinner8d280b92006-03-14 13:13:09 +11001878
1879
1880#ifdef HAVE_PERCPU_SB
1881/*
1882 * Per-cpu incore superblock counters
1883 *
1884 * Simple concept, difficult implementation
1885 *
1886 * Basically, replace the incore superblock counters with a distributed per cpu
1887 * counter for contended fields (e.g. free block count).
1888 *
1889 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1890 * hence needs to be accurately read when we are running low on space. Hence
1891 * there is a method to enable and disable the per-cpu counters based on how
1892 * much "stuff" is available in them.
1893 *
1894 * Basically, a counter is enabled if there is enough free resource to justify
1895 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1896 * ENOSPC), then we disable the counters to synchronise all callers and
1897 * re-distribute the available resources.
1898 *
1899 * If, once we redistributed the available resources, we still get a failure,
1900 * we disable the per-cpu counter and go through the slow path.
1901 *
1902 * The slow path is the current xfs_mod_incore_sb() function. This means that
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001903 * when we disable a per-cpu counter, we need to drain its resources back to
David Chinner8d280b92006-03-14 13:13:09 +11001904 * the global superblock. We do this after disabling the counter to prevent
1905 * more threads from queueing up on the counter.
1906 *
1907 * Essentially, this means that we still need a lock in the fast path to enable
1908 * synchronisation between the global counters and the per-cpu counters. This
1909 * is not a problem because the lock will be local to a CPU almost all the time
1910 * and have little contention except when we get to ENOSPC conditions.
1911 *
1912 * Basically, this lock becomes a barrier that enables us to lock out the fast
1913 * path while we do things like enabling and disabling counters and
1914 * synchronising the counters.
1915 *
1916 * Locking rules:
1917 *
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001918 * 1. m_sb_lock before picking up per-cpu locks
David Chinner8d280b92006-03-14 13:13:09 +11001919 * 2. per-cpu locks always picked up via for_each_online_cpu() order
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001920 * 3. accurate counter sync requires m_sb_lock + per cpu locks
David Chinner8d280b92006-03-14 13:13:09 +11001921 * 4. modifying per-cpu counters requires holding per-cpu lock
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001922 * 5. modifying global counters requires holding m_sb_lock
1923 * 6. enabling or disabling a counter requires holding the m_sb_lock
David Chinner8d280b92006-03-14 13:13:09 +11001924 * and _none_ of the per-cpu locks.
1925 *
1926 * Disabled counters are only ever re-enabled by a balance operation
1927 * that results in more free resources per CPU than a given threshold.
1928 * To ensure counters don't remain disabled, they are rebalanced when
1929 * the global resource goes above a higher threshold (i.e. some hysteresis
1930 * is present to prevent thrashing).
David Chinner8d280b92006-03-14 13:13:09 +11001931 */
David Chinnere8234a62006-03-14 13:23:52 +11001932
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001933#ifdef CONFIG_HOTPLUG_CPU
David Chinnere8234a62006-03-14 13:23:52 +11001934/*
1935 * hot-plug CPU notifier support.
1936 *
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001937 * We need a notifier per filesystem as we need to be able to identify
1938 * the filesystem to balance the counters out. This is achieved by
1939 * having a notifier block embedded in the xfs_mount_t and doing pointer
1940 * magic to get the mount pointer from the notifier block address.
David Chinnere8234a62006-03-14 13:23:52 +11001941 */
1942STATIC int
1943xfs_icsb_cpu_notify(
1944 struct notifier_block *nfb,
1945 unsigned long action,
1946 void *hcpu)
1947{
1948 xfs_icsb_cnts_t *cntp;
1949 xfs_mount_t *mp;
David Chinnere8234a62006-03-14 13:23:52 +11001950
1951 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
1952 cntp = (xfs_icsb_cnts_t *)
1953 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
1954 switch (action) {
1955 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001956 case CPU_UP_PREPARE_FROZEN:
David Chinnere8234a62006-03-14 13:23:52 +11001957 /* Easy Case - initialize the area and locks, and
1958 * then rebalance when online does everything else for us. */
David Chinner01e1b692006-03-14 13:29:16 +11001959 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinnere8234a62006-03-14 13:23:52 +11001960 break;
1961 case CPU_ONLINE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001962 case CPU_ONLINE_FROZEN:
David Chinner03135cf2007-02-10 18:35:15 +11001963 xfs_icsb_lock(mp);
Christoph Hellwig45af6c62008-04-22 17:34:44 +10001964 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
1965 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
1966 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
David Chinner03135cf2007-02-10 18:35:15 +11001967 xfs_icsb_unlock(mp);
David Chinnere8234a62006-03-14 13:23:52 +11001968 break;
1969 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001970 case CPU_DEAD_FROZEN:
David Chinnere8234a62006-03-14 13:23:52 +11001971 /* Disable all the counters, then fold the dead cpu's
1972 * count into the total on the global superblock and
1973 * re-enable the counters. */
David Chinner03135cf2007-02-10 18:35:15 +11001974 xfs_icsb_lock(mp);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001975 spin_lock(&mp->m_sb_lock);
David Chinnere8234a62006-03-14 13:23:52 +11001976 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
1977 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
1978 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
1979
1980 mp->m_sb.sb_icount += cntp->icsb_icount;
1981 mp->m_sb.sb_ifree += cntp->icsb_ifree;
1982 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
1983
David Chinner01e1b692006-03-14 13:29:16 +11001984 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinnere8234a62006-03-14 13:23:52 +11001985
Christoph Hellwig45af6c62008-04-22 17:34:44 +10001986 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
1987 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
1988 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001989 spin_unlock(&mp->m_sb_lock);
David Chinner03135cf2007-02-10 18:35:15 +11001990 xfs_icsb_unlock(mp);
David Chinnere8234a62006-03-14 13:23:52 +11001991 break;
1992 }
1993
1994 return NOTIFY_OK;
1995}
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001996#endif /* CONFIG_HOTPLUG_CPU */
David Chinnere8234a62006-03-14 13:23:52 +11001997
David Chinner8d280b92006-03-14 13:13:09 +11001998int
1999xfs_icsb_init_counters(
2000 xfs_mount_t *mp)
2001{
2002 xfs_icsb_cnts_t *cntp;
2003 int i;
2004
2005 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2006 if (mp->m_sb_cnts == NULL)
2007 return -ENOMEM;
2008
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002009#ifdef CONFIG_HOTPLUG_CPU
David Chinnere8234a62006-03-14 13:23:52 +11002010 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2011 mp->m_icsb_notifier.priority = 0;
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002012 register_hotcpu_notifier(&mp->m_icsb_notifier);
2013#endif /* CONFIG_HOTPLUG_CPU */
David Chinnere8234a62006-03-14 13:23:52 +11002014
David Chinner8d280b92006-03-14 13:13:09 +11002015 for_each_online_cpu(i) {
2016 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11002017 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinner8d280b92006-03-14 13:13:09 +11002018 }
David Chinner20b64282007-02-10 18:35:09 +11002019
2020 mutex_init(&mp->m_icsb_mutex);
2021
David Chinner8d280b92006-03-14 13:13:09 +11002022 /*
2023 * start with all counters disabled so that the
2024 * initial balance kicks us off correctly
2025 */
2026 mp->m_icsb_counters = -1;
2027 return 0;
2028}
2029
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11002030void
2031xfs_icsb_reinit_counters(
2032 xfs_mount_t *mp)
2033{
2034 xfs_icsb_lock(mp);
2035 /*
2036 * start with all counters disabled so that the
2037 * initial balance kicks us off correctly
2038 */
2039 mp->m_icsb_counters = -1;
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002040 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2041 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2042 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11002043 xfs_icsb_unlock(mp);
2044}
2045
Christoph Hellwigc962fb72008-05-20 15:10:52 +10002046void
David Chinner8d280b92006-03-14 13:13:09 +11002047xfs_icsb_destroy_counters(
2048 xfs_mount_t *mp)
2049{
David Chinnere8234a62006-03-14 13:23:52 +11002050 if (mp->m_sb_cnts) {
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002051 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
David Chinner8d280b92006-03-14 13:13:09 +11002052 free_percpu(mp->m_sb_cnts);
David Chinnere8234a62006-03-14 13:23:52 +11002053 }
David Chinner03135cf2007-02-10 18:35:15 +11002054 mutex_destroy(&mp->m_icsb_mutex);
David Chinner8d280b92006-03-14 13:13:09 +11002055}
2056
David Chinner7989cb82007-02-10 18:34:56 +11002057STATIC_INLINE void
David Chinner01e1b692006-03-14 13:29:16 +11002058xfs_icsb_lock_cntr(
2059 xfs_icsb_cnts_t *icsbp)
2060{
2061 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2062 ndelay(1000);
2063 }
2064}
2065
David Chinner7989cb82007-02-10 18:34:56 +11002066STATIC_INLINE void
David Chinner01e1b692006-03-14 13:29:16 +11002067xfs_icsb_unlock_cntr(
2068 xfs_icsb_cnts_t *icsbp)
2069{
2070 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2071}
2072
David Chinner8d280b92006-03-14 13:13:09 +11002073
David Chinner7989cb82007-02-10 18:34:56 +11002074STATIC_INLINE void
David Chinner8d280b92006-03-14 13:13:09 +11002075xfs_icsb_lock_all_counters(
2076 xfs_mount_t *mp)
2077{
2078 xfs_icsb_cnts_t *cntp;
2079 int i;
2080
2081 for_each_online_cpu(i) {
2082 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11002083 xfs_icsb_lock_cntr(cntp);
David Chinner8d280b92006-03-14 13:13:09 +11002084 }
2085}
2086
David Chinner7989cb82007-02-10 18:34:56 +11002087STATIC_INLINE void
David Chinner8d280b92006-03-14 13:13:09 +11002088xfs_icsb_unlock_all_counters(
2089 xfs_mount_t *mp)
2090{
2091 xfs_icsb_cnts_t *cntp;
2092 int i;
2093
2094 for_each_online_cpu(i) {
2095 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11002096 xfs_icsb_unlock_cntr(cntp);
David Chinner8d280b92006-03-14 13:13:09 +11002097 }
2098}
2099
2100STATIC void
2101xfs_icsb_count(
2102 xfs_mount_t *mp,
2103 xfs_icsb_cnts_t *cnt,
2104 int flags)
2105{
2106 xfs_icsb_cnts_t *cntp;
2107 int i;
2108
2109 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2110
2111 if (!(flags & XFS_ICSB_LAZY_COUNT))
2112 xfs_icsb_lock_all_counters(mp);
2113
2114 for_each_online_cpu(i) {
2115 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2116 cnt->icsb_icount += cntp->icsb_icount;
2117 cnt->icsb_ifree += cntp->icsb_ifree;
2118 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2119 }
2120
2121 if (!(flags & XFS_ICSB_LAZY_COUNT))
2122 xfs_icsb_unlock_all_counters(mp);
2123}
2124
2125STATIC int
2126xfs_icsb_counter_disabled(
2127 xfs_mount_t *mp,
2128 xfs_sb_field_t field)
2129{
2130 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2131 return test_bit(field, &mp->m_icsb_counters);
2132}
2133
David Chinner36fbe6e2008-04-10 12:19:56 +10002134STATIC void
David Chinner8d280b92006-03-14 13:13:09 +11002135xfs_icsb_disable_counter(
2136 xfs_mount_t *mp,
2137 xfs_sb_field_t field)
2138{
2139 xfs_icsb_cnts_t cnt;
2140
2141 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2142
David Chinner20b64282007-02-10 18:35:09 +11002143 /*
2144 * If we are already disabled, then there is nothing to do
2145 * here. We check before locking all the counters to avoid
2146 * the expensive lock operation when being called in the
2147 * slow path and the counter is already disabled. This is
2148 * safe because the only time we set or clear this state is under
2149 * the m_icsb_mutex.
2150 */
2151 if (xfs_icsb_counter_disabled(mp, field))
David Chinner36fbe6e2008-04-10 12:19:56 +10002152 return;
David Chinner20b64282007-02-10 18:35:09 +11002153
David Chinner8d280b92006-03-14 13:13:09 +11002154 xfs_icsb_lock_all_counters(mp);
2155 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2156 /* drain back to superblock */
2157
Christoph Hellwigce461932008-04-22 17:34:50 +10002158 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
David Chinner8d280b92006-03-14 13:13:09 +11002159 switch(field) {
2160 case XFS_SBS_ICOUNT:
2161 mp->m_sb.sb_icount = cnt.icsb_icount;
2162 break;
2163 case XFS_SBS_IFREE:
2164 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2165 break;
2166 case XFS_SBS_FDBLOCKS:
2167 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2168 break;
2169 default:
2170 BUG();
2171 }
2172 }
2173
2174 xfs_icsb_unlock_all_counters(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002175}
2176
2177STATIC void
2178xfs_icsb_enable_counter(
2179 xfs_mount_t *mp,
2180 xfs_sb_field_t field,
2181 uint64_t count,
2182 uint64_t resid)
2183{
2184 xfs_icsb_cnts_t *cntp;
2185 int i;
2186
2187 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2188
2189 xfs_icsb_lock_all_counters(mp);
2190 for_each_online_cpu(i) {
2191 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2192 switch (field) {
2193 case XFS_SBS_ICOUNT:
2194 cntp->icsb_icount = count + resid;
2195 break;
2196 case XFS_SBS_IFREE:
2197 cntp->icsb_ifree = count + resid;
2198 break;
2199 case XFS_SBS_FDBLOCKS:
2200 cntp->icsb_fdblocks = count + resid;
2201 break;
2202 default:
2203 BUG();
2204 break;
2205 }
2206 resid = 0;
2207 }
2208 clear_bit(field, &mp->m_icsb_counters);
2209 xfs_icsb_unlock_all_counters(mp);
2210}
2211
David Chinnerdbcabad2007-02-10 18:36:17 +11002212void
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002213xfs_icsb_sync_counters_locked(
David Chinner8d280b92006-03-14 13:13:09 +11002214 xfs_mount_t *mp,
2215 int flags)
2216{
2217 xfs_icsb_cnts_t cnt;
David Chinner8d280b92006-03-14 13:13:09 +11002218
David Chinner8d280b92006-03-14 13:13:09 +11002219 xfs_icsb_count(mp, &cnt, flags);
2220
David Chinner8d280b92006-03-14 13:13:09 +11002221 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2222 mp->m_sb.sb_icount = cnt.icsb_icount;
2223 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2224 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2225 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2226 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
David Chinner8d280b92006-03-14 13:13:09 +11002227}
2228
2229/*
2230 * Accurate update of per-cpu counters to incore superblock
2231 */
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002232void
David Chinner8d280b92006-03-14 13:13:09 +11002233xfs_icsb_sync_counters(
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002234 xfs_mount_t *mp,
2235 int flags)
David Chinner8d280b92006-03-14 13:13:09 +11002236{
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002237 spin_lock(&mp->m_sb_lock);
2238 xfs_icsb_sync_counters_locked(mp, flags);
2239 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11002240}
2241
2242/*
2243 * Balance and enable/disable counters as necessary.
2244 *
David Chinner20b64282007-02-10 18:35:09 +11002245 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2246 * chosen to be the same number as single on disk allocation chunk per CPU, and
2247 * free blocks is something far enough zero that we aren't going thrash when we
2248 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2249 * prevent looping endlessly when xfs_alloc_space asks for more than will
2250 * be distributed to a single CPU but each CPU has enough blocks to be
2251 * reenabled.
2252 *
2253 * Note that we can be called when counters are already disabled.
2254 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2255 * prevent locking every per-cpu counter needlessly.
David Chinner8d280b92006-03-14 13:13:09 +11002256 */
David Chinner20b64282007-02-10 18:35:09 +11002257
2258#define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
David Chinner4be536d2006-09-07 14:26:50 +10002259#define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
David Chinner20b64282007-02-10 18:35:09 +11002260 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
David Chinner8d280b92006-03-14 13:13:09 +11002261STATIC void
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002262xfs_icsb_balance_counter_locked(
David Chinner8d280b92006-03-14 13:13:09 +11002263 xfs_mount_t *mp,
2264 xfs_sb_field_t field,
David Chinner20b64282007-02-10 18:35:09 +11002265 int min_per_cpu)
David Chinner8d280b92006-03-14 13:13:09 +11002266{
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002267 uint64_t count, resid;
David Chinner8d280b92006-03-14 13:13:09 +11002268 int weight = num_online_cpus();
David Chinner20b64282007-02-10 18:35:09 +11002269 uint64_t min = (uint64_t)min_per_cpu;
David Chinner8d280b92006-03-14 13:13:09 +11002270
David Chinner8d280b92006-03-14 13:13:09 +11002271 /* disable counter and sync counter */
2272 xfs_icsb_disable_counter(mp, field);
2273
2274 /* update counters - first CPU gets residual*/
2275 switch (field) {
2276 case XFS_SBS_ICOUNT:
2277 count = mp->m_sb.sb_icount;
2278 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002279 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002280 return;
David Chinner8d280b92006-03-14 13:13:09 +11002281 break;
2282 case XFS_SBS_IFREE:
2283 count = mp->m_sb.sb_ifree;
2284 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002285 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002286 return;
David Chinner8d280b92006-03-14 13:13:09 +11002287 break;
2288 case XFS_SBS_FDBLOCKS:
2289 count = mp->m_sb.sb_fdblocks;
2290 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002291 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002292 return;
David Chinner8d280b92006-03-14 13:13:09 +11002293 break;
2294 default:
2295 BUG();
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002296 count = resid = 0; /* quiet, gcc */
David Chinner8d280b92006-03-14 13:13:09 +11002297 break;
2298 }
2299
2300 xfs_icsb_enable_counter(mp, field, count, resid);
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002301}
2302
2303STATIC void
2304xfs_icsb_balance_counter(
2305 xfs_mount_t *mp,
2306 xfs_sb_field_t fields,
2307 int min_per_cpu)
2308{
2309 spin_lock(&mp->m_sb_lock);
2310 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2311 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11002312}
2313
David Chinnera8272ce2007-11-23 16:28:09 +11002314STATIC int
David Chinner20b64282007-02-10 18:35:09 +11002315xfs_icsb_modify_counters(
David Chinner8d280b92006-03-14 13:13:09 +11002316 xfs_mount_t *mp,
2317 xfs_sb_field_t field,
David Chinner20f4ebf2007-02-10 18:36:10 +11002318 int64_t delta,
David Chinner20b64282007-02-10 18:35:09 +11002319 int rsvd)
David Chinner8d280b92006-03-14 13:13:09 +11002320{
2321 xfs_icsb_cnts_t *icsbp;
2322 long long lcounter; /* long counter for 64 bit fields */
Eric Sandeen007c61c2007-10-11 17:43:56 +10002323 int cpu, ret = 0;
David Chinner8d280b92006-03-14 13:13:09 +11002324
David Chinner20b64282007-02-10 18:35:09 +11002325 might_sleep();
David Chinner8d280b92006-03-14 13:13:09 +11002326again:
2327 cpu = get_cpu();
David Chinner20b64282007-02-10 18:35:09 +11002328 icsbp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, cpu);
2329
2330 /*
2331 * if the counter is disabled, go to slow path
2332 */
David Chinner8d280b92006-03-14 13:13:09 +11002333 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2334 goto slow_path;
David Chinner20b64282007-02-10 18:35:09 +11002335 xfs_icsb_lock_cntr(icsbp);
2336 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2337 xfs_icsb_unlock_cntr(icsbp);
2338 goto slow_path;
2339 }
David Chinner8d280b92006-03-14 13:13:09 +11002340
2341 switch (field) {
2342 case XFS_SBS_ICOUNT:
2343 lcounter = icsbp->icsb_icount;
2344 lcounter += delta;
2345 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002346 goto balance_counter;
David Chinner8d280b92006-03-14 13:13:09 +11002347 icsbp->icsb_icount = lcounter;
2348 break;
2349
2350 case XFS_SBS_IFREE:
2351 lcounter = icsbp->icsb_ifree;
2352 lcounter += delta;
2353 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002354 goto balance_counter;
David Chinner8d280b92006-03-14 13:13:09 +11002355 icsbp->icsb_ifree = lcounter;
2356 break;
2357
2358 case XFS_SBS_FDBLOCKS:
2359 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2360
David Chinner4be536d2006-09-07 14:26:50 +10002361 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002362 lcounter += delta;
2363 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002364 goto balance_counter;
David Chinner4be536d2006-09-07 14:26:50 +10002365 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002366 break;
2367 default:
2368 BUG();
2369 break;
2370 }
David Chinner01e1b692006-03-14 13:29:16 +11002371 xfs_icsb_unlock_cntr(icsbp);
David Chinner8d280b92006-03-14 13:13:09 +11002372 put_cpu();
David Chinner8d280b92006-03-14 13:13:09 +11002373 return 0;
2374
David Chinner8d280b92006-03-14 13:13:09 +11002375slow_path:
David Chinner20b64282007-02-10 18:35:09 +11002376 put_cpu();
2377
2378 /*
2379 * serialise with a mutex so we don't burn lots of cpu on
2380 * the superblock lock. We still need to hold the superblock
2381 * lock, however, when we modify the global structures.
2382 */
David Chinner03135cf2007-02-10 18:35:15 +11002383 xfs_icsb_lock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002384
2385 /*
2386 * Now running atomically.
2387 *
2388 * If the counter is enabled, someone has beaten us to rebalancing.
2389 * Drop the lock and try again in the fast path....
2390 */
2391 if (!(xfs_icsb_counter_disabled(mp, field))) {
David Chinner03135cf2007-02-10 18:35:15 +11002392 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002393 goto again;
2394 }
2395
2396 /*
2397 * The counter is currently disabled. Because we are
2398 * running atomically here, we know a rebalance cannot
2399 * be in progress. Hence we can go straight to operating
2400 * on the global superblock. We do not call xfs_mod_incore_sb()
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002401 * here even though we need to get the m_sb_lock. Doing so
David Chinner20b64282007-02-10 18:35:09 +11002402 * will cause us to re-enter this function and deadlock.
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002403 * Hence we get the m_sb_lock ourselves and then call
David Chinner20b64282007-02-10 18:35:09 +11002404 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2405 * directly on the global counters.
2406 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002407 spin_lock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11002408 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002409 spin_unlock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11002410
2411 /*
2412 * Now that we've modified the global superblock, we
2413 * may be able to re-enable the distributed counters
2414 * (e.g. lots of space just got freed). After that
2415 * we are done.
2416 */
2417 if (ret != ENOSPC)
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002418 xfs_icsb_balance_counter(mp, field, 0);
David Chinner03135cf2007-02-10 18:35:15 +11002419 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002420 return ret;
2421
2422balance_counter:
David Chinner01e1b692006-03-14 13:29:16 +11002423 xfs_icsb_unlock_cntr(icsbp);
David Chinner8d280b92006-03-14 13:13:09 +11002424 put_cpu();
2425
David Chinner20b64282007-02-10 18:35:09 +11002426 /*
2427 * We may have multiple threads here if multiple per-cpu
2428 * counters run dry at the same time. This will mean we can
2429 * do more balances than strictly necessary but it is not
2430 * the common slowpath case.
2431 */
David Chinner03135cf2007-02-10 18:35:15 +11002432 xfs_icsb_lock(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002433
David Chinner20b64282007-02-10 18:35:09 +11002434 /*
2435 * running atomically.
2436 *
2437 * This will leave the counter in the correct state for future
2438 * accesses. After the rebalance, we simply try again and our retry
2439 * will either succeed through the fast path or slow path without
2440 * another balance operation being required.
2441 */
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002442 xfs_icsb_balance_counter(mp, field, delta);
David Chinner03135cf2007-02-10 18:35:15 +11002443 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002444 goto again;
David Chinner8d280b92006-03-14 13:13:09 +11002445}
2446
David Chinner8d280b92006-03-14 13:13:09 +11002447#endif