blob: 536021fb3d4e338a0e694ac069191c0b0d4d0cf0 [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"
Christoph Hellwig211e4d42012-04-23 15:58:34 +100025#include "xfs_trans_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_sb.h"
27#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_dir2.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#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_dinode.h"
34#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110035#include "xfs_btree.h"
36#include "xfs_ialloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_alloc.h"
38#include "xfs_rtalloc.h"
39#include "xfs_bmap.h"
40#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include "xfs_quota.h"
42#include "xfs_fsops.h"
Christoph Hellwig43355092008-03-27 18:01:08 +110043#include "xfs_utils.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000044#include "xfs_trace.h"
45
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
David Chinner8d280b92006-03-14 13:13:09 +110047#ifdef HAVE_PERCPU_SB
David Chinner20f4ebf2007-02-10 18:36:10 +110048STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
Christoph Hellwig45af6c62008-04-22 17:34:44 +100049 int);
50STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t,
51 int);
David Chinner36fbe6e2008-04-10 12:19:56 +100052STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
David Chinner8d280b92006-03-14 13:13:09 +110053#else
54
Christoph Hellwig45af6c62008-04-22 17:34:44 +100055#define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
56#define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
David Chinner8d280b92006-03-14 13:13:09 +110057#endif
58
Christoph Hellwig1df84c92006-01-11 15:29:52 +110059static const struct {
David Chinner8d280b92006-03-14 13:13:09 +110060 short offset;
61 short type; /* 0 = integer
62 * 1 = binary / string (no translation)
63 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070064} xfs_sb_info[] = {
65 { offsetof(xfs_sb_t, sb_magicnum), 0 },
66 { offsetof(xfs_sb_t, sb_blocksize), 0 },
67 { offsetof(xfs_sb_t, sb_dblocks), 0 },
68 { offsetof(xfs_sb_t, sb_rblocks), 0 },
69 { offsetof(xfs_sb_t, sb_rextents), 0 },
70 { offsetof(xfs_sb_t, sb_uuid), 1 },
71 { offsetof(xfs_sb_t, sb_logstart), 0 },
72 { offsetof(xfs_sb_t, sb_rootino), 0 },
73 { offsetof(xfs_sb_t, sb_rbmino), 0 },
74 { offsetof(xfs_sb_t, sb_rsumino), 0 },
75 { offsetof(xfs_sb_t, sb_rextsize), 0 },
76 { offsetof(xfs_sb_t, sb_agblocks), 0 },
77 { offsetof(xfs_sb_t, sb_agcount), 0 },
78 { offsetof(xfs_sb_t, sb_rbmblocks), 0 },
79 { offsetof(xfs_sb_t, sb_logblocks), 0 },
80 { offsetof(xfs_sb_t, sb_versionnum), 0 },
81 { offsetof(xfs_sb_t, sb_sectsize), 0 },
82 { offsetof(xfs_sb_t, sb_inodesize), 0 },
83 { offsetof(xfs_sb_t, sb_inopblock), 0 },
84 { offsetof(xfs_sb_t, sb_fname[0]), 1 },
85 { offsetof(xfs_sb_t, sb_blocklog), 0 },
86 { offsetof(xfs_sb_t, sb_sectlog), 0 },
87 { offsetof(xfs_sb_t, sb_inodelog), 0 },
88 { offsetof(xfs_sb_t, sb_inopblog), 0 },
89 { offsetof(xfs_sb_t, sb_agblklog), 0 },
90 { offsetof(xfs_sb_t, sb_rextslog), 0 },
91 { offsetof(xfs_sb_t, sb_inprogress), 0 },
92 { offsetof(xfs_sb_t, sb_imax_pct), 0 },
93 { offsetof(xfs_sb_t, sb_icount), 0 },
94 { offsetof(xfs_sb_t, sb_ifree), 0 },
95 { offsetof(xfs_sb_t, sb_fdblocks), 0 },
96 { offsetof(xfs_sb_t, sb_frextents), 0 },
97 { offsetof(xfs_sb_t, sb_uquotino), 0 },
98 { offsetof(xfs_sb_t, sb_gquotino), 0 },
99 { offsetof(xfs_sb_t, sb_qflags), 0 },
100 { offsetof(xfs_sb_t, sb_flags), 0 },
101 { offsetof(xfs_sb_t, sb_shared_vn), 0 },
102 { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
103 { offsetof(xfs_sb_t, sb_unit), 0 },
104 { offsetof(xfs_sb_t, sb_width), 0 },
105 { offsetof(xfs_sb_t, sb_dirblklog), 0 },
106 { offsetof(xfs_sb_t, sb_logsectlog), 0 },
107 { offsetof(xfs_sb_t, sb_logsectsize),0 },
108 { offsetof(xfs_sb_t, sb_logsunit), 0 },
109 { offsetof(xfs_sb_t, sb_features2), 0 },
David Chinneree1c0902008-03-06 13:45:50 +1100110 { offsetof(xfs_sb_t, sb_bad_features2), 0 },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 { sizeof(xfs_sb_t), 0 }
112};
113
Christoph Hellwig27174202009-03-30 10:21:31 +0200114static DEFINE_MUTEX(xfs_uuid_table_mutex);
115static int xfs_uuid_table_size;
116static uuid_t *xfs_uuid_table;
117
118/*
119 * See if the UUID is unique among mounted XFS filesystems.
120 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
121 */
122STATIC int
123xfs_uuid_mount(
124 struct xfs_mount *mp)
125{
126 uuid_t *uuid = &mp->m_sb.sb_uuid;
127 int hole, i;
128
129 if (mp->m_flags & XFS_MOUNT_NOUUID)
130 return 0;
131
132 if (uuid_is_nil(uuid)) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100133 xfs_warn(mp, "Filesystem has nil UUID - can't mount");
Christoph Hellwig27174202009-03-30 10:21:31 +0200134 return XFS_ERROR(EINVAL);
135 }
136
137 mutex_lock(&xfs_uuid_table_mutex);
138 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
139 if (uuid_is_nil(&xfs_uuid_table[i])) {
140 hole = i;
141 continue;
142 }
143 if (uuid_equal(uuid, &xfs_uuid_table[i]))
144 goto out_duplicate;
145 }
146
147 if (hole < 0) {
148 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
149 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
150 xfs_uuid_table_size * sizeof(*xfs_uuid_table),
151 KM_SLEEP);
152 hole = xfs_uuid_table_size++;
153 }
154 xfs_uuid_table[hole] = *uuid;
155 mutex_unlock(&xfs_uuid_table_mutex);
156
157 return 0;
158
159 out_duplicate:
160 mutex_unlock(&xfs_uuid_table_mutex);
Mitsuo Hayasaka021000e2012-01-13 05:58:39 +0000161 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
Christoph Hellwig27174202009-03-30 10:21:31 +0200162 return XFS_ERROR(EINVAL);
163}
164
165STATIC void
166xfs_uuid_unmount(
167 struct xfs_mount *mp)
168{
169 uuid_t *uuid = &mp->m_sb.sb_uuid;
170 int i;
171
172 if (mp->m_flags & XFS_MOUNT_NOUUID)
173 return;
174
175 mutex_lock(&xfs_uuid_table_mutex);
176 for (i = 0; i < xfs_uuid_table_size; i++) {
177 if (uuid_is_nil(&xfs_uuid_table[i]))
178 continue;
179 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
180 continue;
181 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
182 break;
183 }
184 ASSERT(i < xfs_uuid_table_size);
185 mutex_unlock(&xfs_uuid_table_mutex);
186}
187
188
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189/*
Dave Chinner0fa800f2010-01-11 11:47:46 +0000190 * Reference counting access wrappers to the perag structures.
Dave Chinnere1765792010-09-22 10:47:20 +1000191 * Because we never free per-ag structures, the only thing we
192 * have to protect against changes is the tree structure itself.
Dave Chinner0fa800f2010-01-11 11:47:46 +0000193 */
194struct xfs_perag *
195xfs_perag_get(struct xfs_mount *mp, xfs_agnumber_t agno)
196{
197 struct xfs_perag *pag;
198 int ref = 0;
199
Dave Chinnere1765792010-09-22 10:47:20 +1000200 rcu_read_lock();
Dave Chinner0fa800f2010-01-11 11:47:46 +0000201 pag = radix_tree_lookup(&mp->m_perag_tree, agno);
202 if (pag) {
203 ASSERT(atomic_read(&pag->pag_ref) >= 0);
Dave Chinner0fa800f2010-01-11 11:47:46 +0000204 ref = atomic_inc_return(&pag->pag_ref);
205 }
Dave Chinnere1765792010-09-22 10:47:20 +1000206 rcu_read_unlock();
Dave Chinner0fa800f2010-01-11 11:47:46 +0000207 trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
208 return pag;
209}
210
Dave Chinner65d0f202010-09-24 18:40:15 +1000211/*
212 * search from @first to find the next perag with the given tag set.
213 */
214struct xfs_perag *
215xfs_perag_get_tag(
216 struct xfs_mount *mp,
217 xfs_agnumber_t first,
218 int tag)
219{
220 struct xfs_perag *pag;
221 int found;
222 int ref;
223
224 rcu_read_lock();
225 found = radix_tree_gang_lookup_tag(&mp->m_perag_tree,
226 (void **)&pag, first, 1, tag);
227 if (found <= 0) {
228 rcu_read_unlock();
229 return NULL;
230 }
231 ref = atomic_inc_return(&pag->pag_ref);
232 rcu_read_unlock();
233 trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_);
234 return pag;
235}
236
Dave Chinner0fa800f2010-01-11 11:47:46 +0000237void
238xfs_perag_put(struct xfs_perag *pag)
239{
240 int ref;
241
242 ASSERT(atomic_read(&pag->pag_ref) > 0);
243 ref = atomic_dec_return(&pag->pag_ref);
244 trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
245}
246
Dave Chinnere1765792010-09-22 10:47:20 +1000247STATIC void
248__xfs_free_perag(
249 struct rcu_head *head)
250{
251 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
252
253 ASSERT(atomic_read(&pag->pag_ref) == 0);
254 kmem_free(pag);
255}
256
Dave Chinner0fa800f2010-01-11 11:47:46 +0000257/*
Dave Chinnere1765792010-09-22 10:47:20 +1000258 * Free up the per-ag resources associated with the mount structure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 */
Christoph Hellwigc962fb72008-05-20 15:10:52 +1000260STATIC void
Christoph Hellwigff4f0382008-08-13 16:50:47 +1000261xfs_free_perag(
Christoph Hellwig745f6912007-08-30 17:20:39 +1000262 xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263{
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000264 xfs_agnumber_t agno;
265 struct xfs_perag *pag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000267 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
268 spin_lock(&mp->m_perag_lock);
269 pag = radix_tree_delete(&mp->m_perag_tree, agno);
270 spin_unlock(&mp->m_perag_lock);
Dave Chinnere1765792010-09-22 10:47:20 +1000271 ASSERT(pag);
Dave Chinnerf83282a2010-11-08 08:55:04 +0000272 ASSERT(atomic_read(&pag->pag_ref) == 0);
Dave Chinnere1765792010-09-22 10:47:20 +1000273 call_rcu(&pag->rcu_head, __xfs_free_perag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275}
276
Nathan Scott4cc929e2007-05-14 18:24:02 +1000277/*
278 * Check size of device based on the (data/realtime) block count.
279 * Note: this check is used by the growfs code as well as mount.
280 */
281int
282xfs_sb_validate_fsb_count(
283 xfs_sb_t *sbp,
284 __uint64_t nblocks)
285{
286 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
287 ASSERT(sbp->sb_blocklog >= BBSHIFT);
288
289#if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
290 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
Eric Sandeen657a4cf2010-04-30 03:42:49 +0000291 return EFBIG;
Nathan Scott4cc929e2007-05-14 18:24:02 +1000292#else /* Limited by UINT_MAX of sectors */
293 if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
Eric Sandeen657a4cf2010-04-30 03:42:49 +0000294 return EFBIG;
Nathan Scott4cc929e2007-05-14 18:24:02 +1000295#endif
296 return 0;
297}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298
299/*
300 * Check the validity of the SB found.
301 */
302STATIC int
303xfs_mount_validate_sb(
304 xfs_mount_t *mp,
Nathan Scott764d1f82006-03-31 13:04:17 +1000305 xfs_sb_t *sbp,
306 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307{
Dave Chinneraf34e092011-03-07 10:04:35 +1100308 int loud = !(flags & XFS_MFSI_QUIET);
309
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 /*
311 * If the log device and data device have the
312 * same device number, the log is internal.
313 * Consequently, the sb_logstart should be non-zero. If
314 * we have a zero sb_logstart in this case, we may be trying to mount
315 * a volume filesystem in a non-volume manner.
316 */
317 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100318 if (loud)
319 xfs_warn(mp, "bad magic number");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 return XFS_ERROR(EWRONGFS);
321 }
322
Eric Sandeen62118702008-03-06 13:44:28 +1100323 if (!xfs_sb_good_version(sbp)) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100324 if (loud)
325 xfs_warn(mp, "bad version");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 return XFS_ERROR(EWRONGFS);
327 }
328
329 if (unlikely(
330 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100331 if (loud)
332 xfs_warn(mp,
333 "filesystem is marked as having an external log; "
334 "specify logdev on the mount command line.");
Nathan Scott764d1f82006-03-31 13:04:17 +1000335 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 }
337
338 if (unlikely(
339 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100340 if (loud)
341 xfs_warn(mp,
342 "filesystem is marked as having an internal log; "
343 "do not specify logdev on the mount command line.");
Nathan Scott764d1f82006-03-31 13:04:17 +1000344 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 }
346
347 /*
Eric Sandeenc0e090c2011-05-20 21:52:17 +0000348 * More sanity checking. Most of these were stolen directly from
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 * xfs_repair.
350 */
351 if (unlikely(
352 sbp->sb_agcount <= 0 ||
353 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
354 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
355 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
356 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
Olaf Weber2ac00af2009-04-17 16:12:45 -0500357 sbp->sb_sectsize != (1 << sbp->sb_sectlog) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
359 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
360 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
361 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
Olaf Weber2ac00af2009-04-17 16:12:45 -0500362 sbp->sb_blocksize != (1 << sbp->sb_blocklog) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
364 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
Nathan Scott9f989c92006-03-14 13:29:32 +1100365 sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
366 sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
Olaf Weber2ac00af2009-04-17 16:12:45 -0500367 sbp->sb_inodesize != (1 << sbp->sb_inodelog) ||
Nathan Scott9f989c92006-03-14 13:29:32 +1100368 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
370 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
Eric Sandeenc0e090c2011-05-20 21:52:17 +0000371 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */) ||
372 sbp->sb_dblocks == 0 ||
373 sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp) ||
374 sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp))) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100375 if (loud)
Eric Sandeenc0e090c2011-05-20 21:52:17 +0000376 XFS_CORRUPTION_ERROR("SB sanity check failed",
377 XFS_ERRLEVEL_LOW, mp, sbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 return XFS_ERROR(EFSCORRUPTED);
379 }
380
Lachlan McIlroy2edbddd2008-06-27 13:34:34 +1000381 /*
382 * Until this is fixed only page-sized or smaller data blocks work.
383 */
384 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100385 if (loud) {
386 xfs_warn(mp,
387 "File system with blocksize %d bytes. "
388 "Only pagesize (%ld) or less will currently work.",
389 sbp->sb_blocksize, PAGE_SIZE);
390 }
Lachlan McIlroy2edbddd2008-06-27 13:34:34 +1000391 return XFS_ERROR(ENOSYS);
392 }
393
Christoph Hellwig1a5902c2009-03-29 19:26:46 +0200394 /*
395 * Currently only very few inode sizes are supported.
396 */
397 switch (sbp->sb_inodesize) {
398 case 256:
399 case 512:
400 case 1024:
401 case 2048:
402 break;
403 default:
Dave Chinneraf34e092011-03-07 10:04:35 +1100404 if (loud)
405 xfs_warn(mp, "inode size of %d bytes not supported",
406 sbp->sb_inodesize);
Christoph Hellwig1a5902c2009-03-29 19:26:46 +0200407 return XFS_ERROR(ENOSYS);
408 }
409
Nathan Scott4cc929e2007-05-14 18:24:02 +1000410 if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
411 xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100412 if (loud)
413 xfs_warn(mp,
414 "file system too large to be mounted on this system.");
Eric Sandeen657a4cf2010-04-30 03:42:49 +0000415 return XFS_ERROR(EFBIG);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 }
417
418 if (unlikely(sbp->sb_inprogress)) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100419 if (loud)
420 xfs_warn(mp, "file system busy");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 return XFS_ERROR(EFSCORRUPTED);
422 }
423
424 /*
Nathan Scottde206142005-05-05 13:24:13 -0700425 * Version 1 directory format has never worked on Linux.
426 */
Eric Sandeen62118702008-03-06 13:44:28 +1100427 if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100428 if (loud)
429 xfs_warn(mp,
430 "file system using version 1 directory format");
Nathan Scottde206142005-05-05 13:24:13 -0700431 return XFS_ERROR(ENOSYS);
432 }
433
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 return 0;
435}
436
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000437int
Nathan Scottc11e2c32005-11-02 15:11:45 +1100438xfs_initialize_perag(
Nathan Scottc11e2c32005-11-02 15:11:45 +1100439 xfs_mount_t *mp,
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000440 xfs_agnumber_t agcount,
441 xfs_agnumber_t *maxagi)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442{
443 xfs_agnumber_t index, max_metadata;
Dave Chinner8b26c582010-01-11 11:47:48 +0000444 xfs_agnumber_t first_initialised = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 xfs_perag_t *pag;
446 xfs_agino_t agino;
447 xfs_ino_t ino;
448 xfs_sb_t *sbp = &mp->m_sb;
Dave Chinner8b26c582010-01-11 11:47:48 +0000449 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000451 /*
452 * Walk the current per-ag tree so we don't try to initialise AGs
453 * that already exist (growfs case). Allocate and insert all the
454 * AGs we don't find ready for initialisation.
455 */
456 for (index = 0; index < agcount; index++) {
457 pag = xfs_perag_get(mp, index);
458 if (pag) {
459 xfs_perag_put(pag);
460 continue;
461 }
Dave Chinner8b26c582010-01-11 11:47:48 +0000462 if (!first_initialised)
463 first_initialised = index;
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000464
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000465 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
466 if (!pag)
Dave Chinner8b26c582010-01-11 11:47:48 +0000467 goto out_unwind;
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000468 pag->pag_agno = index;
469 pag->pag_mount = mp;
Dave Chinner1a427ab2010-12-16 17:08:41 +1100470 spin_lock_init(&pag->pag_ici_lock);
Dave Chinner69b491c2010-09-27 11:09:51 +1000471 mutex_init(&pag->pag_ici_reclaim_lock);
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000472 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
Dave Chinner74f75a02010-09-24 19:59:04 +1000473 spin_lock_init(&pag->pag_buf_lock);
474 pag->pag_buf_tree = RB_ROOT;
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000475
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000476 if (radix_tree_preload(GFP_NOFS))
Dave Chinner8b26c582010-01-11 11:47:48 +0000477 goto out_unwind;
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000478
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000479 spin_lock(&mp->m_perag_lock);
480 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
481 BUG();
482 spin_unlock(&mp->m_perag_lock);
Dave Chinner8b26c582010-01-11 11:47:48 +0000483 radix_tree_preload_end();
484 error = -EEXIST;
485 goto out_unwind;
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000486 }
487 spin_unlock(&mp->m_perag_lock);
488 radix_tree_preload_end();
489 }
490
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000491 /*
492 * If we mount with the inode64 option, or no inode overflows
493 * the legacy 32-bit address space clear the inode32 option.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 */
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000495 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
496 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000498 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
499 mp->m_flags |= XFS_MOUNT_32BITINODES;
500 else
501 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
502
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000504 /*
505 * Calculate how much should be reserved for inodes to meet
506 * the max inode percentage.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 */
508 if (mp->m_maxicount) {
509 __uint64_t icount;
510
511 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
512 do_div(icount, 100);
513 icount += sbp->sb_agblocks - 1;
Eric Sandeena749ee82005-11-02 15:13:42 +1100514 do_div(icount, sbp->sb_agblocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 max_metadata = icount;
516 } else {
517 max_metadata = agcount;
518 }
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000519
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 for (index = 0; index < agcount; index++) {
521 ino = XFS_AGINO_TO_INO(mp, index, agino);
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000522 if (ino > XFS_MAXINUMBER_32) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 index++;
524 break;
525 }
526
Dave Chinner44b56e02010-01-11 11:47:43 +0000527 pag = xfs_perag_get(mp, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 pag->pagi_inodeok = 1;
529 if (index < max_metadata)
530 pag->pagf_metadata = 1;
Dave Chinner44b56e02010-01-11 11:47:43 +0000531 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 }
533 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 for (index = 0; index < agcount; index++) {
Dave Chinner44b56e02010-01-11 11:47:43 +0000535 pag = xfs_perag_get(mp, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 pag->pagi_inodeok = 1;
Dave Chinner44b56e02010-01-11 11:47:43 +0000537 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 }
539 }
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000540
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000541 if (maxagi)
542 *maxagi = index;
543 return 0;
Dave Chinner8b26c582010-01-11 11:47:48 +0000544
545out_unwind:
546 kmem_free(pag);
547 for (; index > first_initialised; index--) {
548 pag = radix_tree_delete(&mp->m_perag_tree, index);
549 kmem_free(pag);
550 }
551 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552}
553
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000554void
555xfs_sb_from_disk(
Chandra Seetharaman6bd92a22012-01-23 17:31:37 +0000556 struct xfs_mount *mp,
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000557 xfs_dsb_t *from)
558{
Chandra Seetharaman6bd92a22012-01-23 17:31:37 +0000559 struct xfs_sb *to = &mp->m_sb;
560
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000561 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
562 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
563 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
564 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
565 to->sb_rextents = be64_to_cpu(from->sb_rextents);
566 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
567 to->sb_logstart = be64_to_cpu(from->sb_logstart);
568 to->sb_rootino = be64_to_cpu(from->sb_rootino);
569 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
570 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
571 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
572 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
573 to->sb_agcount = be32_to_cpu(from->sb_agcount);
574 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
575 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
576 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
577 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
578 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
579 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
580 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
581 to->sb_blocklog = from->sb_blocklog;
582 to->sb_sectlog = from->sb_sectlog;
583 to->sb_inodelog = from->sb_inodelog;
584 to->sb_inopblog = from->sb_inopblog;
585 to->sb_agblklog = from->sb_agblklog;
586 to->sb_rextslog = from->sb_rextslog;
587 to->sb_inprogress = from->sb_inprogress;
588 to->sb_imax_pct = from->sb_imax_pct;
589 to->sb_icount = be64_to_cpu(from->sb_icount);
590 to->sb_ifree = be64_to_cpu(from->sb_ifree);
591 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
592 to->sb_frextents = be64_to_cpu(from->sb_frextents);
593 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
594 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
595 to->sb_qflags = be16_to_cpu(from->sb_qflags);
596 to->sb_flags = from->sb_flags;
597 to->sb_shared_vn = from->sb_shared_vn;
598 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
599 to->sb_unit = be32_to_cpu(from->sb_unit);
600 to->sb_width = be32_to_cpu(from->sb_width);
601 to->sb_dirblklog = from->sb_dirblklog;
602 to->sb_logsectlog = from->sb_logsectlog;
603 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
604 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
605 to->sb_features2 = be32_to_cpu(from->sb_features2);
David Chinneree1c0902008-03-06 13:45:50 +1100606 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000607}
608
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609/*
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000610 * Copy in core superblock to ondisk one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 *
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000612 * The fields argument is mask of superblock fields to copy.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 */
614void
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000615xfs_sb_to_disk(
616 xfs_dsb_t *to,
617 xfs_sb_t *from,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 __int64_t fields)
619{
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000620 xfs_caddr_t to_ptr = (xfs_caddr_t)to;
621 xfs_caddr_t from_ptr = (xfs_caddr_t)from;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 xfs_sb_field_t f;
623 int first;
624 int size;
625
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 ASSERT(fields);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 if (!fields)
628 return;
629
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 while (fields) {
631 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
632 first = xfs_sb_info[f].offset;
633 size = xfs_sb_info[f + 1].offset - first;
634
635 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
636
637 if (size == 1 || xfs_sb_info[f].type == 1) {
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000638 memcpy(to_ptr + first, from_ptr + first, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 } else {
640 switch (size) {
641 case 2:
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000642 *(__be16 *)(to_ptr + first) =
643 cpu_to_be16(*(__u16 *)(from_ptr + first));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 break;
645 case 4:
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000646 *(__be32 *)(to_ptr + first) =
647 cpu_to_be32(*(__u32 *)(from_ptr + first));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 break;
649 case 8:
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000650 *(__be64 *)(to_ptr + first) =
651 cpu_to_be64(*(__u64 *)(from_ptr + first));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 break;
653 default:
654 ASSERT(0);
655 }
656 }
657
658 fields &= ~(1LL << f);
659 }
660}
661
662/*
663 * xfs_readsb
664 *
665 * Does the initial read of the superblock.
666 */
667int
Nathan Scott764d1f82006-03-31 13:04:17 +1000668xfs_readsb(xfs_mount_t *mp, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669{
670 unsigned int sector_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 int error;
Dave Chinneraf34e092011-03-07 10:04:35 +1100673 int loud = !(flags & XFS_MFSI_QUIET);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
675 ASSERT(mp->m_sb_bp == NULL);
676 ASSERT(mp->m_ddev_targp != NULL);
677
678 /*
679 * Allocate a (locked) buffer to hold the superblock.
680 * This will be kept around at all times to optimize
681 * access to the superblock.
682 */
683 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684
Dave Chinner26af6552010-09-22 10:47:20 +1000685reread:
Dave Chinnere70b73f2012-04-23 15:58:49 +1000686 bp = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
687 BTOBB(sector_size), 0);
Dave Chinner26af6552010-09-22 10:47:20 +1000688 if (!bp) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100689 if (loud)
690 xfs_warn(mp, "SB buffer read failed");
Dave Chinner26af6552010-09-22 10:47:20 +1000691 return EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693
694 /*
695 * Initialize the mount structure from the superblock.
696 * But first do some basic consistency checking.
697 */
Chandra Seetharaman6bd92a22012-01-23 17:31:37 +0000698 xfs_sb_from_disk(mp, XFS_BUF_TO_SBP(bp));
Nathan Scott764d1f82006-03-31 13:04:17 +1000699 error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 if (error) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100701 if (loud)
702 xfs_warn(mp, "SB validate failed");
Dave Chinner26af6552010-09-22 10:47:20 +1000703 goto release_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 }
705
706 /*
707 * We must be able to do sector-sized and sector-aligned IO.
708 */
709 if (sector_size > mp->m_sb.sb_sectsize) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100710 if (loud)
711 xfs_warn(mp, "device supports %u byte sectors (not %u)",
712 sector_size, mp->m_sb.sb_sectsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 error = ENOSYS;
Dave Chinner26af6552010-09-22 10:47:20 +1000714 goto release_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 }
716
717 /*
718 * If device sector size is smaller than the superblock size,
719 * re-read the superblock so the buffer is correctly sized.
720 */
721 if (sector_size < mp->m_sb.sb_sectsize) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 xfs_buf_relse(bp);
723 sector_size = mp->m_sb.sb_sectsize;
Dave Chinner26af6552010-09-22 10:47:20 +1000724 goto reread;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 }
726
Lachlan McIlroy5478eea2007-02-10 18:36:29 +1100727 /* Initialize per-cpu counters */
728 xfs_icsb_reinit_counters(mp);
David Chinner8d280b92006-03-14 13:13:09 +1100729
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 mp->m_sb_bp = bp;
Dave Chinner26af6552010-09-22 10:47:20 +1000731 xfs_buf_unlock(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 return 0;
733
Dave Chinner26af6552010-09-22 10:47:20 +1000734release_buf:
735 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 return error;
737}
738
739
740/*
741 * xfs_mount_common
742 *
743 * Mount initialization code establishing various mount
744 * fields from the superblock associated with the given
745 * mount structure
746 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000747STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
749{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 mp->m_agfrotor = mp->m_agirotor = 0;
Eric Sandeen007c61c2007-10-11 17:43:56 +1000751 spin_lock_init(&mp->m_agirotor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 mp->m_maxagi = mp->m_sb.sb_agcount;
753 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
754 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
755 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
756 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
757 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 mp->m_blockmask = sbp->sb_blocksize - 1;
759 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
760 mp->m_blockwmask = mp->m_blockwsize - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761
Christoph Hellwig60197e82008-10-30 17:11:19 +1100762 mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
763 mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
764 mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
765 mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
766
767 mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
768 mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
769 mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
770 mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
771
772 mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
773 mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
774 mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
775 mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776
777 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
778 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
779 sbp->sb_inopblock);
780 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
781}
David Chinner92821e22007-05-24 15:26:31 +1000782
783/*
784 * xfs_initialize_perag_data
785 *
786 * Read in each per-ag structure so we can count up the number of
787 * allocated inodes, free inodes and used filesystem blocks as this
788 * information is no longer persistent in the superblock. Once we have
789 * this information, write it into the in-core superblock structure.
790 */
791STATIC int
792xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
793{
794 xfs_agnumber_t index;
795 xfs_perag_t *pag;
796 xfs_sb_t *sbp = &mp->m_sb;
797 uint64_t ifree = 0;
798 uint64_t ialloc = 0;
799 uint64_t bfree = 0;
800 uint64_t bfreelst = 0;
801 uint64_t btree = 0;
802 int error;
David Chinner92821e22007-05-24 15:26:31 +1000803
804 for (index = 0; index < agcount; index++) {
805 /*
806 * read the agf, then the agi. This gets us
Malcolm Parsons9da096f2009-03-29 09:55:42 +0200807 * all the information we need and populates the
David Chinner92821e22007-05-24 15:26:31 +1000808 * per-ag structures for us.
809 */
810 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
811 if (error)
812 return error;
813
814 error = xfs_ialloc_pagi_init(mp, NULL, index);
815 if (error)
816 return error;
Dave Chinner44b56e02010-01-11 11:47:43 +0000817 pag = xfs_perag_get(mp, index);
David Chinner92821e22007-05-24 15:26:31 +1000818 ifree += pag->pagi_freecount;
819 ialloc += pag->pagi_count;
820 bfree += pag->pagf_freeblks;
821 bfreelst += pag->pagf_flcount;
822 btree += pag->pagf_btreeblks;
Dave Chinner44b56e02010-01-11 11:47:43 +0000823 xfs_perag_put(pag);
David Chinner92821e22007-05-24 15:26:31 +1000824 }
825 /*
826 * Overwrite incore superblock counters with just-read data
827 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +1000828 spin_lock(&mp->m_sb_lock);
David Chinner92821e22007-05-24 15:26:31 +1000829 sbp->sb_ifree = ifree;
830 sbp->sb_icount = ialloc;
831 sbp->sb_fdblocks = bfree + bfreelst + btree;
Eric Sandeen3685c2a2007-10-11 17:42:32 +1000832 spin_unlock(&mp->m_sb_lock);
David Chinner92821e22007-05-24 15:26:31 +1000833
834 /* Fixup the per-cpu counters as well. */
835 xfs_icsb_reinit_counters(mp);
836
837 return 0;
838}
839
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840/*
Eric Sandeen0771fb42007-10-12 11:03:40 +1000841 * Update alignment values based on mount options and sb values
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 */
Eric Sandeen0771fb42007-10-12 11:03:40 +1000843STATIC int
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100844xfs_update_alignment(xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 xfs_sb_t *sbp = &(mp->m_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
Christoph Hellwig42490232008-08-13 16:49:32 +1000848 if (mp->m_dalign) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 /*
850 * If stripe unit and stripe width are not multiples
851 * of the fs blocksize turn off alignment.
852 */
853 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
854 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
855 if (mp->m_flags & XFS_MOUNT_RETERR) {
Eric Sandeenc0e090c2011-05-20 21:52:17 +0000856 xfs_warn(mp, "alignment check failed: "
857 "(sunit/swidth vs. blocksize)");
Eric Sandeen0771fb42007-10-12 11:03:40 +1000858 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 }
860 mp->m_dalign = mp->m_swidth = 0;
861 } else {
862 /*
863 * Convert the stripe unit and width to FSBs.
864 */
865 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
866 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
867 if (mp->m_flags & XFS_MOUNT_RETERR) {
Eric Sandeenc0e090c2011-05-20 21:52:17 +0000868 xfs_warn(mp, "alignment check failed: "
869 "(sunit/swidth vs. ag size)");
Eric Sandeen0771fb42007-10-12 11:03:40 +1000870 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 }
Dave Chinner53487782011-03-07 10:05:35 +1100872 xfs_warn(mp,
873 "stripe alignment turned off: sunit(%d)/swidth(%d) "
874 "incompatible with agsize(%d)",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 mp->m_dalign, mp->m_swidth,
876 sbp->sb_agblocks);
877
878 mp->m_dalign = 0;
879 mp->m_swidth = 0;
880 } else if (mp->m_dalign) {
881 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
882 } else {
883 if (mp->m_flags & XFS_MOUNT_RETERR) {
Eric Sandeenc0e090c2011-05-20 21:52:17 +0000884 xfs_warn(mp, "alignment check failed: "
885 "sunit(%d) less than bsize(%d)",
Dave Chinner53487782011-03-07 10:05:35 +1100886 mp->m_dalign,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 mp->m_blockmask +1);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000888 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 }
890 mp->m_swidth = 0;
891 }
892 }
893
894 /*
895 * Update superblock with new values
896 * and log changes
897 */
Eric Sandeen62118702008-03-06 13:44:28 +1100898 if (xfs_sb_version_hasdalign(sbp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 if (sbp->sb_unit != mp->m_dalign) {
900 sbp->sb_unit = mp->m_dalign;
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100901 mp->m_update_flags |= XFS_SB_UNIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 }
903 if (sbp->sb_width != mp->m_swidth) {
904 sbp->sb_width = mp->m_swidth;
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100905 mp->m_update_flags |= XFS_SB_WIDTH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 }
907 }
908 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
Eric Sandeen62118702008-03-06 13:44:28 +1100909 xfs_sb_version_hasdalign(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 mp->m_dalign = sbp->sb_unit;
911 mp->m_swidth = sbp->sb_width;
912 }
913
Eric Sandeen0771fb42007-10-12 11:03:40 +1000914 return 0;
915}
916
917/*
918 * Set the maximum inode count for this filesystem
919 */
920STATIC void
921xfs_set_maxicount(xfs_mount_t *mp)
922{
923 xfs_sb_t *sbp = &(mp->m_sb);
924 __uint64_t icount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925
926 if (sbp->sb_imax_pct) {
Eric Sandeen0771fb42007-10-12 11:03:40 +1000927 /*
928 * Make sure the maximum inode count is a multiple
929 * of the units we allocate inodes in.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
932 do_div(icount, 100);
933 do_div(icount, mp->m_ialloc_blks);
934 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
935 sbp->sb_inopblog;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000936 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 mp->m_maxicount = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 }
Eric Sandeen0771fb42007-10-12 11:03:40 +1000939}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940
Eric Sandeen0771fb42007-10-12 11:03:40 +1000941/*
942 * Set the default minimum read and write sizes unless
943 * already specified in a mount option.
944 * We use smaller I/O sizes when the file system
945 * is being used for NFS service (wsync mount option).
946 */
947STATIC void
948xfs_set_rw_sizes(xfs_mount_t *mp)
949{
950 xfs_sb_t *sbp = &(mp->m_sb);
951 int readio_log, writeio_log;
952
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
954 if (mp->m_flags & XFS_MOUNT_WSYNC) {
955 readio_log = XFS_WSYNC_READIO_LOG;
956 writeio_log = XFS_WSYNC_WRITEIO_LOG;
957 } else {
958 readio_log = XFS_READIO_LOG_LARGE;
959 writeio_log = XFS_WRITEIO_LOG_LARGE;
960 }
961 } else {
962 readio_log = mp->m_readio_log;
963 writeio_log = mp->m_writeio_log;
964 }
965
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 if (sbp->sb_blocklog > readio_log) {
967 mp->m_readio_log = sbp->sb_blocklog;
968 } else {
969 mp->m_readio_log = readio_log;
970 }
971 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
972 if (sbp->sb_blocklog > writeio_log) {
973 mp->m_writeio_log = sbp->sb_blocklog;
974 } else {
975 mp->m_writeio_log = writeio_log;
976 }
977 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000978}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979
Eric Sandeen0771fb42007-10-12 11:03:40 +1000980/*
Dave Chinner055388a2011-01-04 11:35:03 +1100981 * precalculate the low space thresholds for dynamic speculative preallocation.
982 */
983void
984xfs_set_low_space_thresholds(
985 struct xfs_mount *mp)
986{
987 int i;
988
989 for (i = 0; i < XFS_LOWSP_MAX; i++) {
990 __uint64_t space = mp->m_sb.sb_dblocks;
991
992 do_div(space, 100);
993 mp->m_low_space[i] = space * (i + 1);
994 }
995}
996
997
998/*
Eric Sandeen0771fb42007-10-12 11:03:40 +1000999 * Set whether we're using inode alignment.
1000 */
1001STATIC void
1002xfs_set_inoalignment(xfs_mount_t *mp)
1003{
Eric Sandeen62118702008-03-06 13:44:28 +11001004 if (xfs_sb_version_hasalign(&mp->m_sb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 mp->m_sb.sb_inoalignmt >=
1006 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
1007 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
1008 else
1009 mp->m_inoalign_mask = 0;
1010 /*
1011 * If we are using stripe alignment, check whether
1012 * the stripe unit is a multiple of the inode alignment
1013 */
1014 if (mp->m_dalign && mp->m_inoalign_mask &&
1015 !(mp->m_dalign & mp->m_inoalign_mask))
1016 mp->m_sinoalign = mp->m_dalign;
1017 else
1018 mp->m_sinoalign = 0;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001019}
1020
1021/*
1022 * Check that the data (and log if separate) are an ok size.
1023 */
1024STATIC int
Christoph Hellwig42490232008-08-13 16:49:32 +10001025xfs_check_sizes(xfs_mount_t *mp)
Eric Sandeen0771fb42007-10-12 11:03:40 +10001026{
1027 xfs_buf_t *bp;
1028 xfs_daddr_t d;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001029
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
1031 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001032 xfs_warn(mp, "filesystem size mismatch detected");
Eric Sandeen657a4cf2010-04-30 03:42:49 +00001033 return XFS_ERROR(EFBIG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 }
Dave Chinnere70b73f2012-04-23 15:58:49 +10001035 bp = xfs_buf_read_uncached(mp->m_ddev_targp,
Dave Chinner1922c942010-09-22 10:47:20 +10001036 d - XFS_FSS_TO_BB(mp, 1),
Dave Chinnere70b73f2012-04-23 15:58:49 +10001037 XFS_FSS_TO_BB(mp, 1), 0);
Dave Chinner1922c942010-09-22 10:47:20 +10001038 if (!bp) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001039 xfs_warn(mp, "last sector read failed");
Dave Chinner1922c942010-09-22 10:47:20 +10001040 return EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 }
Dave Chinner1922c942010-09-22 10:47:20 +10001042 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
Christoph Hellwig42490232008-08-13 16:49:32 +10001044 if (mp->m_logdev_targp != mp->m_ddev_targp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
1046 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001047 xfs_warn(mp, "log size mismatch detected");
Eric Sandeen657a4cf2010-04-30 03:42:49 +00001048 return XFS_ERROR(EFBIG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 }
Dave Chinnere70b73f2012-04-23 15:58:49 +10001050 bp = xfs_buf_read_uncached(mp->m_logdev_targp,
Dave Chinner1922c942010-09-22 10:47:20 +10001051 d - XFS_FSB_TO_BB(mp, 1),
Dave Chinnere70b73f2012-04-23 15:58:49 +10001052 XFS_FSB_TO_BB(mp, 1), 0);
Dave Chinner1922c942010-09-22 10:47:20 +10001053 if (!bp) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001054 xfs_warn(mp, "log device read failed");
Dave Chinner1922c942010-09-22 10:47:20 +10001055 return EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 }
Dave Chinner1922c942010-09-22 10:47:20 +10001057 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 }
Eric Sandeen0771fb42007-10-12 11:03:40 +10001059 return 0;
1060}
1061
1062/*
Christoph Hellwig7d095252009-06-08 15:33:32 +02001063 * Clear the quotaflags in memory and in the superblock.
1064 */
1065int
1066xfs_mount_reset_sbqflags(
1067 struct xfs_mount *mp)
1068{
1069 int error;
1070 struct xfs_trans *tp;
1071
1072 mp->m_qflags = 0;
1073
1074 /*
1075 * It is OK to look at sb_qflags here in mount path,
1076 * without m_sb_lock.
1077 */
1078 if (mp->m_sb.sb_qflags == 0)
1079 return 0;
1080 spin_lock(&mp->m_sb_lock);
1081 mp->m_sb.sb_qflags = 0;
1082 spin_unlock(&mp->m_sb_lock);
1083
1084 /*
1085 * If the fs is readonly, let the incore superblock run
1086 * with quotas off but don't flush the update out to disk
1087 */
1088 if (mp->m_flags & XFS_MOUNT_RDONLY)
1089 return 0;
1090
Christoph Hellwig7d095252009-06-08 15:33:32 +02001091 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1092 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1093 XFS_DEFAULT_LOG_COUNT);
1094 if (error) {
1095 xfs_trans_cancel(tp, 0);
Dave Chinner53487782011-03-07 10:05:35 +11001096 xfs_alert(mp, "%s: Superblock update failed!", __func__);
Christoph Hellwig7d095252009-06-08 15:33:32 +02001097 return error;
1098 }
1099
1100 xfs_mod_sb(tp, XFS_SB_QFLAGS);
1101 return xfs_trans_commit(tp, 0);
1102}
1103
Eric Sandeend5db0f92010-02-05 22:59:53 +00001104__uint64_t
1105xfs_default_resblks(xfs_mount_t *mp)
1106{
1107 __uint64_t resblks;
1108
1109 /*
Dave Chinner8babd8a2010-03-04 01:46:25 +00001110 * We default to 5% or 8192 fsbs of space reserved, whichever is
1111 * smaller. This is intended to cover concurrent allocation
1112 * transactions when we initially hit enospc. These each require a 4
1113 * block reservation. Hence by default we cover roughly 2000 concurrent
1114 * allocation reservations.
Eric Sandeend5db0f92010-02-05 22:59:53 +00001115 */
1116 resblks = mp->m_sb.sb_dblocks;
1117 do_div(resblks, 20);
Dave Chinner8babd8a2010-03-04 01:46:25 +00001118 resblks = min_t(__uint64_t, resblks, 8192);
Eric Sandeend5db0f92010-02-05 22:59:53 +00001119 return resblks;
1120}
1121
Christoph Hellwig7d095252009-06-08 15:33:32 +02001122/*
Eric Sandeen0771fb42007-10-12 11:03:40 +10001123 * This function does the following on an initial mount of a file system:
1124 * - reads the superblock from disk and init the mount struct
1125 * - if we're a 32-bit kernel, do a size check on the superblock
1126 * so we don't mount terabyte filesystems
1127 * - init mount struct realtime fields
1128 * - allocate inode hash table for fs
1129 * - init directory manager
1130 * - perform recovery and init the log manager
1131 */
1132int
1133xfs_mountfs(
Christoph Hellwig42490232008-08-13 16:49:32 +10001134 xfs_mount_t *mp)
Eric Sandeen0771fb42007-10-12 11:03:40 +10001135{
1136 xfs_sb_t *sbp = &(mp->m_sb);
1137 xfs_inode_t *rip;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001138 __uint64_t resblks;
Christoph Hellwig7d095252009-06-08 15:33:32 +02001139 uint quotamount = 0;
1140 uint quotaflags = 0;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001141 int error = 0;
1142
Eric Sandeen0771fb42007-10-12 11:03:40 +10001143 xfs_mount_common(mp, sbp);
1144
1145 /*
Eric Sandeene6957ea2008-04-10 12:19:34 +10001146 * Check for a mismatched features2 values. Older kernels
1147 * read & wrote into the wrong sb offset for sb_features2
1148 * on some platforms due to xfs_sb_t not being 64bit size aligned
1149 * when sb_features2 was added, which made older superblock
1150 * reading/writing routines swap it as a 64-bit value.
David Chinneree1c0902008-03-06 13:45:50 +11001151 *
Eric Sandeene6957ea2008-04-10 12:19:34 +10001152 * For backwards compatibility, we make both slots equal.
1153 *
1154 * If we detect a mismatched field, we OR the set bits into the
1155 * existing features2 field in case it has already been modified; we
1156 * don't want to lose any features. We then update the bad location
1157 * with the ORed value so that older kernels will see any features2
1158 * flags, and mark the two fields as needing updates once the
1159 * transaction subsystem is online.
David Chinneree1c0902008-03-06 13:45:50 +11001160 */
Eric Sandeene6957ea2008-04-10 12:19:34 +10001161 if (xfs_sb_has_mismatched_features2(sbp)) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001162 xfs_warn(mp, "correcting sb_features alignment problem");
David Chinneree1c0902008-03-06 13:45:50 +11001163 sbp->sb_features2 |= sbp->sb_bad_features2;
Eric Sandeene6957ea2008-04-10 12:19:34 +10001164 sbp->sb_bad_features2 = sbp->sb_features2;
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001165 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
Eric Sandeene6957ea2008-04-10 12:19:34 +10001166
1167 /*
1168 * Re-check for ATTR2 in case it was found in bad_features2
1169 * slot.
1170 */
Tim Shimmin7c12f292008-04-30 18:15:28 +10001171 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1172 !(mp->m_flags & XFS_MOUNT_NOATTR2))
Eric Sandeene6957ea2008-04-10 12:19:34 +10001173 mp->m_flags |= XFS_MOUNT_ATTR2;
Tim Shimmin7c12f292008-04-30 18:15:28 +10001174 }
Eric Sandeene6957ea2008-04-10 12:19:34 +10001175
Tim Shimmin7c12f292008-04-30 18:15:28 +10001176 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1177 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1178 xfs_sb_version_removeattr2(&mp->m_sb);
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001179 mp->m_update_flags |= XFS_SB_FEATURES2;
Tim Shimmin7c12f292008-04-30 18:15:28 +10001180
1181 /* update sb_versionnum for the clearing of the morebits */
1182 if (!sbp->sb_features2)
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001183 mp->m_update_flags |= XFS_SB_VERSIONNUM;
David Chinneree1c0902008-03-06 13:45:50 +11001184 }
1185
1186 /*
Eric Sandeen0771fb42007-10-12 11:03:40 +10001187 * Check if sb_agblocks is aligned at stripe boundary
1188 * If sb_agblocks is NOT aligned turn off m_dalign since
1189 * allocator alignment is within an ag, therefore ag has
1190 * to be aligned at stripe boundary.
1191 */
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001192 error = xfs_update_alignment(mp);
Eric Sandeen0771fb42007-10-12 11:03:40 +10001193 if (error)
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001194 goto out;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001195
1196 xfs_alloc_compute_maxlevels(mp);
1197 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
1198 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
1199 xfs_ialloc_compute_maxlevels(mp);
1200
1201 xfs_set_maxicount(mp);
1202
1203 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
1204
Christoph Hellwig27174202009-03-30 10:21:31 +02001205 error = xfs_uuid_mount(mp);
1206 if (error)
1207 goto out;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001208
1209 /*
1210 * Set the minimum read and write sizes
1211 */
1212 xfs_set_rw_sizes(mp);
1213
Dave Chinner055388a2011-01-04 11:35:03 +11001214 /* set the low space thresholds for dynamic preallocation */
1215 xfs_set_low_space_thresholds(mp);
1216
Eric Sandeen0771fb42007-10-12 11:03:40 +10001217 /*
1218 * Set the inode cluster size.
1219 * This may still be overridden by the file system
1220 * block size if it is larger than the chosen cluster size.
1221 */
1222 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1223
1224 /*
1225 * Set inode alignment fields
1226 */
1227 xfs_set_inoalignment(mp);
1228
1229 /*
1230 * Check that the data (and log if separate) are an ok size.
1231 */
Christoph Hellwig42490232008-08-13 16:49:32 +10001232 error = xfs_check_sizes(mp);
Eric Sandeen0771fb42007-10-12 11:03:40 +10001233 if (error)
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001234 goto out_remove_uuid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235
1236 /*
1237 * Initialize realtime fields in the mount structure
1238 */
Eric Sandeen0771fb42007-10-12 11:03:40 +10001239 error = xfs_rtmount_init(mp);
1240 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001241 xfs_warn(mp, "RT mount failed");
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001242 goto out_remove_uuid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 }
1244
1245 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 * Copies the low order bits of the timestamp and the randomly
1247 * set "sequence" number out of a UUID.
1248 */
1249 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1250
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 mp->m_dmevmask = 0; /* not persistent; set after each mount */
1252
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001253 xfs_dir_mount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254
1255 /*
1256 * Initialize the attribute manager's entries.
1257 */
1258 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1259
1260 /*
1261 * Initialize the precomputed transaction reservations values.
1262 */
1263 xfs_trans_init(mp);
1264
1265 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 * Allocate and initialize the per-ag data.
1267 */
Dave Chinner1c1c6eb2010-01-11 11:47:44 +00001268 spin_lock_init(&mp->m_perag_lock);
Dave Chinner9b98b6f2010-05-27 01:58:13 +00001269 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
Dave Chinner1c1c6eb2010-01-11 11:47:44 +00001270 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
1271 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001272 xfs_warn(mp, "Failed per-ag init: %d", error);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001273 goto out_remove_uuid;
Dave Chinner1c1c6eb2010-01-11 11:47:44 +00001274 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001276 if (!sbp->sb_logblocks) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001277 xfs_warn(mp, "no log defined");
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001278 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
1279 error = XFS_ERROR(EFSCORRUPTED);
1280 goto out_free_perag;
1281 }
1282
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 /*
1284 * log's mount-time initialization. Perform 1st part recovery if needed
1285 */
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001286 error = xfs_log_mount(mp, mp->m_logdev_targp,
1287 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1288 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1289 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001290 xfs_warn(mp, "log mount failed");
Dave Chinnerd4f35122012-04-23 15:59:06 +10001291 goto out_fail_wait;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 }
1293
1294 /*
David Chinner92821e22007-05-24 15:26:31 +10001295 * Now the log is mounted, we know if it was an unclean shutdown or
1296 * not. If it was, with the first phase of recovery has completed, we
1297 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1298 * but they are recovered transactionally in the second recovery phase
1299 * later.
1300 *
1301 * Hence we can safely re-initialise incore superblock counters from
1302 * the per-ag data. These may not be correct if the filesystem was not
1303 * cleanly unmounted, so we need to wait for recovery to finish before
1304 * doing this.
1305 *
1306 * If the filesystem was cleanly unmounted, then we can trust the
1307 * values in the superblock to be correct and we don't need to do
1308 * anything here.
1309 *
1310 * If we are currently making the filesystem, the initialisation will
1311 * fail as the perag data is in an undefined state.
1312 */
David Chinner92821e22007-05-24 15:26:31 +10001313 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1314 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1315 !mp->m_sb.sb_inprogress) {
1316 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001317 if (error)
Dave Chinnerd4f35122012-04-23 15:59:06 +10001318 goto out_fail_wait;
David Chinner92821e22007-05-24 15:26:31 +10001319 }
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001320
David Chinner92821e22007-05-24 15:26:31 +10001321 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 * Get and sanity-check the root inode.
1323 * Save the pointer to it in the mount structure.
1324 */
Dave Chinner7b6259e2010-06-24 11:35:17 +10001325 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001327 xfs_warn(mp, "failed to read root inode");
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001328 goto out_log_dealloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 }
1330
1331 ASSERT(rip != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
Al Viroabbede12011-07-26 02:31:30 -04001333 if (unlikely(!S_ISDIR(rip->i_d.di_mode))) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001334 xfs_warn(mp, "corrupted root inode %llu: not a directory",
Nathan Scottb6574522006-06-09 15:29:40 +10001335 (unsigned long long)rip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1337 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1338 mp);
1339 error = XFS_ERROR(EFSCORRUPTED);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001340 goto out_rele_rip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 }
1342 mp->m_rootip = rip; /* save it */
1343
1344 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1345
1346 /*
1347 * Initialize realtime inode pointers in the mount structure
1348 */
Eric Sandeen0771fb42007-10-12 11:03:40 +10001349 error = xfs_rtmount_inodes(mp);
1350 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 /*
1352 * Free up the root inode.
1353 */
Dave Chinner0b932cc2011-03-07 10:08:35 +11001354 xfs_warn(mp, "failed to read RT inodes");
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001355 goto out_rele_rip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 }
1357
1358 /*
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001359 * If this is a read-only mount defer the superblock updates until
1360 * the next remount into writeable mode. Otherwise we would never
1361 * perform the update e.g. for the root filesystem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 */
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001363 if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1364 error = xfs_mount_log_sb(mp, mp->m_update_flags);
David Chinnere5720ee2008-04-10 12:21:18 +10001365 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001366 xfs_warn(mp, "failed to write sb changes");
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001367 goto out_rtunmount;
David Chinnere5720ee2008-04-10 12:21:18 +10001368 }
1369 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370
1371 /*
1372 * Initialise the XFS quota management subsystem for this mount
1373 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001374 if (XFS_IS_QUOTA_RUNNING(mp)) {
1375 error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
1376 if (error)
1377 goto out_rtunmount;
1378 } else {
1379 ASSERT(!XFS_IS_QUOTA_ON(mp));
1380
1381 /*
1382 * If a file system had quotas running earlier, but decided to
1383 * mount without -o uquota/pquota/gquota options, revoke the
1384 * quotachecked license.
1385 */
1386 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001387 xfs_notice(mp, "resetting quota flags");
Christoph Hellwig7d095252009-06-08 15:33:32 +02001388 error = xfs_mount_reset_sbqflags(mp);
1389 if (error)
1390 return error;
1391 }
1392 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393
1394 /*
1395 * Finish recovering the file system. This part needed to be
1396 * delayed until after the root and real-time bitmap inodes
1397 * were consistently read in.
1398 */
Christoph Hellwig42490232008-08-13 16:49:32 +10001399 error = xfs_log_mount_finish(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001401 xfs_warn(mp, "log mount finish failed");
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001402 goto out_rtunmount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 }
1404
1405 /*
1406 * Complete the quota initialisation, post-log-replay component.
1407 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001408 if (quotamount) {
1409 ASSERT(mp->m_qflags == 0);
1410 mp->m_qflags = quotaflags;
1411
1412 xfs_qm_mount_quotas(mp);
1413 }
1414
David Chinner84e1e992007-06-18 16:50:27 +10001415 /*
1416 * Now we are mounted, reserve a small amount of unused space for
1417 * privileged transactions. This is needed so that transaction
1418 * space required for critical operations can dip into this pool
1419 * when at ENOSPC. This is needed for operations like create with
1420 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1421 * are not allowed to use this reserved space.
Dave Chinner8babd8a2010-03-04 01:46:25 +00001422 *
1423 * This may drive us straight to ENOSPC on mount, but that implies
1424 * we were already there on the last unmount. Warn if this occurs.
David Chinner84e1e992007-06-18 16:50:27 +10001425 */
Eric Sandeend5db0f92010-02-05 22:59:53 +00001426 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
1427 resblks = xfs_default_resblks(mp);
1428 error = xfs_reserve_blocks(mp, &resblks, NULL);
1429 if (error)
Dave Chinner0b932cc2011-03-07 10:08:35 +11001430 xfs_warn(mp,
1431 "Unable to allocate reserve blocks. Continuing without reserve pool.");
Eric Sandeend5db0f92010-02-05 22:59:53 +00001432 }
David Chinner84e1e992007-06-18 16:50:27 +10001433
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 return 0;
1435
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001436 out_rtunmount:
1437 xfs_rtunmount_inodes(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001438 out_rele_rip:
Christoph Hellwig43355092008-03-27 18:01:08 +11001439 IRELE(rip);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001440 out_log_dealloc:
Christoph Hellwig21b699c2009-03-16 08:19:29 +01001441 xfs_log_unmount(mp);
Dave Chinnerd4f35122012-04-23 15:59:06 +10001442 out_fail_wait:
1443 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1444 xfs_wait_buftarg(mp->m_logdev_targp);
1445 xfs_wait_buftarg(mp->m_ddev_targp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001446 out_free_perag:
Christoph Hellwigff4f0382008-08-13 16:50:47 +10001447 xfs_free_perag(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001448 out_remove_uuid:
Christoph Hellwig27174202009-03-30 10:21:31 +02001449 xfs_uuid_unmount(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001450 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 return error;
1452}
1453
1454/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 * This flushes out the inodes,dquots and the superblock, unmounts the
1456 * log and makes sure that incore structures are freed.
1457 */
Christoph Hellwig41b5c2e2008-08-13 16:49:57 +10001458void
1459xfs_unmountfs(
1460 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461{
Christoph Hellwig41b5c2e2008-08-13 16:49:57 +10001462 __uint64_t resblks;
1463 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
Christoph Hellwig7d095252009-06-08 15:33:32 +02001465 xfs_qm_unmount_quotas(mp);
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001466 xfs_rtunmount_inodes(mp);
Christoph Hellwig77508ec2008-08-13 16:49:04 +10001467 IRELE(mp->m_rootip);
1468
David Chinner641c56f2007-06-18 16:50:17 +10001469 /*
1470 * We can potentially deadlock here if we have an inode cluster
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001471 * that has been freed has its buffer still pinned in memory because
David Chinner641c56f2007-06-18 16:50:17 +10001472 * the transaction is still sitting in a iclog. The stale inodes
1473 * on that buffer will have their flush locks held until the
1474 * transaction hits the disk and the callbacks run. the inode
1475 * flush takes the flush lock unconditionally and with nothing to
1476 * push out the iclog we will never get that unlocked. hence we
1477 * need to force the log first.
1478 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00001479 xfs_log_force(mp, XFS_LOG_SYNC);
Dave Chinnerc8543632010-02-06 12:39:36 +11001480
1481 /*
Christoph Hellwig211e4d42012-04-23 15:58:34 +10001482 * Flush all pending changes from the AIL.
Dave Chinnerc8543632010-02-06 12:39:36 +11001483 */
Christoph Hellwig211e4d42012-04-23 15:58:34 +10001484 xfs_ail_push_all_sync(mp->m_ail);
1485
1486 /*
1487 * And reclaim all inodes. At this point there should be no dirty
1488 * inode, and none should be pinned or locked, but use synchronous
1489 * reclaim just to be sure.
1490 */
Dave Chinnerc8543632010-02-06 12:39:36 +11001491 xfs_reclaim_inodes(mp, SYNC_WAIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492
Christoph Hellwig7d095252009-06-08 15:33:32 +02001493 xfs_qm_unmount(mp);
Lachlan McIlroya357a122008-10-30 16:53:25 +11001494
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 /*
1496 * Flush out the log synchronously so that we know for sure
1497 * that nothing is pinned. This is important because bflush()
1498 * will skip pinned buffers.
1499 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00001500 xfs_log_force(mp, XFS_LOG_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501
David Chinner84e1e992007-06-18 16:50:27 +10001502 /*
1503 * Unreserve any blocks we have so that when we unmount we don't account
1504 * the reserved free space as used. This is really only necessary for
1505 * lazy superblock counting because it trusts the incore superblock
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001506 * counters to be absolutely correct on clean unmount.
David Chinner84e1e992007-06-18 16:50:27 +10001507 *
1508 * We don't bother correcting this elsewhere for lazy superblock
1509 * counting because on mount of an unclean filesystem we reconstruct the
1510 * correct counter value and this is irrelevant.
1511 *
1512 * For non-lazy counter filesystems, this doesn't matter at all because
1513 * we only every apply deltas to the superblock and hence the incore
1514 * value does not matter....
1515 */
1516 resblks = 0;
David Chinner714082b2008-04-10 12:20:03 +10001517 error = xfs_reserve_blocks(mp, &resblks, NULL);
1518 if (error)
Dave Chinner0b932cc2011-03-07 10:08:35 +11001519 xfs_warn(mp, "Unable to free reserved block pool. "
David Chinner714082b2008-04-10 12:20:03 +10001520 "Freespace may not be correct on next mount.");
1521
Chandra Seetharamanadab0f62011-06-29 22:10:14 +00001522 error = xfs_log_sbcount(mp);
David Chinnere5720ee2008-04-10 12:21:18 +10001523 if (error)
Dave Chinner0b932cc2011-03-07 10:08:35 +11001524 xfs_warn(mp, "Unable to update superblock counters. "
David Chinnere5720ee2008-04-10 12:21:18 +10001525 "Freespace may not be correct on next mount.");
Christoph Hellwig87c7bec2011-09-14 14:08:26 +00001526
1527 /*
Christoph Hellwig211e4d42012-04-23 15:58:34 +10001528 * At this point we might have modified the superblock again and thus
1529 * added an item to the AIL, thus flush it again.
Christoph Hellwig87c7bec2011-09-14 14:08:26 +00001530 */
Christoph Hellwig211e4d42012-04-23 15:58:34 +10001531 xfs_ail_push_all_sync(mp->m_ail);
Christoph Hellwig87c7bec2011-09-14 14:08:26 +00001532 xfs_wait_buftarg(mp->m_ddev_targp);
1533
Christoph Hellwig21b699c2009-03-16 08:19:29 +01001534 xfs_log_unmount_write(mp);
1535 xfs_log_unmount(mp);
Christoph Hellwig27174202009-03-30 10:21:31 +02001536 xfs_uuid_unmount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537
Christoph Hellwig1550d0b2008-08-13 16:17:37 +10001538#if defined(DEBUG)
Christoph Hellwig0ce4cfd2007-08-30 17:20:53 +10001539 xfs_errortag_clearall(mp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540#endif
Christoph Hellwigff4f0382008-08-13 16:50:47 +10001541 xfs_free_perag(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542}
1543
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544int
David Chinner92821e22007-05-24 15:26:31 +10001545xfs_fs_writable(xfs_mount_t *mp)
1546{
Christoph Hellwigb267ce92007-08-30 17:21:30 +10001547 return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001548 (mp->m_flags & XFS_MOUNT_RDONLY));
David Chinner92821e22007-05-24 15:26:31 +10001549}
1550
1551/*
Alex Elderb2ce3972011-07-11 09:51:44 -05001552 * xfs_log_sbcount
1553 *
Chandra Seetharamanadab0f62011-06-29 22:10:14 +00001554 * Sync the superblock counters to disk.
Alex Elderb2ce3972011-07-11 09:51:44 -05001555 *
1556 * Note this code can be called during the process of freezing, so
Chandra Seetharamanadab0f62011-06-29 22:10:14 +00001557 * we may need to use the transaction allocator which does not
Alex Elderb2ce3972011-07-11 09:51:44 -05001558 * block when the transaction subsystem is in its frozen state.
David Chinner92821e22007-05-24 15:26:31 +10001559 */
1560int
Chandra Seetharamanadab0f62011-06-29 22:10:14 +00001561xfs_log_sbcount(xfs_mount_t *mp)
David Chinner92821e22007-05-24 15:26:31 +10001562{
1563 xfs_trans_t *tp;
1564 int error;
1565
1566 if (!xfs_fs_writable(mp))
1567 return 0;
1568
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10001569 xfs_icsb_sync_counters(mp, 0);
David Chinner92821e22007-05-24 15:26:31 +10001570
1571 /*
1572 * we don't need to do this if we are updating the superblock
1573 * counters on every modification.
1574 */
1575 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1576 return 0;
1577
Alex Elderb2ce3972011-07-11 09:51:44 -05001578 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
David Chinner92821e22007-05-24 15:26:31 +10001579 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1580 XFS_DEFAULT_LOG_COUNT);
1581 if (error) {
1582 xfs_trans_cancel(tp, 0);
1583 return error;
1584 }
1585
1586 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
Chandra Seetharamanadab0f62011-06-29 22:10:14 +00001587 xfs_trans_set_sync(tp);
David Chinnere5720ee2008-04-10 12:21:18 +10001588 error = xfs_trans_commit(tp, 0);
1589 return error;
David Chinner92821e22007-05-24 15:26:31 +10001590}
1591
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592/*
1593 * xfs_mod_sb() can be used to copy arbitrary changes to the
1594 * in-core superblock into the superblock buffer to be logged.
1595 * It does not provide the higher level of locking that is
1596 * needed to protect the in-core superblock from concurrent
1597 * access.
1598 */
1599void
1600xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1601{
1602 xfs_buf_t *bp;
1603 int first;
1604 int last;
1605 xfs_mount_t *mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 xfs_sb_field_t f;
1607
1608 ASSERT(fields);
1609 if (!fields)
1610 return;
1611 mp = tp->t_mountp;
1612 bp = xfs_trans_getsb(tp, mp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 first = sizeof(xfs_sb_t);
1614 last = 0;
1615
1616 /* translate/copy */
1617
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +10001618 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619
1620 /* find modified range */
Dave Chinner587aa0f2010-01-20 12:04:53 +11001621 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1622 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1623 last = xfs_sb_info[f + 1].offset - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624
1625 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1626 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1627 first = xfs_sb_info[f].offset;
1628
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 xfs_trans_log_buf(tp, bp, first, last);
1630}
Yingping Lud210a282006-06-09 14:55:18 +10001631
Yingping Lud210a282006-06-09 14:55:18 +10001632
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633/*
1634 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1635 * a delta to a specified field in the in-core superblock. Simply
1636 * switch on the field indicated and apply the delta to that field.
1637 * Fields are not allowed to dip below zero, so if the delta would
1638 * do this do not apply it and return EINVAL.
1639 *
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001640 * The m_sb_lock must be held when this routine is called.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05001642STATIC int
David Chinner20f4ebf2007-02-10 18:36:10 +11001643xfs_mod_incore_sb_unlocked(
1644 xfs_mount_t *mp,
1645 xfs_sb_field_t field,
1646 int64_t delta,
1647 int rsvd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648{
1649 int scounter; /* short counter for 32 bit fields */
1650 long long lcounter; /* long counter for 64 bit fields */
1651 long long res_used, rem;
1652
1653 /*
1654 * With the in-core superblock spin lock held, switch
1655 * on the indicated field. Apply the delta to the
1656 * proper field. If the fields value would dip below
1657 * 0, then do not apply the delta and return EINVAL.
1658 */
1659 switch (field) {
1660 case XFS_SBS_ICOUNT:
1661 lcounter = (long long)mp->m_sb.sb_icount;
1662 lcounter += delta;
1663 if (lcounter < 0) {
1664 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001665 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 }
1667 mp->m_sb.sb_icount = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001668 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 case XFS_SBS_IFREE:
1670 lcounter = (long long)mp->m_sb.sb_ifree;
1671 lcounter += delta;
1672 if (lcounter < 0) {
1673 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001674 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 }
1676 mp->m_sb.sb_ifree = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001677 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 case XFS_SBS_FDBLOCKS:
David Chinner4be536d2006-09-07 14:26:50 +10001679 lcounter = (long long)
1680 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1682
1683 if (delta > 0) { /* Putting blocks back */
1684 if (res_used > delta) {
1685 mp->m_resblks_avail += delta;
1686 } else {
1687 rem = delta - res_used;
1688 mp->m_resblks_avail = mp->m_resblks;
1689 lcounter += rem;
1690 }
1691 } else { /* Taking blocks away */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 lcounter += delta;
Dave Chinner8babd8a2010-03-04 01:46:25 +00001693 if (lcounter >= 0) {
1694 mp->m_sb.sb_fdblocks = lcounter +
1695 XFS_ALLOC_SET_ASIDE(mp);
1696 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 }
Dave Chinner8babd8a2010-03-04 01:46:25 +00001698
1699 /*
1700 * We are out of blocks, use any available reserved
1701 * blocks if were allowed to.
1702 */
1703 if (!rsvd)
1704 return XFS_ERROR(ENOSPC);
1705
1706 lcounter = (long long)mp->m_resblks_avail + delta;
1707 if (lcounter >= 0) {
1708 mp->m_resblks_avail = lcounter;
1709 return 0;
1710 }
1711 printk_once(KERN_WARNING
1712 "Filesystem \"%s\": reserve blocks depleted! "
1713 "Consider increasing reserve pool size.",
1714 mp->m_fsname);
1715 return XFS_ERROR(ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 }
1717
David Chinner4be536d2006-09-07 14:26:50 +10001718 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
Jesper Juhl014c2542006-01-15 02:37:08 +01001719 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 case XFS_SBS_FREXTENTS:
1721 lcounter = (long long)mp->m_sb.sb_frextents;
1722 lcounter += delta;
1723 if (lcounter < 0) {
Jesper Juhl014c2542006-01-15 02:37:08 +01001724 return XFS_ERROR(ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 }
1726 mp->m_sb.sb_frextents = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001727 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 case XFS_SBS_DBLOCKS:
1729 lcounter = (long long)mp->m_sb.sb_dblocks;
1730 lcounter += delta;
1731 if (lcounter < 0) {
1732 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001733 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 }
1735 mp->m_sb.sb_dblocks = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001736 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 case XFS_SBS_AGCOUNT:
1738 scounter = mp->m_sb.sb_agcount;
1739 scounter += delta;
1740 if (scounter < 0) {
1741 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001742 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 }
1744 mp->m_sb.sb_agcount = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001745 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 case XFS_SBS_IMAX_PCT:
1747 scounter = mp->m_sb.sb_imax_pct;
1748 scounter += delta;
1749 if (scounter < 0) {
1750 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001751 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 }
1753 mp->m_sb.sb_imax_pct = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001754 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 case XFS_SBS_REXTSIZE:
1756 scounter = mp->m_sb.sb_rextsize;
1757 scounter += delta;
1758 if (scounter < 0) {
1759 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001760 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 }
1762 mp->m_sb.sb_rextsize = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001763 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 case XFS_SBS_RBMBLOCKS:
1765 scounter = mp->m_sb.sb_rbmblocks;
1766 scounter += delta;
1767 if (scounter < 0) {
1768 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001769 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 }
1771 mp->m_sb.sb_rbmblocks = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001772 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 case XFS_SBS_RBLOCKS:
1774 lcounter = (long long)mp->m_sb.sb_rblocks;
1775 lcounter += delta;
1776 if (lcounter < 0) {
1777 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001778 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 }
1780 mp->m_sb.sb_rblocks = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001781 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 case XFS_SBS_REXTENTS:
1783 lcounter = (long long)mp->m_sb.sb_rextents;
1784 lcounter += delta;
1785 if (lcounter < 0) {
1786 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001787 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 }
1789 mp->m_sb.sb_rextents = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001790 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 case XFS_SBS_REXTSLOG:
1792 scounter = mp->m_sb.sb_rextslog;
1793 scounter += delta;
1794 if (scounter < 0) {
1795 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001796 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 }
1798 mp->m_sb.sb_rextslog = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001799 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 default:
1801 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001802 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 }
1804}
1805
1806/*
1807 * xfs_mod_incore_sb() is used to change a field in the in-core
1808 * superblock structure by the specified delta. This modification
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001809 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 * routine to do the work.
1811 */
1812int
David Chinner20f4ebf2007-02-10 18:36:10 +11001813xfs_mod_incore_sb(
Christoph Hellwig96540c72010-09-30 02:25:55 +00001814 struct xfs_mount *mp,
1815 xfs_sb_field_t field,
1816 int64_t delta,
1817 int rsvd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818{
Christoph Hellwig96540c72010-09-30 02:25:55 +00001819 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820
David Chinner8d280b92006-03-14 13:13:09 +11001821#ifdef HAVE_PERCPU_SB
Christoph Hellwig96540c72010-09-30 02:25:55 +00001822 ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS);
David Chinner8d280b92006-03-14 13:13:09 +11001823#endif
Christoph Hellwig96540c72010-09-30 02:25:55 +00001824 spin_lock(&mp->m_sb_lock);
1825 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1826 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001827
Jesper Juhl014c2542006-01-15 02:37:08 +01001828 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829}
1830
1831/*
Christoph Hellwig1b040712010-09-30 02:25:56 +00001832 * Change more than one field in the in-core superblock structure at a time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 *
Christoph Hellwig1b040712010-09-30 02:25:56 +00001834 * The fields and changes to those fields are specified in the array of
1835 * xfs_mod_sb structures passed in. Either all of the specified deltas
1836 * will be applied or none of them will. If any modified field dips below 0,
1837 * then all modifications will be backed out and EINVAL will be returned.
1838 *
1839 * Note that this function may not be used for the superblock values that
1840 * are tracked with the in-memory per-cpu counters - a direct call to
1841 * xfs_icsb_modify_counters is required for these.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 */
1843int
Christoph Hellwig1b040712010-09-30 02:25:56 +00001844xfs_mod_incore_sb_batch(
1845 struct xfs_mount *mp,
1846 xfs_mod_sb_t *msb,
1847 uint nmsb,
1848 int rsvd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849{
David Sterba45c51b92011-04-13 22:03:28 +00001850 xfs_mod_sb_t *msbp;
Christoph Hellwig1b040712010-09-30 02:25:56 +00001851 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852
1853 /*
Christoph Hellwig1b040712010-09-30 02:25:56 +00001854 * Loop through the array of mod structures and apply each individually.
1855 * If any fail, then back out all those which have already been applied.
1856 * Do all of this within the scope of the m_sb_lock so that all of the
1857 * changes will be atomic.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001859 spin_lock(&mp->m_sb_lock);
David Sterba45c51b92011-04-13 22:03:28 +00001860 for (msbp = msb; msbp < (msb + nmsb); msbp++) {
Christoph Hellwig1b040712010-09-30 02:25:56 +00001861 ASSERT(msbp->msb_field < XFS_SBS_ICOUNT ||
1862 msbp->msb_field > XFS_SBS_FDBLOCKS);
David Chinner8d280b92006-03-14 13:13:09 +11001863
Christoph Hellwig1b040712010-09-30 02:25:56 +00001864 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1865 msbp->msb_delta, rsvd);
1866 if (error)
1867 goto unwind;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 }
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001869 spin_unlock(&mp->m_sb_lock);
Christoph Hellwig1b040712010-09-30 02:25:56 +00001870 return 0;
1871
1872unwind:
1873 while (--msbp >= msb) {
1874 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1875 -msbp->msb_delta, rsvd);
1876 ASSERT(error == 0);
1877 }
1878 spin_unlock(&mp->m_sb_lock);
1879 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880}
1881
1882/*
1883 * xfs_getsb() is called to obtain the buffer for the superblock.
1884 * The buffer is returned locked and read in from disk.
1885 * The buffer should be released with a call to xfs_brelse().
1886 *
1887 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1888 * the superblock buffer if it can be locked without sleeping.
1889 * If it can't then we'll return NULL.
1890 */
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001891struct xfs_buf *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892xfs_getsb(
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001893 struct xfs_mount *mp,
1894 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895{
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001896 struct xfs_buf *bp = mp->m_sb_bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001898 if (!xfs_buf_trylock(bp)) {
1899 if (flags & XBF_TRYLOCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 return NULL;
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001901 xfs_buf_lock(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 }
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001903
Chandra Seetharaman72790aa2011-07-22 23:40:04 +00001904 xfs_buf_hold(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 ASSERT(XFS_BUF_ISDONE(bp));
Jesper Juhl014c2542006-01-15 02:37:08 +01001906 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907}
1908
1909/*
1910 * Used to free the superblock along various error paths.
1911 */
1912void
1913xfs_freesb(
Dave Chinner26af6552010-09-22 10:47:20 +10001914 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915{
Dave Chinner26af6552010-09-22 10:47:20 +10001916 struct xfs_buf *bp = mp->m_sb_bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917
Dave Chinner26af6552010-09-22 10:47:20 +10001918 xfs_buf_lock(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 mp->m_sb_bp = NULL;
Dave Chinner26af6552010-09-22 10:47:20 +10001920 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921}
1922
1923/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 * Used to log changes to the superblock unit and width fields which could
Eric Sandeene6957ea2008-04-10 12:19:34 +10001925 * be altered by the mount options, as well as any potential sb_features2
1926 * fixup. Only the first superblock is updated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 */
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001928int
David Chinneree1c0902008-03-06 13:45:50 +11001929xfs_mount_log_sb(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 xfs_mount_t *mp,
1931 __int64_t fields)
1932{
1933 xfs_trans_t *tp;
David Chinnere5720ee2008-04-10 12:21:18 +10001934 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935
David Chinneree1c0902008-03-06 13:45:50 +11001936 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
David Chinner4b166de2008-05-20 11:30:27 +10001937 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1938 XFS_SB_VERSIONNUM));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939
1940 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
David Chinnere5720ee2008-04-10 12:21:18 +10001941 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1942 XFS_DEFAULT_LOG_COUNT);
1943 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 xfs_trans_cancel(tp, 0);
David Chinnere5720ee2008-04-10 12:21:18 +10001945 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 }
1947 xfs_mod_sb(tp, fields);
David Chinnere5720ee2008-04-10 12:21:18 +10001948 error = xfs_trans_commit(tp, 0);
1949 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950}
David Chinner8d280b92006-03-14 13:13:09 +11001951
Christoph Hellwigdda35b82010-02-15 09:44:46 +00001952/*
1953 * If the underlying (data/log/rt) device is readonly, there are some
1954 * operations that cannot proceed.
1955 */
1956int
1957xfs_dev_is_read_only(
1958 struct xfs_mount *mp,
1959 char *message)
1960{
1961 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1962 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1963 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001964 xfs_notice(mp, "%s required on read-only device.", message);
1965 xfs_notice(mp, "write access unavailable, cannot proceed.");
Christoph Hellwigdda35b82010-02-15 09:44:46 +00001966 return EROFS;
1967 }
1968 return 0;
1969}
David Chinner8d280b92006-03-14 13:13:09 +11001970
1971#ifdef HAVE_PERCPU_SB
1972/*
1973 * Per-cpu incore superblock counters
1974 *
1975 * Simple concept, difficult implementation
1976 *
1977 * Basically, replace the incore superblock counters with a distributed per cpu
1978 * counter for contended fields (e.g. free block count).
1979 *
1980 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1981 * hence needs to be accurately read when we are running low on space. Hence
1982 * there is a method to enable and disable the per-cpu counters based on how
1983 * much "stuff" is available in them.
1984 *
1985 * Basically, a counter is enabled if there is enough free resource to justify
1986 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1987 * ENOSPC), then we disable the counters to synchronise all callers and
1988 * re-distribute the available resources.
1989 *
1990 * If, once we redistributed the available resources, we still get a failure,
1991 * we disable the per-cpu counter and go through the slow path.
1992 *
1993 * The slow path is the current xfs_mod_incore_sb() function. This means that
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001994 * when we disable a per-cpu counter, we need to drain its resources back to
David Chinner8d280b92006-03-14 13:13:09 +11001995 * the global superblock. We do this after disabling the counter to prevent
1996 * more threads from queueing up on the counter.
1997 *
1998 * Essentially, this means that we still need a lock in the fast path to enable
1999 * synchronisation between the global counters and the per-cpu counters. This
2000 * is not a problem because the lock will be local to a CPU almost all the time
2001 * and have little contention except when we get to ENOSPC conditions.
2002 *
2003 * Basically, this lock becomes a barrier that enables us to lock out the fast
2004 * path while we do things like enabling and disabling counters and
2005 * synchronising the counters.
2006 *
2007 * Locking rules:
2008 *
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002009 * 1. m_sb_lock before picking up per-cpu locks
David Chinner8d280b92006-03-14 13:13:09 +11002010 * 2. per-cpu locks always picked up via for_each_online_cpu() order
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002011 * 3. accurate counter sync requires m_sb_lock + per cpu locks
David Chinner8d280b92006-03-14 13:13:09 +11002012 * 4. modifying per-cpu counters requires holding per-cpu lock
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002013 * 5. modifying global counters requires holding m_sb_lock
2014 * 6. enabling or disabling a counter requires holding the m_sb_lock
David Chinner8d280b92006-03-14 13:13:09 +11002015 * and _none_ of the per-cpu locks.
2016 *
2017 * Disabled counters are only ever re-enabled by a balance operation
2018 * that results in more free resources per CPU than a given threshold.
2019 * To ensure counters don't remain disabled, they are rebalanced when
2020 * the global resource goes above a higher threshold (i.e. some hysteresis
2021 * is present to prevent thrashing).
David Chinner8d280b92006-03-14 13:13:09 +11002022 */
David Chinnere8234a62006-03-14 13:23:52 +11002023
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002024#ifdef CONFIG_HOTPLUG_CPU
David Chinnere8234a62006-03-14 13:23:52 +11002025/*
2026 * hot-plug CPU notifier support.
2027 *
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002028 * We need a notifier per filesystem as we need to be able to identify
2029 * the filesystem to balance the counters out. This is achieved by
2030 * having a notifier block embedded in the xfs_mount_t and doing pointer
2031 * magic to get the mount pointer from the notifier block address.
David Chinnere8234a62006-03-14 13:23:52 +11002032 */
2033STATIC int
2034xfs_icsb_cpu_notify(
2035 struct notifier_block *nfb,
2036 unsigned long action,
2037 void *hcpu)
2038{
2039 xfs_icsb_cnts_t *cntp;
2040 xfs_mount_t *mp;
David Chinnere8234a62006-03-14 13:23:52 +11002041
2042 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
2043 cntp = (xfs_icsb_cnts_t *)
2044 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
2045 switch (action) {
2046 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002047 case CPU_UP_PREPARE_FROZEN:
David Chinnere8234a62006-03-14 13:23:52 +11002048 /* Easy Case - initialize the area and locks, and
2049 * then rebalance when online does everything else for us. */
David Chinner01e1b692006-03-14 13:29:16 +11002050 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinnere8234a62006-03-14 13:23:52 +11002051 break;
2052 case CPU_ONLINE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002053 case CPU_ONLINE_FROZEN:
David Chinner03135cf2007-02-10 18:35:15 +11002054 xfs_icsb_lock(mp);
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002055 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2056 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2057 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
David Chinner03135cf2007-02-10 18:35:15 +11002058 xfs_icsb_unlock(mp);
David Chinnere8234a62006-03-14 13:23:52 +11002059 break;
2060 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002061 case CPU_DEAD_FROZEN:
David Chinnere8234a62006-03-14 13:23:52 +11002062 /* Disable all the counters, then fold the dead cpu's
2063 * count into the total on the global superblock and
2064 * re-enable the counters. */
David Chinner03135cf2007-02-10 18:35:15 +11002065 xfs_icsb_lock(mp);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002066 spin_lock(&mp->m_sb_lock);
David Chinnere8234a62006-03-14 13:23:52 +11002067 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
2068 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
2069 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
2070
2071 mp->m_sb.sb_icount += cntp->icsb_icount;
2072 mp->m_sb.sb_ifree += cntp->icsb_ifree;
2073 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
2074
David Chinner01e1b692006-03-14 13:29:16 +11002075 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinnere8234a62006-03-14 13:23:52 +11002076
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002077 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
2078 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
2079 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002080 spin_unlock(&mp->m_sb_lock);
David Chinner03135cf2007-02-10 18:35:15 +11002081 xfs_icsb_unlock(mp);
David Chinnere8234a62006-03-14 13:23:52 +11002082 break;
2083 }
2084
2085 return NOTIFY_OK;
2086}
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002087#endif /* CONFIG_HOTPLUG_CPU */
David Chinnere8234a62006-03-14 13:23:52 +11002088
David Chinner8d280b92006-03-14 13:13:09 +11002089int
2090xfs_icsb_init_counters(
2091 xfs_mount_t *mp)
2092{
2093 xfs_icsb_cnts_t *cntp;
2094 int i;
2095
2096 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2097 if (mp->m_sb_cnts == NULL)
2098 return -ENOMEM;
2099
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002100#ifdef CONFIG_HOTPLUG_CPU
David Chinnere8234a62006-03-14 13:23:52 +11002101 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2102 mp->m_icsb_notifier.priority = 0;
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002103 register_hotcpu_notifier(&mp->m_icsb_notifier);
2104#endif /* CONFIG_HOTPLUG_CPU */
David Chinnere8234a62006-03-14 13:23:52 +11002105
David Chinner8d280b92006-03-14 13:13:09 +11002106 for_each_online_cpu(i) {
2107 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11002108 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinner8d280b92006-03-14 13:13:09 +11002109 }
David Chinner20b64282007-02-10 18:35:09 +11002110
2111 mutex_init(&mp->m_icsb_mutex);
2112
David Chinner8d280b92006-03-14 13:13:09 +11002113 /*
2114 * start with all counters disabled so that the
2115 * initial balance kicks us off correctly
2116 */
2117 mp->m_icsb_counters = -1;
2118 return 0;
2119}
2120
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11002121void
2122xfs_icsb_reinit_counters(
2123 xfs_mount_t *mp)
2124{
2125 xfs_icsb_lock(mp);
2126 /*
2127 * start with all counters disabled so that the
2128 * initial balance kicks us off correctly
2129 */
2130 mp->m_icsb_counters = -1;
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002131 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2132 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2133 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11002134 xfs_icsb_unlock(mp);
2135}
2136
Christoph Hellwigc962fb72008-05-20 15:10:52 +10002137void
David Chinner8d280b92006-03-14 13:13:09 +11002138xfs_icsb_destroy_counters(
2139 xfs_mount_t *mp)
2140{
David Chinnere8234a62006-03-14 13:23:52 +11002141 if (mp->m_sb_cnts) {
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002142 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
David Chinner8d280b92006-03-14 13:13:09 +11002143 free_percpu(mp->m_sb_cnts);
David Chinnere8234a62006-03-14 13:23:52 +11002144 }
David Chinner03135cf2007-02-10 18:35:15 +11002145 mutex_destroy(&mp->m_icsb_mutex);
David Chinner8d280b92006-03-14 13:13:09 +11002146}
2147
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00002148STATIC void
David Chinner01e1b692006-03-14 13:29:16 +11002149xfs_icsb_lock_cntr(
2150 xfs_icsb_cnts_t *icsbp)
2151{
2152 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2153 ndelay(1000);
2154 }
2155}
2156
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00002157STATIC void
David Chinner01e1b692006-03-14 13:29:16 +11002158xfs_icsb_unlock_cntr(
2159 xfs_icsb_cnts_t *icsbp)
2160{
2161 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2162}
2163
David Chinner8d280b92006-03-14 13:13:09 +11002164
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00002165STATIC void
David Chinner8d280b92006-03-14 13:13:09 +11002166xfs_icsb_lock_all_counters(
2167 xfs_mount_t *mp)
2168{
2169 xfs_icsb_cnts_t *cntp;
2170 int i;
2171
2172 for_each_online_cpu(i) {
2173 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11002174 xfs_icsb_lock_cntr(cntp);
David Chinner8d280b92006-03-14 13:13:09 +11002175 }
2176}
2177
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00002178STATIC void
David Chinner8d280b92006-03-14 13:13:09 +11002179xfs_icsb_unlock_all_counters(
2180 xfs_mount_t *mp)
2181{
2182 xfs_icsb_cnts_t *cntp;
2183 int i;
2184
2185 for_each_online_cpu(i) {
2186 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11002187 xfs_icsb_unlock_cntr(cntp);
David Chinner8d280b92006-03-14 13:13:09 +11002188 }
2189}
2190
2191STATIC void
2192xfs_icsb_count(
2193 xfs_mount_t *mp,
2194 xfs_icsb_cnts_t *cnt,
2195 int flags)
2196{
2197 xfs_icsb_cnts_t *cntp;
2198 int i;
2199
2200 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2201
2202 if (!(flags & XFS_ICSB_LAZY_COUNT))
2203 xfs_icsb_lock_all_counters(mp);
2204
2205 for_each_online_cpu(i) {
2206 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2207 cnt->icsb_icount += cntp->icsb_icount;
2208 cnt->icsb_ifree += cntp->icsb_ifree;
2209 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2210 }
2211
2212 if (!(flags & XFS_ICSB_LAZY_COUNT))
2213 xfs_icsb_unlock_all_counters(mp);
2214}
2215
2216STATIC int
2217xfs_icsb_counter_disabled(
2218 xfs_mount_t *mp,
2219 xfs_sb_field_t field)
2220{
2221 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2222 return test_bit(field, &mp->m_icsb_counters);
2223}
2224
David Chinner36fbe6e2008-04-10 12:19:56 +10002225STATIC void
David Chinner8d280b92006-03-14 13:13:09 +11002226xfs_icsb_disable_counter(
2227 xfs_mount_t *mp,
2228 xfs_sb_field_t field)
2229{
2230 xfs_icsb_cnts_t cnt;
2231
2232 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2233
David Chinner20b64282007-02-10 18:35:09 +11002234 /*
2235 * If we are already disabled, then there is nothing to do
2236 * here. We check before locking all the counters to avoid
2237 * the expensive lock operation when being called in the
2238 * slow path and the counter is already disabled. This is
2239 * safe because the only time we set or clear this state is under
2240 * the m_icsb_mutex.
2241 */
2242 if (xfs_icsb_counter_disabled(mp, field))
David Chinner36fbe6e2008-04-10 12:19:56 +10002243 return;
David Chinner20b64282007-02-10 18:35:09 +11002244
David Chinner8d280b92006-03-14 13:13:09 +11002245 xfs_icsb_lock_all_counters(mp);
2246 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2247 /* drain back to superblock */
2248
Christoph Hellwigce461932008-04-22 17:34:50 +10002249 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
David Chinner8d280b92006-03-14 13:13:09 +11002250 switch(field) {
2251 case XFS_SBS_ICOUNT:
2252 mp->m_sb.sb_icount = cnt.icsb_icount;
2253 break;
2254 case XFS_SBS_IFREE:
2255 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2256 break;
2257 case XFS_SBS_FDBLOCKS:
2258 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2259 break;
2260 default:
2261 BUG();
2262 }
2263 }
2264
2265 xfs_icsb_unlock_all_counters(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002266}
2267
2268STATIC void
2269xfs_icsb_enable_counter(
2270 xfs_mount_t *mp,
2271 xfs_sb_field_t field,
2272 uint64_t count,
2273 uint64_t resid)
2274{
2275 xfs_icsb_cnts_t *cntp;
2276 int i;
2277
2278 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2279
2280 xfs_icsb_lock_all_counters(mp);
2281 for_each_online_cpu(i) {
2282 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2283 switch (field) {
2284 case XFS_SBS_ICOUNT:
2285 cntp->icsb_icount = count + resid;
2286 break;
2287 case XFS_SBS_IFREE:
2288 cntp->icsb_ifree = count + resid;
2289 break;
2290 case XFS_SBS_FDBLOCKS:
2291 cntp->icsb_fdblocks = count + resid;
2292 break;
2293 default:
2294 BUG();
2295 break;
2296 }
2297 resid = 0;
2298 }
2299 clear_bit(field, &mp->m_icsb_counters);
2300 xfs_icsb_unlock_all_counters(mp);
2301}
2302
David Chinnerdbcabad2007-02-10 18:36:17 +11002303void
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002304xfs_icsb_sync_counters_locked(
David Chinner8d280b92006-03-14 13:13:09 +11002305 xfs_mount_t *mp,
2306 int flags)
2307{
2308 xfs_icsb_cnts_t cnt;
David Chinner8d280b92006-03-14 13:13:09 +11002309
David Chinner8d280b92006-03-14 13:13:09 +11002310 xfs_icsb_count(mp, &cnt, flags);
2311
David Chinner8d280b92006-03-14 13:13:09 +11002312 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2313 mp->m_sb.sb_icount = cnt.icsb_icount;
2314 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2315 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2316 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2317 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
David Chinner8d280b92006-03-14 13:13:09 +11002318}
2319
2320/*
2321 * Accurate update of per-cpu counters to incore superblock
2322 */
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002323void
David Chinner8d280b92006-03-14 13:13:09 +11002324xfs_icsb_sync_counters(
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002325 xfs_mount_t *mp,
2326 int flags)
David Chinner8d280b92006-03-14 13:13:09 +11002327{
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002328 spin_lock(&mp->m_sb_lock);
2329 xfs_icsb_sync_counters_locked(mp, flags);
2330 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11002331}
2332
2333/*
2334 * Balance and enable/disable counters as necessary.
2335 *
David Chinner20b64282007-02-10 18:35:09 +11002336 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2337 * chosen to be the same number as single on disk allocation chunk per CPU, and
2338 * free blocks is something far enough zero that we aren't going thrash when we
2339 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2340 * prevent looping endlessly when xfs_alloc_space asks for more than will
2341 * be distributed to a single CPU but each CPU has enough blocks to be
2342 * reenabled.
2343 *
2344 * Note that we can be called when counters are already disabled.
2345 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2346 * prevent locking every per-cpu counter needlessly.
David Chinner8d280b92006-03-14 13:13:09 +11002347 */
David Chinner20b64282007-02-10 18:35:09 +11002348
2349#define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
David Chinner4be536d2006-09-07 14:26:50 +10002350#define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
David Chinner20b64282007-02-10 18:35:09 +11002351 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
David Chinner8d280b92006-03-14 13:13:09 +11002352STATIC void
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002353xfs_icsb_balance_counter_locked(
David Chinner8d280b92006-03-14 13:13:09 +11002354 xfs_mount_t *mp,
2355 xfs_sb_field_t field,
David Chinner20b64282007-02-10 18:35:09 +11002356 int min_per_cpu)
David Chinner8d280b92006-03-14 13:13:09 +11002357{
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002358 uint64_t count, resid;
David Chinner8d280b92006-03-14 13:13:09 +11002359 int weight = num_online_cpus();
David Chinner20b64282007-02-10 18:35:09 +11002360 uint64_t min = (uint64_t)min_per_cpu;
David Chinner8d280b92006-03-14 13:13:09 +11002361
David Chinner8d280b92006-03-14 13:13:09 +11002362 /* disable counter and sync counter */
2363 xfs_icsb_disable_counter(mp, field);
2364
2365 /* update counters - first CPU gets residual*/
2366 switch (field) {
2367 case XFS_SBS_ICOUNT:
2368 count = mp->m_sb.sb_icount;
2369 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002370 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002371 return;
David Chinner8d280b92006-03-14 13:13:09 +11002372 break;
2373 case XFS_SBS_IFREE:
2374 count = mp->m_sb.sb_ifree;
2375 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002376 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002377 return;
David Chinner8d280b92006-03-14 13:13:09 +11002378 break;
2379 case XFS_SBS_FDBLOCKS:
2380 count = mp->m_sb.sb_fdblocks;
2381 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002382 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002383 return;
David Chinner8d280b92006-03-14 13:13:09 +11002384 break;
2385 default:
2386 BUG();
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002387 count = resid = 0; /* quiet, gcc */
David Chinner8d280b92006-03-14 13:13:09 +11002388 break;
2389 }
2390
2391 xfs_icsb_enable_counter(mp, field, count, resid);
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002392}
2393
2394STATIC void
2395xfs_icsb_balance_counter(
2396 xfs_mount_t *mp,
2397 xfs_sb_field_t fields,
2398 int min_per_cpu)
2399{
2400 spin_lock(&mp->m_sb_lock);
2401 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2402 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11002403}
2404
Christoph Hellwig1b040712010-09-30 02:25:56 +00002405int
David Chinner20b64282007-02-10 18:35:09 +11002406xfs_icsb_modify_counters(
David Chinner8d280b92006-03-14 13:13:09 +11002407 xfs_mount_t *mp,
2408 xfs_sb_field_t field,
David Chinner20f4ebf2007-02-10 18:36:10 +11002409 int64_t delta,
David Chinner20b64282007-02-10 18:35:09 +11002410 int rsvd)
David Chinner8d280b92006-03-14 13:13:09 +11002411{
2412 xfs_icsb_cnts_t *icsbp;
2413 long long lcounter; /* long counter for 64 bit fields */
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002414 int ret = 0;
David Chinner8d280b92006-03-14 13:13:09 +11002415
David Chinner20b64282007-02-10 18:35:09 +11002416 might_sleep();
David Chinner8d280b92006-03-14 13:13:09 +11002417again:
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002418 preempt_disable();
2419 icsbp = this_cpu_ptr(mp->m_sb_cnts);
David Chinner20b64282007-02-10 18:35:09 +11002420
2421 /*
2422 * if the counter is disabled, go to slow path
2423 */
David Chinner8d280b92006-03-14 13:13:09 +11002424 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2425 goto slow_path;
David Chinner20b64282007-02-10 18:35:09 +11002426 xfs_icsb_lock_cntr(icsbp);
2427 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2428 xfs_icsb_unlock_cntr(icsbp);
2429 goto slow_path;
2430 }
David Chinner8d280b92006-03-14 13:13:09 +11002431
2432 switch (field) {
2433 case XFS_SBS_ICOUNT:
2434 lcounter = icsbp->icsb_icount;
2435 lcounter += delta;
2436 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002437 goto balance_counter;
David Chinner8d280b92006-03-14 13:13:09 +11002438 icsbp->icsb_icount = lcounter;
2439 break;
2440
2441 case XFS_SBS_IFREE:
2442 lcounter = icsbp->icsb_ifree;
2443 lcounter += delta;
2444 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002445 goto balance_counter;
David Chinner8d280b92006-03-14 13:13:09 +11002446 icsbp->icsb_ifree = lcounter;
2447 break;
2448
2449 case XFS_SBS_FDBLOCKS:
2450 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2451
David Chinner4be536d2006-09-07 14:26:50 +10002452 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002453 lcounter += delta;
2454 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002455 goto balance_counter;
David Chinner4be536d2006-09-07 14:26:50 +10002456 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002457 break;
2458 default:
2459 BUG();
2460 break;
2461 }
David Chinner01e1b692006-03-14 13:29:16 +11002462 xfs_icsb_unlock_cntr(icsbp);
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002463 preempt_enable();
David Chinner8d280b92006-03-14 13:13:09 +11002464 return 0;
2465
David Chinner8d280b92006-03-14 13:13:09 +11002466slow_path:
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002467 preempt_enable();
David Chinner20b64282007-02-10 18:35:09 +11002468
2469 /*
2470 * serialise with a mutex so we don't burn lots of cpu on
2471 * the superblock lock. We still need to hold the superblock
2472 * lock, however, when we modify the global structures.
2473 */
David Chinner03135cf2007-02-10 18:35:15 +11002474 xfs_icsb_lock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002475
2476 /*
2477 * Now running atomically.
2478 *
2479 * If the counter is enabled, someone has beaten us to rebalancing.
2480 * Drop the lock and try again in the fast path....
2481 */
2482 if (!(xfs_icsb_counter_disabled(mp, field))) {
David Chinner03135cf2007-02-10 18:35:15 +11002483 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002484 goto again;
2485 }
2486
2487 /*
2488 * The counter is currently disabled. Because we are
2489 * running atomically here, we know a rebalance cannot
2490 * be in progress. Hence we can go straight to operating
2491 * on the global superblock. We do not call xfs_mod_incore_sb()
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002492 * here even though we need to get the m_sb_lock. Doing so
David Chinner20b64282007-02-10 18:35:09 +11002493 * will cause us to re-enter this function and deadlock.
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002494 * Hence we get the m_sb_lock ourselves and then call
David Chinner20b64282007-02-10 18:35:09 +11002495 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2496 * directly on the global counters.
2497 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002498 spin_lock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11002499 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002500 spin_unlock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11002501
2502 /*
2503 * Now that we've modified the global superblock, we
2504 * may be able to re-enable the distributed counters
2505 * (e.g. lots of space just got freed). After that
2506 * we are done.
2507 */
2508 if (ret != ENOSPC)
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002509 xfs_icsb_balance_counter(mp, field, 0);
David Chinner03135cf2007-02-10 18:35:15 +11002510 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002511 return ret;
2512
2513balance_counter:
David Chinner01e1b692006-03-14 13:29:16 +11002514 xfs_icsb_unlock_cntr(icsbp);
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002515 preempt_enable();
David Chinner8d280b92006-03-14 13:13:09 +11002516
David Chinner20b64282007-02-10 18:35:09 +11002517 /*
2518 * We may have multiple threads here if multiple per-cpu
2519 * counters run dry at the same time. This will mean we can
2520 * do more balances than strictly necessary but it is not
2521 * the common slowpath case.
2522 */
David Chinner03135cf2007-02-10 18:35:15 +11002523 xfs_icsb_lock(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002524
David Chinner20b64282007-02-10 18:35:09 +11002525 /*
2526 * running atomically.
2527 *
2528 * This will leave the counter in the correct state for future
2529 * accesses. After the rebalance, we simply try again and our retry
2530 * will either succeed through the fast path or slow path without
2531 * another balance operation being required.
2532 */
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002533 xfs_icsb_balance_counter(mp, field, delta);
David Chinner03135cf2007-02-10 18:35:15 +11002534 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002535 goto again;
David Chinner8d280b92006-03-14 13:13:09 +11002536}
2537
David Chinner8d280b92006-03-14 13:13:09 +11002538#endif