blob: 0081657ad985ee27417ba4b73c142cb3d43e78b7 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_dir2.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110030#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_dinode.h"
33#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110034#include "xfs_btree.h"
35#include "xfs_ialloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_alloc.h"
37#include "xfs_rtalloc.h"
38#include "xfs_bmap.h"
39#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include "xfs_rw.h"
41#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
Christoph Hellwigba0f32d2005-06-21 15:36:52 +100047STATIC void xfs_unmountfs_wait(xfs_mount_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
David Chinner8d280b92006-03-14 13:13:09 +110049
50#ifdef HAVE_PERCPU_SB
David Chinner20f4ebf2007-02-10 18:36:10 +110051STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
Christoph Hellwig45af6c62008-04-22 17:34:44 +100052 int);
53STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t,
54 int);
David Chinner36fbe6e2008-04-10 12:19:56 +100055STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
David Chinner8d280b92006-03-14 13:13:09 +110056#else
57
Christoph Hellwig45af6c62008-04-22 17:34:44 +100058#define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
59#define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
David Chinner8d280b92006-03-14 13:13:09 +110060#endif
61
Christoph Hellwig1df84c92006-01-11 15:29:52 +110062static const struct {
David Chinner8d280b92006-03-14 13:13:09 +110063 short offset;
64 short type; /* 0 = integer
65 * 1 = binary / string (no translation)
66 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070067} xfs_sb_info[] = {
68 { offsetof(xfs_sb_t, sb_magicnum), 0 },
69 { offsetof(xfs_sb_t, sb_blocksize), 0 },
70 { offsetof(xfs_sb_t, sb_dblocks), 0 },
71 { offsetof(xfs_sb_t, sb_rblocks), 0 },
72 { offsetof(xfs_sb_t, sb_rextents), 0 },
73 { offsetof(xfs_sb_t, sb_uuid), 1 },
74 { offsetof(xfs_sb_t, sb_logstart), 0 },
75 { offsetof(xfs_sb_t, sb_rootino), 0 },
76 { offsetof(xfs_sb_t, sb_rbmino), 0 },
77 { offsetof(xfs_sb_t, sb_rsumino), 0 },
78 { offsetof(xfs_sb_t, sb_rextsize), 0 },
79 { offsetof(xfs_sb_t, sb_agblocks), 0 },
80 { offsetof(xfs_sb_t, sb_agcount), 0 },
81 { offsetof(xfs_sb_t, sb_rbmblocks), 0 },
82 { offsetof(xfs_sb_t, sb_logblocks), 0 },
83 { offsetof(xfs_sb_t, sb_versionnum), 0 },
84 { offsetof(xfs_sb_t, sb_sectsize), 0 },
85 { offsetof(xfs_sb_t, sb_inodesize), 0 },
86 { offsetof(xfs_sb_t, sb_inopblock), 0 },
87 { offsetof(xfs_sb_t, sb_fname[0]), 1 },
88 { offsetof(xfs_sb_t, sb_blocklog), 0 },
89 { offsetof(xfs_sb_t, sb_sectlog), 0 },
90 { offsetof(xfs_sb_t, sb_inodelog), 0 },
91 { offsetof(xfs_sb_t, sb_inopblog), 0 },
92 { offsetof(xfs_sb_t, sb_agblklog), 0 },
93 { offsetof(xfs_sb_t, sb_rextslog), 0 },
94 { offsetof(xfs_sb_t, sb_inprogress), 0 },
95 { offsetof(xfs_sb_t, sb_imax_pct), 0 },
96 { offsetof(xfs_sb_t, sb_icount), 0 },
97 { offsetof(xfs_sb_t, sb_ifree), 0 },
98 { offsetof(xfs_sb_t, sb_fdblocks), 0 },
99 { offsetof(xfs_sb_t, sb_frextents), 0 },
100 { offsetof(xfs_sb_t, sb_uquotino), 0 },
101 { offsetof(xfs_sb_t, sb_gquotino), 0 },
102 { offsetof(xfs_sb_t, sb_qflags), 0 },
103 { offsetof(xfs_sb_t, sb_flags), 0 },
104 { offsetof(xfs_sb_t, sb_shared_vn), 0 },
105 { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
106 { offsetof(xfs_sb_t, sb_unit), 0 },
107 { offsetof(xfs_sb_t, sb_width), 0 },
108 { offsetof(xfs_sb_t, sb_dirblklog), 0 },
109 { offsetof(xfs_sb_t, sb_logsectlog), 0 },
110 { offsetof(xfs_sb_t, sb_logsectsize),0 },
111 { offsetof(xfs_sb_t, sb_logsunit), 0 },
112 { offsetof(xfs_sb_t, sb_features2), 0 },
David Chinneree1c0902008-03-06 13:45:50 +1100113 { offsetof(xfs_sb_t, sb_bad_features2), 0 },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 { sizeof(xfs_sb_t), 0 }
115};
116
Christoph Hellwig27174202009-03-30 10:21:31 +0200117static DEFINE_MUTEX(xfs_uuid_table_mutex);
118static int xfs_uuid_table_size;
119static uuid_t *xfs_uuid_table;
120
121/*
122 * See if the UUID is unique among mounted XFS filesystems.
123 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
124 */
125STATIC int
126xfs_uuid_mount(
127 struct xfs_mount *mp)
128{
129 uuid_t *uuid = &mp->m_sb.sb_uuid;
130 int hole, i;
131
132 if (mp->m_flags & XFS_MOUNT_NOUUID)
133 return 0;
134
135 if (uuid_is_nil(uuid)) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100136 xfs_warn(mp, "Filesystem has nil UUID - can't mount");
Christoph Hellwig27174202009-03-30 10:21:31 +0200137 return XFS_ERROR(EINVAL);
138 }
139
140 mutex_lock(&xfs_uuid_table_mutex);
141 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
142 if (uuid_is_nil(&xfs_uuid_table[i])) {
143 hole = i;
144 continue;
145 }
146 if (uuid_equal(uuid, &xfs_uuid_table[i]))
147 goto out_duplicate;
148 }
149
150 if (hole < 0) {
151 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
152 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
153 xfs_uuid_table_size * sizeof(*xfs_uuid_table),
154 KM_SLEEP);
155 hole = xfs_uuid_table_size++;
156 }
157 xfs_uuid_table[hole] = *uuid;
158 mutex_unlock(&xfs_uuid_table_mutex);
159
160 return 0;
161
162 out_duplicate:
163 mutex_unlock(&xfs_uuid_table_mutex);
Dave Chinner0b932cc2011-03-07 10:08:35 +1100164 xfs_warn(mp, "Filesystem has duplicate UUID - can't mount");
Christoph Hellwig27174202009-03-30 10:21:31 +0200165 return XFS_ERROR(EINVAL);
166}
167
168STATIC void
169xfs_uuid_unmount(
170 struct xfs_mount *mp)
171{
172 uuid_t *uuid = &mp->m_sb.sb_uuid;
173 int i;
174
175 if (mp->m_flags & XFS_MOUNT_NOUUID)
176 return;
177
178 mutex_lock(&xfs_uuid_table_mutex);
179 for (i = 0; i < xfs_uuid_table_size; i++) {
180 if (uuid_is_nil(&xfs_uuid_table[i]))
181 continue;
182 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
183 continue;
184 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
185 break;
186 }
187 ASSERT(i < xfs_uuid_table_size);
188 mutex_unlock(&xfs_uuid_table_mutex);
189}
190
191
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192/*
Dave Chinner0fa800f2010-01-11 11:47:46 +0000193 * Reference counting access wrappers to the perag structures.
Dave Chinnere1765792010-09-22 10:47:20 +1000194 * Because we never free per-ag structures, the only thing we
195 * have to protect against changes is the tree structure itself.
Dave Chinner0fa800f2010-01-11 11:47:46 +0000196 */
197struct xfs_perag *
198xfs_perag_get(struct xfs_mount *mp, xfs_agnumber_t agno)
199{
200 struct xfs_perag *pag;
201 int ref = 0;
202
Dave Chinnere1765792010-09-22 10:47:20 +1000203 rcu_read_lock();
Dave Chinner0fa800f2010-01-11 11:47:46 +0000204 pag = radix_tree_lookup(&mp->m_perag_tree, agno);
205 if (pag) {
206 ASSERT(atomic_read(&pag->pag_ref) >= 0);
Dave Chinner0fa800f2010-01-11 11:47:46 +0000207 ref = atomic_inc_return(&pag->pag_ref);
208 }
Dave Chinnere1765792010-09-22 10:47:20 +1000209 rcu_read_unlock();
Dave Chinner0fa800f2010-01-11 11:47:46 +0000210 trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
211 return pag;
212}
213
Dave Chinner65d0f202010-09-24 18:40:15 +1000214/*
215 * search from @first to find the next perag with the given tag set.
216 */
217struct xfs_perag *
218xfs_perag_get_tag(
219 struct xfs_mount *mp,
220 xfs_agnumber_t first,
221 int tag)
222{
223 struct xfs_perag *pag;
224 int found;
225 int ref;
226
227 rcu_read_lock();
228 found = radix_tree_gang_lookup_tag(&mp->m_perag_tree,
229 (void **)&pag, first, 1, tag);
230 if (found <= 0) {
231 rcu_read_unlock();
232 return NULL;
233 }
234 ref = atomic_inc_return(&pag->pag_ref);
235 rcu_read_unlock();
236 trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_);
237 return pag;
238}
239
Dave Chinner0fa800f2010-01-11 11:47:46 +0000240void
241xfs_perag_put(struct xfs_perag *pag)
242{
243 int ref;
244
245 ASSERT(atomic_read(&pag->pag_ref) > 0);
246 ref = atomic_dec_return(&pag->pag_ref);
247 trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
248}
249
Dave Chinnere1765792010-09-22 10:47:20 +1000250STATIC void
251__xfs_free_perag(
252 struct rcu_head *head)
253{
254 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
255
256 ASSERT(atomic_read(&pag->pag_ref) == 0);
257 kmem_free(pag);
258}
259
Dave Chinner0fa800f2010-01-11 11:47:46 +0000260/*
Dave Chinnere1765792010-09-22 10:47:20 +1000261 * Free up the per-ag resources associated with the mount structure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 */
Christoph Hellwigc962fb72008-05-20 15:10:52 +1000263STATIC void
Christoph Hellwigff4f0382008-08-13 16:50:47 +1000264xfs_free_perag(
Christoph Hellwig745f6912007-08-30 17:20:39 +1000265 xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266{
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000267 xfs_agnumber_t agno;
268 struct xfs_perag *pag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000270 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
271 spin_lock(&mp->m_perag_lock);
272 pag = radix_tree_delete(&mp->m_perag_tree, agno);
273 spin_unlock(&mp->m_perag_lock);
Dave Chinnere1765792010-09-22 10:47:20 +1000274 ASSERT(pag);
Dave Chinnerf83282a2010-11-08 08:55:04 +0000275 ASSERT(atomic_read(&pag->pag_ref) == 0);
Dave Chinnere1765792010-09-22 10:47:20 +1000276 call_rcu(&pag->rcu_head, __xfs_free_perag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278}
279
Nathan Scott4cc929e2007-05-14 18:24:02 +1000280/*
281 * Check size of device based on the (data/realtime) block count.
282 * Note: this check is used by the growfs code as well as mount.
283 */
284int
285xfs_sb_validate_fsb_count(
286 xfs_sb_t *sbp,
287 __uint64_t nblocks)
288{
289 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
290 ASSERT(sbp->sb_blocklog >= BBSHIFT);
291
292#if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
293 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
Eric Sandeen657a4cf2010-04-30 03:42:49 +0000294 return EFBIG;
Nathan Scott4cc929e2007-05-14 18:24:02 +1000295#else /* Limited by UINT_MAX of sectors */
296 if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
Eric Sandeen657a4cf2010-04-30 03:42:49 +0000297 return EFBIG;
Nathan Scott4cc929e2007-05-14 18:24:02 +1000298#endif
299 return 0;
300}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301
302/*
303 * Check the validity of the SB found.
304 */
305STATIC int
306xfs_mount_validate_sb(
307 xfs_mount_t *mp,
Nathan Scott764d1f82006-03-31 13:04:17 +1000308 xfs_sb_t *sbp,
309 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310{
Dave Chinneraf34e092011-03-07 10:04:35 +1100311 int loud = !(flags & XFS_MFSI_QUIET);
312
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 /*
314 * If the log device and data device have the
315 * same device number, the log is internal.
316 * Consequently, the sb_logstart should be non-zero. If
317 * we have a zero sb_logstart in this case, we may be trying to mount
318 * a volume filesystem in a non-volume manner.
319 */
320 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100321 if (loud)
322 xfs_warn(mp, "bad magic number");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 return XFS_ERROR(EWRONGFS);
324 }
325
Eric Sandeen62118702008-03-06 13:44:28 +1100326 if (!xfs_sb_good_version(sbp)) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100327 if (loud)
328 xfs_warn(mp, "bad version");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 return XFS_ERROR(EWRONGFS);
330 }
331
332 if (unlikely(
333 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100334 if (loud)
335 xfs_warn(mp,
336 "filesystem is marked as having an external log; "
337 "specify logdev on the mount command line.");
Nathan Scott764d1f82006-03-31 13:04:17 +1000338 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 }
340
341 if (unlikely(
342 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100343 if (loud)
344 xfs_warn(mp,
345 "filesystem is marked as having an internal log; "
346 "do not specify logdev on the mount command line.");
Nathan Scott764d1f82006-03-31 13:04:17 +1000347 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 }
349
350 /*
Eric Sandeenc0e090c2011-05-20 21:52:17 +0000351 * More sanity checking. Most of these were stolen directly from
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 * xfs_repair.
353 */
354 if (unlikely(
355 sbp->sb_agcount <= 0 ||
356 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
357 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
358 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
359 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
Olaf Weber2ac00af2009-04-17 16:12:45 -0500360 sbp->sb_sectsize != (1 << sbp->sb_sectlog) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
362 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
363 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
364 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
Olaf Weber2ac00af2009-04-17 16:12:45 -0500365 sbp->sb_blocksize != (1 << sbp->sb_blocklog) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
367 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
Nathan Scott9f989c92006-03-14 13:29:32 +1100368 sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
369 sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
Olaf Weber2ac00af2009-04-17 16:12:45 -0500370 sbp->sb_inodesize != (1 << sbp->sb_inodelog) ||
Nathan Scott9f989c92006-03-14 13:29:32 +1100371 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
373 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
Eric Sandeenc0e090c2011-05-20 21:52:17 +0000374 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */) ||
375 sbp->sb_dblocks == 0 ||
376 sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp) ||
377 sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp))) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100378 if (loud)
Eric Sandeenc0e090c2011-05-20 21:52:17 +0000379 XFS_CORRUPTION_ERROR("SB sanity check failed",
380 XFS_ERRLEVEL_LOW, mp, sbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 return XFS_ERROR(EFSCORRUPTED);
382 }
383
Lachlan McIlroy2edbddd2008-06-27 13:34:34 +1000384 /*
385 * Until this is fixed only page-sized or smaller data blocks work.
386 */
387 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100388 if (loud) {
389 xfs_warn(mp,
390 "File system with blocksize %d bytes. "
391 "Only pagesize (%ld) or less will currently work.",
392 sbp->sb_blocksize, PAGE_SIZE);
393 }
Lachlan McIlroy2edbddd2008-06-27 13:34:34 +1000394 return XFS_ERROR(ENOSYS);
395 }
396
Christoph Hellwig1a5902c2009-03-29 19:26:46 +0200397 /*
398 * Currently only very few inode sizes are supported.
399 */
400 switch (sbp->sb_inodesize) {
401 case 256:
402 case 512:
403 case 1024:
404 case 2048:
405 break;
406 default:
Dave Chinneraf34e092011-03-07 10:04:35 +1100407 if (loud)
408 xfs_warn(mp, "inode size of %d bytes not supported",
409 sbp->sb_inodesize);
Christoph Hellwig1a5902c2009-03-29 19:26:46 +0200410 return XFS_ERROR(ENOSYS);
411 }
412
Nathan Scott4cc929e2007-05-14 18:24:02 +1000413 if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
414 xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100415 if (loud)
416 xfs_warn(mp,
417 "file system too large to be mounted on this system.");
Eric Sandeen657a4cf2010-04-30 03:42:49 +0000418 return XFS_ERROR(EFBIG);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 }
420
421 if (unlikely(sbp->sb_inprogress)) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100422 if (loud)
423 xfs_warn(mp, "file system busy");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 return XFS_ERROR(EFSCORRUPTED);
425 }
426
427 /*
Nathan Scottde206142005-05-05 13:24:13 -0700428 * Version 1 directory format has never worked on Linux.
429 */
Eric Sandeen62118702008-03-06 13:44:28 +1100430 if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100431 if (loud)
432 xfs_warn(mp,
433 "file system using version 1 directory format");
Nathan Scottde206142005-05-05 13:24:13 -0700434 return XFS_ERROR(ENOSYS);
435 }
436
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 return 0;
438}
439
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000440int
Nathan Scottc11e2c32005-11-02 15:11:45 +1100441xfs_initialize_perag(
Nathan Scottc11e2c32005-11-02 15:11:45 +1100442 xfs_mount_t *mp,
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000443 xfs_agnumber_t agcount,
444 xfs_agnumber_t *maxagi)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445{
446 xfs_agnumber_t index, max_metadata;
Dave Chinner8b26c582010-01-11 11:47:48 +0000447 xfs_agnumber_t first_initialised = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 xfs_perag_t *pag;
449 xfs_agino_t agino;
450 xfs_ino_t ino;
451 xfs_sb_t *sbp = &mp->m_sb;
Dave Chinner8b26c582010-01-11 11:47:48 +0000452 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000454 /*
455 * Walk the current per-ag tree so we don't try to initialise AGs
456 * that already exist (growfs case). Allocate and insert all the
457 * AGs we don't find ready for initialisation.
458 */
459 for (index = 0; index < agcount; index++) {
460 pag = xfs_perag_get(mp, index);
461 if (pag) {
462 xfs_perag_put(pag);
463 continue;
464 }
Dave Chinner8b26c582010-01-11 11:47:48 +0000465 if (!first_initialised)
466 first_initialised = index;
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000467
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000468 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
469 if (!pag)
Dave Chinner8b26c582010-01-11 11:47:48 +0000470 goto out_unwind;
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000471 pag->pag_agno = index;
472 pag->pag_mount = mp;
Dave Chinner1a427ab2010-12-16 17:08:41 +1100473 spin_lock_init(&pag->pag_ici_lock);
Dave Chinner69b491c2010-09-27 11:09:51 +1000474 mutex_init(&pag->pag_ici_reclaim_lock);
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000475 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
Dave Chinner74f75a02010-09-24 19:59:04 +1000476 spin_lock_init(&pag->pag_buf_lock);
477 pag->pag_buf_tree = RB_ROOT;
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000478
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000479 if (radix_tree_preload(GFP_NOFS))
Dave Chinner8b26c582010-01-11 11:47:48 +0000480 goto out_unwind;
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000481
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000482 spin_lock(&mp->m_perag_lock);
483 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
484 BUG();
485 spin_unlock(&mp->m_perag_lock);
Dave Chinner8b26c582010-01-11 11:47:48 +0000486 radix_tree_preload_end();
487 error = -EEXIST;
488 goto out_unwind;
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000489 }
490 spin_unlock(&mp->m_perag_lock);
491 radix_tree_preload_end();
492 }
493
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000494 /*
495 * If we mount with the inode64 option, or no inode overflows
496 * the legacy 32-bit address space clear the inode32 option.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 */
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000498 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
499 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000501 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
502 mp->m_flags |= XFS_MOUNT_32BITINODES;
503 else
504 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
505
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000507 /*
508 * Calculate how much should be reserved for inodes to meet
509 * the max inode percentage.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 */
511 if (mp->m_maxicount) {
512 __uint64_t icount;
513
514 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
515 do_div(icount, 100);
516 icount += sbp->sb_agblocks - 1;
Eric Sandeena749ee82005-11-02 15:13:42 +1100517 do_div(icount, sbp->sb_agblocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 max_metadata = icount;
519 } else {
520 max_metadata = agcount;
521 }
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000522
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 for (index = 0; index < agcount; index++) {
524 ino = XFS_AGINO_TO_INO(mp, index, agino);
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000525 if (ino > XFS_MAXINUMBER_32) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 index++;
527 break;
528 }
529
Dave Chinner44b56e02010-01-11 11:47:43 +0000530 pag = xfs_perag_get(mp, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 pag->pagi_inodeok = 1;
532 if (index < max_metadata)
533 pag->pagf_metadata = 1;
Dave Chinner44b56e02010-01-11 11:47:43 +0000534 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 }
536 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 for (index = 0; index < agcount; index++) {
Dave Chinner44b56e02010-01-11 11:47:43 +0000538 pag = xfs_perag_get(mp, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 pag->pagi_inodeok = 1;
Dave Chinner44b56e02010-01-11 11:47:43 +0000540 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 }
542 }
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000543
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000544 if (maxagi)
545 *maxagi = index;
546 return 0;
Dave Chinner8b26c582010-01-11 11:47:48 +0000547
548out_unwind:
549 kmem_free(pag);
550 for (; index > first_initialised; index--) {
551 pag = radix_tree_delete(&mp->m_perag_tree, index);
552 kmem_free(pag);
553 }
554 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555}
556
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000557void
558xfs_sb_from_disk(
559 xfs_sb_t *to,
560 xfs_dsb_t *from)
561{
562 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
563 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
564 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
565 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
566 to->sb_rextents = be64_to_cpu(from->sb_rextents);
567 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
568 to->sb_logstart = be64_to_cpu(from->sb_logstart);
569 to->sb_rootino = be64_to_cpu(from->sb_rootino);
570 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
571 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
572 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
573 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
574 to->sb_agcount = be32_to_cpu(from->sb_agcount);
575 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
576 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
577 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
578 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
579 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
580 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
581 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
582 to->sb_blocklog = from->sb_blocklog;
583 to->sb_sectlog = from->sb_sectlog;
584 to->sb_inodelog = from->sb_inodelog;
585 to->sb_inopblog = from->sb_inopblog;
586 to->sb_agblklog = from->sb_agblklog;
587 to->sb_rextslog = from->sb_rextslog;
588 to->sb_inprogress = from->sb_inprogress;
589 to->sb_imax_pct = from->sb_imax_pct;
590 to->sb_icount = be64_to_cpu(from->sb_icount);
591 to->sb_ifree = be64_to_cpu(from->sb_ifree);
592 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
593 to->sb_frextents = be64_to_cpu(from->sb_frextents);
594 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
595 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
596 to->sb_qflags = be16_to_cpu(from->sb_qflags);
597 to->sb_flags = from->sb_flags;
598 to->sb_shared_vn = from->sb_shared_vn;
599 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
600 to->sb_unit = be32_to_cpu(from->sb_unit);
601 to->sb_width = be32_to_cpu(from->sb_width);
602 to->sb_dirblklog = from->sb_dirblklog;
603 to->sb_logsectlog = from->sb_logsectlog;
604 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
605 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
606 to->sb_features2 = be32_to_cpu(from->sb_features2);
David Chinneree1c0902008-03-06 13:45:50 +1100607 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000608}
609
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610/*
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000611 * Copy in core superblock to ondisk one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 *
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000613 * The fields argument is mask of superblock fields to copy.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 */
615void
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000616xfs_sb_to_disk(
617 xfs_dsb_t *to,
618 xfs_sb_t *from,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 __int64_t fields)
620{
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000621 xfs_caddr_t to_ptr = (xfs_caddr_t)to;
622 xfs_caddr_t from_ptr = (xfs_caddr_t)from;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 xfs_sb_field_t f;
624 int first;
625 int size;
626
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 ASSERT(fields);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 if (!fields)
629 return;
630
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 while (fields) {
632 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
633 first = xfs_sb_info[f].offset;
634 size = xfs_sb_info[f + 1].offset - first;
635
636 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
637
638 if (size == 1 || xfs_sb_info[f].type == 1) {
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000639 memcpy(to_ptr + first, from_ptr + first, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 } else {
641 switch (size) {
642 case 2:
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000643 *(__be16 *)(to_ptr + first) =
644 cpu_to_be16(*(__u16 *)(from_ptr + first));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 break;
646 case 4:
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000647 *(__be32 *)(to_ptr + first) =
648 cpu_to_be32(*(__u32 *)(from_ptr + first));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 break;
650 case 8:
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000651 *(__be64 *)(to_ptr + first) =
652 cpu_to_be64(*(__u64 *)(from_ptr + first));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 break;
654 default:
655 ASSERT(0);
656 }
657 }
658
659 fields &= ~(1LL << f);
660 }
661}
662
663/*
664 * xfs_readsb
665 *
666 * Does the initial read of the superblock.
667 */
668int
Nathan Scott764d1f82006-03-31 13:04:17 +1000669xfs_readsb(xfs_mount_t *mp, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670{
671 unsigned int sector_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 int error;
Dave Chinneraf34e092011-03-07 10:04:35 +1100674 int loud = !(flags & XFS_MFSI_QUIET);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675
676 ASSERT(mp->m_sb_bp == NULL);
677 ASSERT(mp->m_ddev_targp != NULL);
678
679 /*
680 * Allocate a (locked) buffer to hold the superblock.
681 * This will be kept around at all times to optimize
682 * access to the superblock.
683 */
684 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685
Dave Chinner26af6552010-09-22 10:47:20 +1000686reread:
687 bp = xfs_buf_read_uncached(mp, mp->m_ddev_targp,
688 XFS_SB_DADDR, sector_size, 0);
689 if (!bp) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100690 if (loud)
691 xfs_warn(mp, "SB buffer read failed");
Dave Chinner26af6552010-09-22 10:47:20 +1000692 return EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
695 /*
696 * Initialize the mount structure from the superblock.
697 * But first do some basic consistency checking.
698 */
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000699 xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
Nathan Scott764d1f82006-03-31 13:04:17 +1000700 error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 if (error) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100702 if (loud)
703 xfs_warn(mp, "SB validate failed");
Dave Chinner26af6552010-09-22 10:47:20 +1000704 goto release_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 }
706
707 /*
708 * We must be able to do sector-sized and sector-aligned IO.
709 */
710 if (sector_size > mp->m_sb.sb_sectsize) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100711 if (loud)
712 xfs_warn(mp, "device supports %u byte sectors (not %u)",
713 sector_size, mp->m_sb.sb_sectsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 error = ENOSYS;
Dave Chinner26af6552010-09-22 10:47:20 +1000715 goto release_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 }
717
718 /*
719 * If device sector size is smaller than the superblock size,
720 * re-read the superblock so the buffer is correctly sized.
721 */
722 if (sector_size < mp->m_sb.sb_sectsize) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 xfs_buf_relse(bp);
724 sector_size = mp->m_sb.sb_sectsize;
Dave Chinner26af6552010-09-22 10:47:20 +1000725 goto reread;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 }
727
Lachlan McIlroy5478eea2007-02-10 18:36:29 +1100728 /* Initialize per-cpu counters */
729 xfs_icsb_reinit_counters(mp);
David Chinner8d280b92006-03-14 13:13:09 +1100730
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 mp->m_sb_bp = bp;
Dave Chinner26af6552010-09-22 10:47:20 +1000732 xfs_buf_unlock(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 return 0;
734
Dave Chinner26af6552010-09-22 10:47:20 +1000735release_buf:
736 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 return error;
738}
739
740
741/*
742 * xfs_mount_common
743 *
744 * Mount initialization code establishing various mount
745 * fields from the superblock associated with the given
746 * mount structure
747 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000748STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
750{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 mp->m_agfrotor = mp->m_agirotor = 0;
Eric Sandeen007c61c2007-10-11 17:43:56 +1000752 spin_lock_init(&mp->m_agirotor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 mp->m_maxagi = mp->m_sb.sb_agcount;
754 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
755 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
756 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
757 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
758 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 mp->m_blockmask = sbp->sb_blocksize - 1;
760 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
761 mp->m_blockwmask = mp->m_blockwsize - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762
Christoph Hellwig60197e82008-10-30 17:11:19 +1100763 mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
764 mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
765 mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
766 mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
767
768 mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
769 mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
770 mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
771 mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
772
773 mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
774 mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
775 mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
776 mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
778 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
779 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
780 sbp->sb_inopblock);
781 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
782}
David Chinner92821e22007-05-24 15:26:31 +1000783
784/*
785 * xfs_initialize_perag_data
786 *
787 * Read in each per-ag structure so we can count up the number of
788 * allocated inodes, free inodes and used filesystem blocks as this
789 * information is no longer persistent in the superblock. Once we have
790 * this information, write it into the in-core superblock structure.
791 */
792STATIC int
793xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
794{
795 xfs_agnumber_t index;
796 xfs_perag_t *pag;
797 xfs_sb_t *sbp = &mp->m_sb;
798 uint64_t ifree = 0;
799 uint64_t ialloc = 0;
800 uint64_t bfree = 0;
801 uint64_t bfreelst = 0;
802 uint64_t btree = 0;
803 int error;
David Chinner92821e22007-05-24 15:26:31 +1000804
805 for (index = 0; index < agcount; index++) {
806 /*
807 * read the agf, then the agi. This gets us
Malcolm Parsons9da096f2009-03-29 09:55:42 +0200808 * all the information we need and populates the
David Chinner92821e22007-05-24 15:26:31 +1000809 * per-ag structures for us.
810 */
811 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
812 if (error)
813 return error;
814
815 error = xfs_ialloc_pagi_init(mp, NULL, index);
816 if (error)
817 return error;
Dave Chinner44b56e02010-01-11 11:47:43 +0000818 pag = xfs_perag_get(mp, index);
David Chinner92821e22007-05-24 15:26:31 +1000819 ifree += pag->pagi_freecount;
820 ialloc += pag->pagi_count;
821 bfree += pag->pagf_freeblks;
822 bfreelst += pag->pagf_flcount;
823 btree += pag->pagf_btreeblks;
Dave Chinner44b56e02010-01-11 11:47:43 +0000824 xfs_perag_put(pag);
David Chinner92821e22007-05-24 15:26:31 +1000825 }
826 /*
827 * Overwrite incore superblock counters with just-read data
828 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +1000829 spin_lock(&mp->m_sb_lock);
David Chinner92821e22007-05-24 15:26:31 +1000830 sbp->sb_ifree = ifree;
831 sbp->sb_icount = ialloc;
832 sbp->sb_fdblocks = bfree + bfreelst + btree;
Eric Sandeen3685c2a2007-10-11 17:42:32 +1000833 spin_unlock(&mp->m_sb_lock);
David Chinner92821e22007-05-24 15:26:31 +1000834
835 /* Fixup the per-cpu counters as well. */
836 xfs_icsb_reinit_counters(mp);
837
838 return 0;
839}
840
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841/*
Eric Sandeen0771fb42007-10-12 11:03:40 +1000842 * Update alignment values based on mount options and sb values
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 */
Eric Sandeen0771fb42007-10-12 11:03:40 +1000844STATIC int
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100845xfs_update_alignment(xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 xfs_sb_t *sbp = &(mp->m_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
Christoph Hellwig42490232008-08-13 16:49:32 +1000849 if (mp->m_dalign) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 /*
851 * If stripe unit and stripe width are not multiples
852 * of the fs blocksize turn off alignment.
853 */
854 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
855 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
856 if (mp->m_flags & XFS_MOUNT_RETERR) {
Eric Sandeenc0e090c2011-05-20 21:52:17 +0000857 xfs_warn(mp, "alignment check failed: "
858 "(sunit/swidth vs. blocksize)");
Eric Sandeen0771fb42007-10-12 11:03:40 +1000859 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 }
861 mp->m_dalign = mp->m_swidth = 0;
862 } else {
863 /*
864 * Convert the stripe unit and width to FSBs.
865 */
866 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
867 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
868 if (mp->m_flags & XFS_MOUNT_RETERR) {
Eric Sandeenc0e090c2011-05-20 21:52:17 +0000869 xfs_warn(mp, "alignment check failed: "
870 "(sunit/swidth vs. ag size)");
Eric Sandeen0771fb42007-10-12 11:03:40 +1000871 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 }
Dave Chinner53487782011-03-07 10:05:35 +1100873 xfs_warn(mp,
874 "stripe alignment turned off: sunit(%d)/swidth(%d) "
875 "incompatible with agsize(%d)",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 mp->m_dalign, mp->m_swidth,
877 sbp->sb_agblocks);
878
879 mp->m_dalign = 0;
880 mp->m_swidth = 0;
881 } else if (mp->m_dalign) {
882 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
883 } else {
884 if (mp->m_flags & XFS_MOUNT_RETERR) {
Eric Sandeenc0e090c2011-05-20 21:52:17 +0000885 xfs_warn(mp, "alignment check failed: "
886 "sunit(%d) less than bsize(%d)",
Dave Chinner53487782011-03-07 10:05:35 +1100887 mp->m_dalign,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 mp->m_blockmask +1);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000889 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 }
891 mp->m_swidth = 0;
892 }
893 }
894
895 /*
896 * Update superblock with new values
897 * and log changes
898 */
Eric Sandeen62118702008-03-06 13:44:28 +1100899 if (xfs_sb_version_hasdalign(sbp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 if (sbp->sb_unit != mp->m_dalign) {
901 sbp->sb_unit = mp->m_dalign;
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100902 mp->m_update_flags |= XFS_SB_UNIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 }
904 if (sbp->sb_width != mp->m_swidth) {
905 sbp->sb_width = mp->m_swidth;
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100906 mp->m_update_flags |= XFS_SB_WIDTH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 }
908 }
909 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
Eric Sandeen62118702008-03-06 13:44:28 +1100910 xfs_sb_version_hasdalign(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 mp->m_dalign = sbp->sb_unit;
912 mp->m_swidth = sbp->sb_width;
913 }
914
Eric Sandeen0771fb42007-10-12 11:03:40 +1000915 return 0;
916}
917
918/*
919 * Set the maximum inode count for this filesystem
920 */
921STATIC void
922xfs_set_maxicount(xfs_mount_t *mp)
923{
924 xfs_sb_t *sbp = &(mp->m_sb);
925 __uint64_t icount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926
927 if (sbp->sb_imax_pct) {
Eric Sandeen0771fb42007-10-12 11:03:40 +1000928 /*
929 * Make sure the maximum inode count is a multiple
930 * of the units we allocate inodes in.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
933 do_div(icount, 100);
934 do_div(icount, mp->m_ialloc_blks);
935 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
936 sbp->sb_inopblog;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000937 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 mp->m_maxicount = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 }
Eric Sandeen0771fb42007-10-12 11:03:40 +1000940}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941
Eric Sandeen0771fb42007-10-12 11:03:40 +1000942/*
943 * Set the default minimum read and write sizes unless
944 * already specified in a mount option.
945 * We use smaller I/O sizes when the file system
946 * is being used for NFS service (wsync mount option).
947 */
948STATIC void
949xfs_set_rw_sizes(xfs_mount_t *mp)
950{
951 xfs_sb_t *sbp = &(mp->m_sb);
952 int readio_log, writeio_log;
953
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
955 if (mp->m_flags & XFS_MOUNT_WSYNC) {
956 readio_log = XFS_WSYNC_READIO_LOG;
957 writeio_log = XFS_WSYNC_WRITEIO_LOG;
958 } else {
959 readio_log = XFS_READIO_LOG_LARGE;
960 writeio_log = XFS_WRITEIO_LOG_LARGE;
961 }
962 } else {
963 readio_log = mp->m_readio_log;
964 writeio_log = mp->m_writeio_log;
965 }
966
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 if (sbp->sb_blocklog > readio_log) {
968 mp->m_readio_log = sbp->sb_blocklog;
969 } else {
970 mp->m_readio_log = readio_log;
971 }
972 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
973 if (sbp->sb_blocklog > writeio_log) {
974 mp->m_writeio_log = sbp->sb_blocklog;
975 } else {
976 mp->m_writeio_log = writeio_log;
977 }
978 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000979}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
Eric Sandeen0771fb42007-10-12 11:03:40 +1000981/*
Dave Chinner055388a2011-01-04 11:35:03 +1100982 * precalculate the low space thresholds for dynamic speculative preallocation.
983 */
984void
985xfs_set_low_space_thresholds(
986 struct xfs_mount *mp)
987{
988 int i;
989
990 for (i = 0; i < XFS_LOWSP_MAX; i++) {
991 __uint64_t space = mp->m_sb.sb_dblocks;
992
993 do_div(space, 100);
994 mp->m_low_space[i] = space * (i + 1);
995 }
996}
997
998
999/*
Eric Sandeen0771fb42007-10-12 11:03:40 +10001000 * Set whether we're using inode alignment.
1001 */
1002STATIC void
1003xfs_set_inoalignment(xfs_mount_t *mp)
1004{
Eric Sandeen62118702008-03-06 13:44:28 +11001005 if (xfs_sb_version_hasalign(&mp->m_sb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 mp->m_sb.sb_inoalignmt >=
1007 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
1008 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
1009 else
1010 mp->m_inoalign_mask = 0;
1011 /*
1012 * If we are using stripe alignment, check whether
1013 * the stripe unit is a multiple of the inode alignment
1014 */
1015 if (mp->m_dalign && mp->m_inoalign_mask &&
1016 !(mp->m_dalign & mp->m_inoalign_mask))
1017 mp->m_sinoalign = mp->m_dalign;
1018 else
1019 mp->m_sinoalign = 0;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001020}
1021
1022/*
1023 * Check that the data (and log if separate) are an ok size.
1024 */
1025STATIC int
Christoph Hellwig42490232008-08-13 16:49:32 +10001026xfs_check_sizes(xfs_mount_t *mp)
Eric Sandeen0771fb42007-10-12 11:03:40 +10001027{
1028 xfs_buf_t *bp;
1029 xfs_daddr_t d;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001030
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
1032 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001033 xfs_warn(mp, "filesystem size mismatch detected");
Eric Sandeen657a4cf2010-04-30 03:42:49 +00001034 return XFS_ERROR(EFBIG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 }
Dave Chinner1922c942010-09-22 10:47:20 +10001036 bp = xfs_buf_read_uncached(mp, mp->m_ddev_targp,
1037 d - XFS_FSS_TO_BB(mp, 1),
1038 BBTOB(XFS_FSS_TO_BB(mp, 1)), 0);
1039 if (!bp) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001040 xfs_warn(mp, "last sector read failed");
Dave Chinner1922c942010-09-22 10:47:20 +10001041 return EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 }
Dave Chinner1922c942010-09-22 10:47:20 +10001043 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
Christoph Hellwig42490232008-08-13 16:49:32 +10001045 if (mp->m_logdev_targp != mp->m_ddev_targp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
1047 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001048 xfs_warn(mp, "log size mismatch detected");
Eric Sandeen657a4cf2010-04-30 03:42:49 +00001049 return XFS_ERROR(EFBIG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 }
Dave Chinner1922c942010-09-22 10:47:20 +10001051 bp = xfs_buf_read_uncached(mp, mp->m_logdev_targp,
1052 d - XFS_FSB_TO_BB(mp, 1),
1053 XFS_FSB_TO_B(mp, 1), 0);
1054 if (!bp) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001055 xfs_warn(mp, "log device read failed");
Dave Chinner1922c942010-09-22 10:47:20 +10001056 return EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 }
Dave Chinner1922c942010-09-22 10:47:20 +10001058 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 }
Eric Sandeen0771fb42007-10-12 11:03:40 +10001060 return 0;
1061}
1062
1063/*
Christoph Hellwig7d095252009-06-08 15:33:32 +02001064 * Clear the quotaflags in memory and in the superblock.
1065 */
1066int
1067xfs_mount_reset_sbqflags(
1068 struct xfs_mount *mp)
1069{
1070 int error;
1071 struct xfs_trans *tp;
1072
1073 mp->m_qflags = 0;
1074
1075 /*
1076 * It is OK to look at sb_qflags here in mount path,
1077 * without m_sb_lock.
1078 */
1079 if (mp->m_sb.sb_qflags == 0)
1080 return 0;
1081 spin_lock(&mp->m_sb_lock);
1082 mp->m_sb.sb_qflags = 0;
1083 spin_unlock(&mp->m_sb_lock);
1084
1085 /*
1086 * If the fs is readonly, let the incore superblock run
1087 * with quotas off but don't flush the update out to disk
1088 */
1089 if (mp->m_flags & XFS_MOUNT_RDONLY)
1090 return 0;
1091
Christoph Hellwig7d095252009-06-08 15:33:32 +02001092 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1093 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1094 XFS_DEFAULT_LOG_COUNT);
1095 if (error) {
1096 xfs_trans_cancel(tp, 0);
Dave Chinner53487782011-03-07 10:05:35 +11001097 xfs_alert(mp, "%s: Superblock update failed!", __func__);
Christoph Hellwig7d095252009-06-08 15:33:32 +02001098 return error;
1099 }
1100
1101 xfs_mod_sb(tp, XFS_SB_QFLAGS);
1102 return xfs_trans_commit(tp, 0);
1103}
1104
Eric Sandeend5db0f92010-02-05 22:59:53 +00001105__uint64_t
1106xfs_default_resblks(xfs_mount_t *mp)
1107{
1108 __uint64_t resblks;
1109
1110 /*
Dave Chinner8babd8a2010-03-04 01:46:25 +00001111 * We default to 5% or 8192 fsbs of space reserved, whichever is
1112 * smaller. This is intended to cover concurrent allocation
1113 * transactions when we initially hit enospc. These each require a 4
1114 * block reservation. Hence by default we cover roughly 2000 concurrent
1115 * allocation reservations.
Eric Sandeend5db0f92010-02-05 22:59:53 +00001116 */
1117 resblks = mp->m_sb.sb_dblocks;
1118 do_div(resblks, 20);
Dave Chinner8babd8a2010-03-04 01:46:25 +00001119 resblks = min_t(__uint64_t, resblks, 8192);
Eric Sandeend5db0f92010-02-05 22:59:53 +00001120 return resblks;
1121}
1122
Christoph Hellwig7d095252009-06-08 15:33:32 +02001123/*
Eric Sandeen0771fb42007-10-12 11:03:40 +10001124 * This function does the following on an initial mount of a file system:
1125 * - reads the superblock from disk and init the mount struct
1126 * - if we're a 32-bit kernel, do a size check on the superblock
1127 * so we don't mount terabyte filesystems
1128 * - init mount struct realtime fields
1129 * - allocate inode hash table for fs
1130 * - init directory manager
1131 * - perform recovery and init the log manager
1132 */
1133int
1134xfs_mountfs(
Christoph Hellwig42490232008-08-13 16:49:32 +10001135 xfs_mount_t *mp)
Eric Sandeen0771fb42007-10-12 11:03:40 +10001136{
1137 xfs_sb_t *sbp = &(mp->m_sb);
1138 xfs_inode_t *rip;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001139 __uint64_t resblks;
Christoph Hellwig7d095252009-06-08 15:33:32 +02001140 uint quotamount = 0;
1141 uint quotaflags = 0;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001142 int error = 0;
1143
Eric Sandeen0771fb42007-10-12 11:03:40 +10001144 xfs_mount_common(mp, sbp);
1145
1146 /*
Eric Sandeene6957ea2008-04-10 12:19:34 +10001147 * Check for a mismatched features2 values. Older kernels
1148 * read & wrote into the wrong sb offset for sb_features2
1149 * on some platforms due to xfs_sb_t not being 64bit size aligned
1150 * when sb_features2 was added, which made older superblock
1151 * reading/writing routines swap it as a 64-bit value.
David Chinneree1c0902008-03-06 13:45:50 +11001152 *
Eric Sandeene6957ea2008-04-10 12:19:34 +10001153 * For backwards compatibility, we make both slots equal.
1154 *
1155 * If we detect a mismatched field, we OR the set bits into the
1156 * existing features2 field in case it has already been modified; we
1157 * don't want to lose any features. We then update the bad location
1158 * with the ORed value so that older kernels will see any features2
1159 * flags, and mark the two fields as needing updates once the
1160 * transaction subsystem is online.
David Chinneree1c0902008-03-06 13:45:50 +11001161 */
Eric Sandeene6957ea2008-04-10 12:19:34 +10001162 if (xfs_sb_has_mismatched_features2(sbp)) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001163 xfs_warn(mp, "correcting sb_features alignment problem");
David Chinneree1c0902008-03-06 13:45:50 +11001164 sbp->sb_features2 |= sbp->sb_bad_features2;
Eric Sandeene6957ea2008-04-10 12:19:34 +10001165 sbp->sb_bad_features2 = sbp->sb_features2;
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001166 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
Eric Sandeene6957ea2008-04-10 12:19:34 +10001167
1168 /*
1169 * Re-check for ATTR2 in case it was found in bad_features2
1170 * slot.
1171 */
Tim Shimmin7c12f292008-04-30 18:15:28 +10001172 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1173 !(mp->m_flags & XFS_MOUNT_NOATTR2))
Eric Sandeene6957ea2008-04-10 12:19:34 +10001174 mp->m_flags |= XFS_MOUNT_ATTR2;
Tim Shimmin7c12f292008-04-30 18:15:28 +10001175 }
Eric Sandeene6957ea2008-04-10 12:19:34 +10001176
Tim Shimmin7c12f292008-04-30 18:15:28 +10001177 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1178 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1179 xfs_sb_version_removeattr2(&mp->m_sb);
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001180 mp->m_update_flags |= XFS_SB_FEATURES2;
Tim Shimmin7c12f292008-04-30 18:15:28 +10001181
1182 /* update sb_versionnum for the clearing of the morebits */
1183 if (!sbp->sb_features2)
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001184 mp->m_update_flags |= XFS_SB_VERSIONNUM;
David Chinneree1c0902008-03-06 13:45:50 +11001185 }
1186
1187 /*
Eric Sandeen0771fb42007-10-12 11:03:40 +10001188 * Check if sb_agblocks is aligned at stripe boundary
1189 * If sb_agblocks is NOT aligned turn off m_dalign since
1190 * allocator alignment is within an ag, therefore ag has
1191 * to be aligned at stripe boundary.
1192 */
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001193 error = xfs_update_alignment(mp);
Eric Sandeen0771fb42007-10-12 11:03:40 +10001194 if (error)
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001195 goto out;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001196
1197 xfs_alloc_compute_maxlevels(mp);
1198 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
1199 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
1200 xfs_ialloc_compute_maxlevels(mp);
1201
1202 xfs_set_maxicount(mp);
1203
1204 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
1205
Christoph Hellwig27174202009-03-30 10:21:31 +02001206 error = xfs_uuid_mount(mp);
1207 if (error)
1208 goto out;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001209
1210 /*
1211 * Set the minimum read and write sizes
1212 */
1213 xfs_set_rw_sizes(mp);
1214
Dave Chinner055388a2011-01-04 11:35:03 +11001215 /* set the low space thresholds for dynamic preallocation */
1216 xfs_set_low_space_thresholds(mp);
1217
Eric Sandeen0771fb42007-10-12 11:03:40 +10001218 /*
1219 * Set the inode cluster size.
1220 * This may still be overridden by the file system
1221 * block size if it is larger than the chosen cluster size.
1222 */
1223 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1224
1225 /*
1226 * Set inode alignment fields
1227 */
1228 xfs_set_inoalignment(mp);
1229
1230 /*
1231 * Check that the data (and log if separate) are an ok size.
1232 */
Christoph Hellwig42490232008-08-13 16:49:32 +10001233 error = xfs_check_sizes(mp);
Eric Sandeen0771fb42007-10-12 11:03:40 +10001234 if (error)
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001235 goto out_remove_uuid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236
1237 /*
1238 * Initialize realtime fields in the mount structure
1239 */
Eric Sandeen0771fb42007-10-12 11:03:40 +10001240 error = xfs_rtmount_init(mp);
1241 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001242 xfs_warn(mp, "RT mount failed");
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001243 goto out_remove_uuid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 }
1245
1246 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 * Copies the low order bits of the timestamp and the randomly
1248 * set "sequence" number out of a UUID.
1249 */
1250 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1251
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 mp->m_dmevmask = 0; /* not persistent; set after each mount */
1253
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001254 xfs_dir_mount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255
1256 /*
1257 * Initialize the attribute manager's entries.
1258 */
1259 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1260
1261 /*
1262 * Initialize the precomputed transaction reservations values.
1263 */
1264 xfs_trans_init(mp);
1265
1266 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 * Allocate and initialize the per-ag data.
1268 */
Dave Chinner1c1c6eb2010-01-11 11:47:44 +00001269 spin_lock_init(&mp->m_perag_lock);
Dave Chinner9b98b6f2010-05-27 01:58:13 +00001270 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
Dave Chinner1c1c6eb2010-01-11 11:47:44 +00001271 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
1272 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001273 xfs_warn(mp, "Failed per-ag init: %d", error);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001274 goto out_remove_uuid;
Dave Chinner1c1c6eb2010-01-11 11:47:44 +00001275 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001277 if (!sbp->sb_logblocks) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001278 xfs_warn(mp, "no log defined");
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001279 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
1280 error = XFS_ERROR(EFSCORRUPTED);
1281 goto out_free_perag;
1282 }
1283
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 /*
1285 * log's mount-time initialization. Perform 1st part recovery if needed
1286 */
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001287 error = xfs_log_mount(mp, mp->m_logdev_targp,
1288 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1289 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1290 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001291 xfs_warn(mp, "log mount failed");
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001292 goto out_free_perag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 }
1294
1295 /*
David Chinner92821e22007-05-24 15:26:31 +10001296 * Now the log is mounted, we know if it was an unclean shutdown or
1297 * not. If it was, with the first phase of recovery has completed, we
1298 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1299 * but they are recovered transactionally in the second recovery phase
1300 * later.
1301 *
1302 * Hence we can safely re-initialise incore superblock counters from
1303 * the per-ag data. These may not be correct if the filesystem was not
1304 * cleanly unmounted, so we need to wait for recovery to finish before
1305 * doing this.
1306 *
1307 * If the filesystem was cleanly unmounted, then we can trust the
1308 * values in the superblock to be correct and we don't need to do
1309 * anything here.
1310 *
1311 * If we are currently making the filesystem, the initialisation will
1312 * fail as the perag data is in an undefined state.
1313 */
David Chinner92821e22007-05-24 15:26:31 +10001314 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1315 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1316 !mp->m_sb.sb_inprogress) {
1317 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001318 if (error)
1319 goto out_free_perag;
David Chinner92821e22007-05-24 15:26:31 +10001320 }
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001321
David Chinner92821e22007-05-24 15:26:31 +10001322 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 * Get and sanity-check the root inode.
1324 * Save the pointer to it in the mount structure.
1325 */
Dave Chinner7b6259e2010-06-24 11:35:17 +10001326 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001328 xfs_warn(mp, "failed to read root inode");
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001329 goto out_log_dealloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 }
1331
1332 ASSERT(rip != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
Al Viroabbede12011-07-26 02:31:30 -04001334 if (unlikely(!S_ISDIR(rip->i_d.di_mode))) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001335 xfs_warn(mp, "corrupted root inode %llu: not a directory",
Nathan Scottb6574522006-06-09 15:29:40 +10001336 (unsigned long long)rip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1338 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1339 mp);
1340 error = XFS_ERROR(EFSCORRUPTED);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001341 goto out_rele_rip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 }
1343 mp->m_rootip = rip; /* save it */
1344
1345 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1346
1347 /*
1348 * Initialize realtime inode pointers in the mount structure
1349 */
Eric Sandeen0771fb42007-10-12 11:03:40 +10001350 error = xfs_rtmount_inodes(mp);
1351 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 /*
1353 * Free up the root inode.
1354 */
Dave Chinner0b932cc2011-03-07 10:08:35 +11001355 xfs_warn(mp, "failed to read RT inodes");
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001356 goto out_rele_rip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 }
1358
1359 /*
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001360 * If this is a read-only mount defer the superblock updates until
1361 * the next remount into writeable mode. Otherwise we would never
1362 * perform the update e.g. for the root filesystem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 */
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001364 if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1365 error = xfs_mount_log_sb(mp, mp->m_update_flags);
David Chinnere5720ee2008-04-10 12:21:18 +10001366 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001367 xfs_warn(mp, "failed to write sb changes");
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001368 goto out_rtunmount;
David Chinnere5720ee2008-04-10 12:21:18 +10001369 }
1370 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
1372 /*
1373 * Initialise the XFS quota management subsystem for this mount
1374 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001375 if (XFS_IS_QUOTA_RUNNING(mp)) {
1376 error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
1377 if (error)
1378 goto out_rtunmount;
1379 } else {
1380 ASSERT(!XFS_IS_QUOTA_ON(mp));
1381
1382 /*
1383 * If a file system had quotas running earlier, but decided to
1384 * mount without -o uquota/pquota/gquota options, revoke the
1385 * quotachecked license.
1386 */
1387 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001388 xfs_notice(mp, "resetting quota flags");
Christoph Hellwig7d095252009-06-08 15:33:32 +02001389 error = xfs_mount_reset_sbqflags(mp);
1390 if (error)
1391 return error;
1392 }
1393 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394
1395 /*
1396 * Finish recovering the file system. This part needed to be
1397 * delayed until after the root and real-time bitmap inodes
1398 * were consistently read in.
1399 */
Christoph Hellwig42490232008-08-13 16:49:32 +10001400 error = xfs_log_mount_finish(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001402 xfs_warn(mp, "log mount finish failed");
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001403 goto out_rtunmount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 }
1405
1406 /*
1407 * Complete the quota initialisation, post-log-replay component.
1408 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001409 if (quotamount) {
1410 ASSERT(mp->m_qflags == 0);
1411 mp->m_qflags = quotaflags;
1412
1413 xfs_qm_mount_quotas(mp);
1414 }
1415
David Chinner84e1e992007-06-18 16:50:27 +10001416 /*
1417 * Now we are mounted, reserve a small amount of unused space for
1418 * privileged transactions. This is needed so that transaction
1419 * space required for critical operations can dip into this pool
1420 * when at ENOSPC. This is needed for operations like create with
1421 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1422 * are not allowed to use this reserved space.
Dave Chinner8babd8a2010-03-04 01:46:25 +00001423 *
1424 * This may drive us straight to ENOSPC on mount, but that implies
1425 * we were already there on the last unmount. Warn if this occurs.
David Chinner84e1e992007-06-18 16:50:27 +10001426 */
Eric Sandeend5db0f92010-02-05 22:59:53 +00001427 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
1428 resblks = xfs_default_resblks(mp);
1429 error = xfs_reserve_blocks(mp, &resblks, NULL);
1430 if (error)
Dave Chinner0b932cc2011-03-07 10:08:35 +11001431 xfs_warn(mp,
1432 "Unable to allocate reserve blocks. Continuing without reserve pool.");
Eric Sandeend5db0f92010-02-05 22:59:53 +00001433 }
David Chinner84e1e992007-06-18 16:50:27 +10001434
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 return 0;
1436
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001437 out_rtunmount:
1438 xfs_rtunmount_inodes(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001439 out_rele_rip:
Christoph Hellwig43355092008-03-27 18:01:08 +11001440 IRELE(rip);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001441 out_log_dealloc:
Christoph Hellwig21b699c2009-03-16 08:19:29 +01001442 xfs_log_unmount(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001443 out_free_perag:
Christoph Hellwigff4f0382008-08-13 16:50:47 +10001444 xfs_free_perag(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001445 out_remove_uuid:
Christoph Hellwig27174202009-03-30 10:21:31 +02001446 xfs_uuid_unmount(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001447 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 return error;
1449}
1450
1451/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 * This flushes out the inodes,dquots and the superblock, unmounts the
1453 * log and makes sure that incore structures are freed.
1454 */
Christoph Hellwig41b5c2e2008-08-13 16:49:57 +10001455void
1456xfs_unmountfs(
1457 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458{
Christoph Hellwig41b5c2e2008-08-13 16:49:57 +10001459 __uint64_t resblks;
1460 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461
Christoph Hellwig7d095252009-06-08 15:33:32 +02001462 xfs_qm_unmount_quotas(mp);
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001463 xfs_rtunmount_inodes(mp);
Christoph Hellwig77508ec2008-08-13 16:49:04 +10001464 IRELE(mp->m_rootip);
1465
David Chinner641c56f2007-06-18 16:50:17 +10001466 /*
1467 * We can potentially deadlock here if we have an inode cluster
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001468 * that has been freed has its buffer still pinned in memory because
David Chinner641c56f2007-06-18 16:50:17 +10001469 * the transaction is still sitting in a iclog. The stale inodes
1470 * on that buffer will have their flush locks held until the
1471 * transaction hits the disk and the callbacks run. the inode
1472 * flush takes the flush lock unconditionally and with nothing to
1473 * push out the iclog we will never get that unlocked. hence we
1474 * need to force the log first.
1475 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00001476 xfs_log_force(mp, XFS_LOG_SYNC);
Dave Chinnerc8543632010-02-06 12:39:36 +11001477
1478 /*
1479 * Do a delwri reclaim pass first so that as many dirty inodes are
1480 * queued up for IO as possible. Then flush the buffers before making
1481 * a synchronous path to catch all the remaining inodes are reclaimed.
1482 * This makes the reclaim process as quick as possible by avoiding
1483 * synchronous writeout and blocking on inodes already in the delwri
1484 * state as much as possible.
1485 */
1486 xfs_reclaim_inodes(mp, 0);
1487 XFS_bflush(mp->m_ddev_targp);
1488 xfs_reclaim_inodes(mp, SYNC_WAIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489
Christoph Hellwig7d095252009-06-08 15:33:32 +02001490 xfs_qm_unmount(mp);
Lachlan McIlroya357a122008-10-30 16:53:25 +11001491
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 /*
1493 * Flush out the log synchronously so that we know for sure
1494 * that nothing is pinned. This is important because bflush()
1495 * will skip pinned buffers.
1496 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00001497 xfs_log_force(mp, XFS_LOG_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498
1499 xfs_binval(mp->m_ddev_targp);
1500 if (mp->m_rtdev_targp) {
1501 xfs_binval(mp->m_rtdev_targp);
1502 }
1503
David Chinner84e1e992007-06-18 16:50:27 +10001504 /*
1505 * Unreserve any blocks we have so that when we unmount we don't account
1506 * the reserved free space as used. This is really only necessary for
1507 * lazy superblock counting because it trusts the incore superblock
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001508 * counters to be absolutely correct on clean unmount.
David Chinner84e1e992007-06-18 16:50:27 +10001509 *
1510 * We don't bother correcting this elsewhere for lazy superblock
1511 * counting because on mount of an unclean filesystem we reconstruct the
1512 * correct counter value and this is irrelevant.
1513 *
1514 * For non-lazy counter filesystems, this doesn't matter at all because
1515 * we only every apply deltas to the superblock and hence the incore
1516 * value does not matter....
1517 */
1518 resblks = 0;
David Chinner714082b2008-04-10 12:20:03 +10001519 error = xfs_reserve_blocks(mp, &resblks, NULL);
1520 if (error)
Dave Chinner0b932cc2011-03-07 10:08:35 +11001521 xfs_warn(mp, "Unable to free reserved block pool. "
David Chinner714082b2008-04-10 12:20:03 +10001522 "Freespace may not be correct on next mount.");
1523
Chandra Seetharamanadab0f62011-06-29 22:10:14 +00001524 error = xfs_log_sbcount(mp);
David Chinnere5720ee2008-04-10 12:21:18 +10001525 if (error)
Dave Chinner0b932cc2011-03-07 10:08:35 +11001526 xfs_warn(mp, "Unable to update superblock counters. "
David Chinnere5720ee2008-04-10 12:21:18 +10001527 "Freespace may not be correct on next mount.");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 xfs_unmountfs_writesb(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 xfs_unmountfs_wait(mp); /* wait for async bufs */
Christoph Hellwig21b699c2009-03-16 08:19:29 +01001530 xfs_log_unmount_write(mp);
1531 xfs_log_unmount(mp);
Christoph Hellwig27174202009-03-30 10:21:31 +02001532 xfs_uuid_unmount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533
Christoph Hellwig1550d0b2008-08-13 16:17:37 +10001534#if defined(DEBUG)
Christoph Hellwig0ce4cfd2007-08-30 17:20:53 +10001535 xfs_errortag_clearall(mp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536#endif
Christoph Hellwigff4f0382008-08-13 16:50:47 +10001537 xfs_free_perag(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538}
1539
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10001540STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541xfs_unmountfs_wait(xfs_mount_t *mp)
1542{
1543 if (mp->m_logdev_targp != mp->m_ddev_targp)
1544 xfs_wait_buftarg(mp->m_logdev_targp);
1545 if (mp->m_rtdev_targp)
1546 xfs_wait_buftarg(mp->m_rtdev_targp);
1547 xfs_wait_buftarg(mp->m_ddev_targp);
1548}
1549
1550int
David Chinner92821e22007-05-24 15:26:31 +10001551xfs_fs_writable(xfs_mount_t *mp)
1552{
Christoph Hellwigb267ce92007-08-30 17:21:30 +10001553 return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001554 (mp->m_flags & XFS_MOUNT_RDONLY));
David Chinner92821e22007-05-24 15:26:31 +10001555}
1556
1557/*
Alex Elderb2ce3972011-07-11 09:51:44 -05001558 * xfs_log_sbcount
1559 *
Chandra Seetharamanadab0f62011-06-29 22:10:14 +00001560 * Sync the superblock counters to disk.
Alex Elderb2ce3972011-07-11 09:51:44 -05001561 *
1562 * Note this code can be called during the process of freezing, so
Chandra Seetharamanadab0f62011-06-29 22:10:14 +00001563 * we may need to use the transaction allocator which does not
Alex Elderb2ce3972011-07-11 09:51:44 -05001564 * block when the transaction subsystem is in its frozen state.
David Chinner92821e22007-05-24 15:26:31 +10001565 */
1566int
Chandra Seetharamanadab0f62011-06-29 22:10:14 +00001567xfs_log_sbcount(xfs_mount_t *mp)
David Chinner92821e22007-05-24 15:26:31 +10001568{
1569 xfs_trans_t *tp;
1570 int error;
1571
1572 if (!xfs_fs_writable(mp))
1573 return 0;
1574
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10001575 xfs_icsb_sync_counters(mp, 0);
David Chinner92821e22007-05-24 15:26:31 +10001576
1577 /*
1578 * we don't need to do this if we are updating the superblock
1579 * counters on every modification.
1580 */
1581 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1582 return 0;
1583
Alex Elderb2ce3972011-07-11 09:51:44 -05001584 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
David Chinner92821e22007-05-24 15:26:31 +10001585 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1586 XFS_DEFAULT_LOG_COUNT);
1587 if (error) {
1588 xfs_trans_cancel(tp, 0);
1589 return error;
1590 }
1591
1592 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
Chandra Seetharamanadab0f62011-06-29 22:10:14 +00001593 xfs_trans_set_sync(tp);
David Chinnere5720ee2008-04-10 12:21:18 +10001594 error = xfs_trans_commit(tp, 0);
1595 return error;
David Chinner92821e22007-05-24 15:26:31 +10001596}
1597
1598int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599xfs_unmountfs_writesb(xfs_mount_t *mp)
1600{
1601 xfs_buf_t *sbp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 int error = 0;
1603
1604 /*
1605 * skip superblock write if fs is read-only, or
1606 * if we are doing a forced umount.
1607 */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001608 if (!((mp->m_flags & XFS_MOUNT_RDONLY) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 XFS_FORCED_SHUTDOWN(mp))) {
David Chinner8d280b92006-03-14 13:13:09 +11001610
David Chinner92821e22007-05-24 15:26:31 +10001611 sbp = xfs_getsb(mp, 0);
David Chinner8d280b92006-03-14 13:13:09 +11001612
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 XFS_BUF_UNDONE(sbp);
1614 XFS_BUF_UNREAD(sbp);
1615 XFS_BUF_UNDELAYWRITE(sbp);
1616 XFS_BUF_WRITE(sbp);
1617 XFS_BUF_UNASYNC(sbp);
Chandra Seetharaman49074c02011-07-22 23:40:40 +00001618 ASSERT(sbp->b_target == mp->m_ddev_targp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 xfsbdstrat(mp, sbp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00001620 error = xfs_buf_iowait(sbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 if (error)
1622 xfs_ioerror_alert("xfs_unmountfs_writesb",
1623 mp, sbp, XFS_BUF_ADDR(sbp));
David Chinner92821e22007-05-24 15:26:31 +10001624 xfs_buf_relse(sbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001626 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627}
1628
1629/*
1630 * xfs_mod_sb() can be used to copy arbitrary changes to the
1631 * in-core superblock into the superblock buffer to be logged.
1632 * It does not provide the higher level of locking that is
1633 * needed to protect the in-core superblock from concurrent
1634 * access.
1635 */
1636void
1637xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1638{
1639 xfs_buf_t *bp;
1640 int first;
1641 int last;
1642 xfs_mount_t *mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 xfs_sb_field_t f;
1644
1645 ASSERT(fields);
1646 if (!fields)
1647 return;
1648 mp = tp->t_mountp;
1649 bp = xfs_trans_getsb(tp, mp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 first = sizeof(xfs_sb_t);
1651 last = 0;
1652
1653 /* translate/copy */
1654
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +10001655 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656
1657 /* find modified range */
Dave Chinner587aa0f2010-01-20 12:04:53 +11001658 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1659 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1660 last = xfs_sb_info[f + 1].offset - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661
1662 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1663 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1664 first = xfs_sb_info[f].offset;
1665
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 xfs_trans_log_buf(tp, bp, first, last);
1667}
Yingping Lud210a282006-06-09 14:55:18 +10001668
Yingping Lud210a282006-06-09 14:55:18 +10001669
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670/*
1671 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1672 * a delta to a specified field in the in-core superblock. Simply
1673 * switch on the field indicated and apply the delta to that field.
1674 * Fields are not allowed to dip below zero, so if the delta would
1675 * do this do not apply it and return EINVAL.
1676 *
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001677 * The m_sb_lock must be held when this routine is called.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05001679STATIC int
David Chinner20f4ebf2007-02-10 18:36:10 +11001680xfs_mod_incore_sb_unlocked(
1681 xfs_mount_t *mp,
1682 xfs_sb_field_t field,
1683 int64_t delta,
1684 int rsvd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685{
1686 int scounter; /* short counter for 32 bit fields */
1687 long long lcounter; /* long counter for 64 bit fields */
1688 long long res_used, rem;
1689
1690 /*
1691 * With the in-core superblock spin lock held, switch
1692 * on the indicated field. Apply the delta to the
1693 * proper field. If the fields value would dip below
1694 * 0, then do not apply the delta and return EINVAL.
1695 */
1696 switch (field) {
1697 case XFS_SBS_ICOUNT:
1698 lcounter = (long long)mp->m_sb.sb_icount;
1699 lcounter += delta;
1700 if (lcounter < 0) {
1701 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001702 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 }
1704 mp->m_sb.sb_icount = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001705 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 case XFS_SBS_IFREE:
1707 lcounter = (long long)mp->m_sb.sb_ifree;
1708 lcounter += delta;
1709 if (lcounter < 0) {
1710 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001711 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 }
1713 mp->m_sb.sb_ifree = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001714 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 case XFS_SBS_FDBLOCKS:
David Chinner4be536d2006-09-07 14:26:50 +10001716 lcounter = (long long)
1717 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1719
1720 if (delta > 0) { /* Putting blocks back */
1721 if (res_used > delta) {
1722 mp->m_resblks_avail += delta;
1723 } else {
1724 rem = delta - res_used;
1725 mp->m_resblks_avail = mp->m_resblks;
1726 lcounter += rem;
1727 }
1728 } else { /* Taking blocks away */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 lcounter += delta;
Dave Chinner8babd8a2010-03-04 01:46:25 +00001730 if (lcounter >= 0) {
1731 mp->m_sb.sb_fdblocks = lcounter +
1732 XFS_ALLOC_SET_ASIDE(mp);
1733 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 }
Dave Chinner8babd8a2010-03-04 01:46:25 +00001735
1736 /*
1737 * We are out of blocks, use any available reserved
1738 * blocks if were allowed to.
1739 */
1740 if (!rsvd)
1741 return XFS_ERROR(ENOSPC);
1742
1743 lcounter = (long long)mp->m_resblks_avail + delta;
1744 if (lcounter >= 0) {
1745 mp->m_resblks_avail = lcounter;
1746 return 0;
1747 }
1748 printk_once(KERN_WARNING
1749 "Filesystem \"%s\": reserve blocks depleted! "
1750 "Consider increasing reserve pool size.",
1751 mp->m_fsname);
1752 return XFS_ERROR(ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 }
1754
David Chinner4be536d2006-09-07 14:26:50 +10001755 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
Jesper Juhl014c2542006-01-15 02:37:08 +01001756 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 case XFS_SBS_FREXTENTS:
1758 lcounter = (long long)mp->m_sb.sb_frextents;
1759 lcounter += delta;
1760 if (lcounter < 0) {
Jesper Juhl014c2542006-01-15 02:37:08 +01001761 return XFS_ERROR(ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 }
1763 mp->m_sb.sb_frextents = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001764 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 case XFS_SBS_DBLOCKS:
1766 lcounter = (long long)mp->m_sb.sb_dblocks;
1767 lcounter += delta;
1768 if (lcounter < 0) {
1769 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001770 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 }
1772 mp->m_sb.sb_dblocks = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001773 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 case XFS_SBS_AGCOUNT:
1775 scounter = mp->m_sb.sb_agcount;
1776 scounter += delta;
1777 if (scounter < 0) {
1778 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001779 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 }
1781 mp->m_sb.sb_agcount = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001782 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 case XFS_SBS_IMAX_PCT:
1784 scounter = mp->m_sb.sb_imax_pct;
1785 scounter += delta;
1786 if (scounter < 0) {
1787 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001788 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 }
1790 mp->m_sb.sb_imax_pct = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001791 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 case XFS_SBS_REXTSIZE:
1793 scounter = mp->m_sb.sb_rextsize;
1794 scounter += delta;
1795 if (scounter < 0) {
1796 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001797 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 }
1799 mp->m_sb.sb_rextsize = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001800 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 case XFS_SBS_RBMBLOCKS:
1802 scounter = mp->m_sb.sb_rbmblocks;
1803 scounter += delta;
1804 if (scounter < 0) {
1805 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001806 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 }
1808 mp->m_sb.sb_rbmblocks = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001809 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 case XFS_SBS_RBLOCKS:
1811 lcounter = (long long)mp->m_sb.sb_rblocks;
1812 lcounter += delta;
1813 if (lcounter < 0) {
1814 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001815 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 }
1817 mp->m_sb.sb_rblocks = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001818 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 case XFS_SBS_REXTENTS:
1820 lcounter = (long long)mp->m_sb.sb_rextents;
1821 lcounter += delta;
1822 if (lcounter < 0) {
1823 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001824 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 }
1826 mp->m_sb.sb_rextents = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001827 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 case XFS_SBS_REXTSLOG:
1829 scounter = mp->m_sb.sb_rextslog;
1830 scounter += delta;
1831 if (scounter < 0) {
1832 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001833 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 }
1835 mp->m_sb.sb_rextslog = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001836 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 default:
1838 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001839 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 }
1841}
1842
1843/*
1844 * xfs_mod_incore_sb() is used to change a field in the in-core
1845 * superblock structure by the specified delta. This modification
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001846 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 * routine to do the work.
1848 */
1849int
David Chinner20f4ebf2007-02-10 18:36:10 +11001850xfs_mod_incore_sb(
Christoph Hellwig96540c72010-09-30 02:25:55 +00001851 struct xfs_mount *mp,
1852 xfs_sb_field_t field,
1853 int64_t delta,
1854 int rsvd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855{
Christoph Hellwig96540c72010-09-30 02:25:55 +00001856 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857
David Chinner8d280b92006-03-14 13:13:09 +11001858#ifdef HAVE_PERCPU_SB
Christoph Hellwig96540c72010-09-30 02:25:55 +00001859 ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS);
David Chinner8d280b92006-03-14 13:13:09 +11001860#endif
Christoph Hellwig96540c72010-09-30 02:25:55 +00001861 spin_lock(&mp->m_sb_lock);
1862 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1863 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001864
Jesper Juhl014c2542006-01-15 02:37:08 +01001865 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866}
1867
1868/*
Christoph Hellwig1b040712010-09-30 02:25:56 +00001869 * Change more than one field in the in-core superblock structure at a time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 *
Christoph Hellwig1b040712010-09-30 02:25:56 +00001871 * The fields and changes to those fields are specified in the array of
1872 * xfs_mod_sb structures passed in. Either all of the specified deltas
1873 * will be applied or none of them will. If any modified field dips below 0,
1874 * then all modifications will be backed out and EINVAL will be returned.
1875 *
1876 * Note that this function may not be used for the superblock values that
1877 * are tracked with the in-memory per-cpu counters - a direct call to
1878 * xfs_icsb_modify_counters is required for these.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 */
1880int
Christoph Hellwig1b040712010-09-30 02:25:56 +00001881xfs_mod_incore_sb_batch(
1882 struct xfs_mount *mp,
1883 xfs_mod_sb_t *msb,
1884 uint nmsb,
1885 int rsvd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886{
David Sterba45c51b92011-04-13 22:03:28 +00001887 xfs_mod_sb_t *msbp;
Christoph Hellwig1b040712010-09-30 02:25:56 +00001888 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889
1890 /*
Christoph Hellwig1b040712010-09-30 02:25:56 +00001891 * Loop through the array of mod structures and apply each individually.
1892 * If any fail, then back out all those which have already been applied.
1893 * Do all of this within the scope of the m_sb_lock so that all of the
1894 * changes will be atomic.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001896 spin_lock(&mp->m_sb_lock);
David Sterba45c51b92011-04-13 22:03:28 +00001897 for (msbp = msb; msbp < (msb + nmsb); msbp++) {
Christoph Hellwig1b040712010-09-30 02:25:56 +00001898 ASSERT(msbp->msb_field < XFS_SBS_ICOUNT ||
1899 msbp->msb_field > XFS_SBS_FDBLOCKS);
David Chinner8d280b92006-03-14 13:13:09 +11001900
Christoph Hellwig1b040712010-09-30 02:25:56 +00001901 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1902 msbp->msb_delta, rsvd);
1903 if (error)
1904 goto unwind;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 }
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001906 spin_unlock(&mp->m_sb_lock);
Christoph Hellwig1b040712010-09-30 02:25:56 +00001907 return 0;
1908
1909unwind:
1910 while (--msbp >= msb) {
1911 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1912 -msbp->msb_delta, rsvd);
1913 ASSERT(error == 0);
1914 }
1915 spin_unlock(&mp->m_sb_lock);
1916 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917}
1918
1919/*
1920 * xfs_getsb() is called to obtain the buffer for the superblock.
1921 * The buffer is returned locked and read in from disk.
1922 * The buffer should be released with a call to xfs_brelse().
1923 *
1924 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1925 * the superblock buffer if it can be locked without sleeping.
1926 * If it can't then we'll return NULL.
1927 */
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001928struct xfs_buf *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929xfs_getsb(
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001930 struct xfs_mount *mp,
1931 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932{
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001933 struct xfs_buf *bp = mp->m_sb_bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001935 if (!xfs_buf_trylock(bp)) {
1936 if (flags & XBF_TRYLOCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 return NULL;
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001938 xfs_buf_lock(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 }
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001940
Chandra Seetharaman72790aa2011-07-22 23:40:04 +00001941 xfs_buf_hold(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 ASSERT(XFS_BUF_ISDONE(bp));
Jesper Juhl014c2542006-01-15 02:37:08 +01001943 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944}
1945
1946/*
1947 * Used to free the superblock along various error paths.
1948 */
1949void
1950xfs_freesb(
Dave Chinner26af6552010-09-22 10:47:20 +10001951 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952{
Dave Chinner26af6552010-09-22 10:47:20 +10001953 struct xfs_buf *bp = mp->m_sb_bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954
Dave Chinner26af6552010-09-22 10:47:20 +10001955 xfs_buf_lock(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 mp->m_sb_bp = NULL;
Dave Chinner26af6552010-09-22 10:47:20 +10001957 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958}
1959
1960/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 * Used to log changes to the superblock unit and width fields which could
Eric Sandeene6957ea2008-04-10 12:19:34 +10001962 * be altered by the mount options, as well as any potential sb_features2
1963 * fixup. Only the first superblock is updated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 */
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001965int
David Chinneree1c0902008-03-06 13:45:50 +11001966xfs_mount_log_sb(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 xfs_mount_t *mp,
1968 __int64_t fields)
1969{
1970 xfs_trans_t *tp;
David Chinnere5720ee2008-04-10 12:21:18 +10001971 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972
David Chinneree1c0902008-03-06 13:45:50 +11001973 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
David Chinner4b166de2008-05-20 11:30:27 +10001974 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1975 XFS_SB_VERSIONNUM));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976
1977 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
David Chinnere5720ee2008-04-10 12:21:18 +10001978 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1979 XFS_DEFAULT_LOG_COUNT);
1980 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 xfs_trans_cancel(tp, 0);
David Chinnere5720ee2008-04-10 12:21:18 +10001982 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 }
1984 xfs_mod_sb(tp, fields);
David Chinnere5720ee2008-04-10 12:21:18 +10001985 error = xfs_trans_commit(tp, 0);
1986 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987}
David Chinner8d280b92006-03-14 13:13:09 +11001988
Christoph Hellwigdda35b82010-02-15 09:44:46 +00001989/*
1990 * If the underlying (data/log/rt) device is readonly, there are some
1991 * operations that cannot proceed.
1992 */
1993int
1994xfs_dev_is_read_only(
1995 struct xfs_mount *mp,
1996 char *message)
1997{
1998 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1999 xfs_readonly_buftarg(mp->m_logdev_targp) ||
2000 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11002001 xfs_notice(mp, "%s required on read-only device.", message);
2002 xfs_notice(mp, "write access unavailable, cannot proceed.");
Christoph Hellwigdda35b82010-02-15 09:44:46 +00002003 return EROFS;
2004 }
2005 return 0;
2006}
David Chinner8d280b92006-03-14 13:13:09 +11002007
2008#ifdef HAVE_PERCPU_SB
2009/*
2010 * Per-cpu incore superblock counters
2011 *
2012 * Simple concept, difficult implementation
2013 *
2014 * Basically, replace the incore superblock counters with a distributed per cpu
2015 * counter for contended fields (e.g. free block count).
2016 *
2017 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
2018 * hence needs to be accurately read when we are running low on space. Hence
2019 * there is a method to enable and disable the per-cpu counters based on how
2020 * much "stuff" is available in them.
2021 *
2022 * Basically, a counter is enabled if there is enough free resource to justify
2023 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
2024 * ENOSPC), then we disable the counters to synchronise all callers and
2025 * re-distribute the available resources.
2026 *
2027 * If, once we redistributed the available resources, we still get a failure,
2028 * we disable the per-cpu counter and go through the slow path.
2029 *
2030 * The slow path is the current xfs_mod_incore_sb() function. This means that
Malcolm Parsons9da096f2009-03-29 09:55:42 +02002031 * when we disable a per-cpu counter, we need to drain its resources back to
David Chinner8d280b92006-03-14 13:13:09 +11002032 * the global superblock. We do this after disabling the counter to prevent
2033 * more threads from queueing up on the counter.
2034 *
2035 * Essentially, this means that we still need a lock in the fast path to enable
2036 * synchronisation between the global counters and the per-cpu counters. This
2037 * is not a problem because the lock will be local to a CPU almost all the time
2038 * and have little contention except when we get to ENOSPC conditions.
2039 *
2040 * Basically, this lock becomes a barrier that enables us to lock out the fast
2041 * path while we do things like enabling and disabling counters and
2042 * synchronising the counters.
2043 *
2044 * Locking rules:
2045 *
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002046 * 1. m_sb_lock before picking up per-cpu locks
David Chinner8d280b92006-03-14 13:13:09 +11002047 * 2. per-cpu locks always picked up via for_each_online_cpu() order
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002048 * 3. accurate counter sync requires m_sb_lock + per cpu locks
David Chinner8d280b92006-03-14 13:13:09 +11002049 * 4. modifying per-cpu counters requires holding per-cpu lock
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002050 * 5. modifying global counters requires holding m_sb_lock
2051 * 6. enabling or disabling a counter requires holding the m_sb_lock
David Chinner8d280b92006-03-14 13:13:09 +11002052 * and _none_ of the per-cpu locks.
2053 *
2054 * Disabled counters are only ever re-enabled by a balance operation
2055 * that results in more free resources per CPU than a given threshold.
2056 * To ensure counters don't remain disabled, they are rebalanced when
2057 * the global resource goes above a higher threshold (i.e. some hysteresis
2058 * is present to prevent thrashing).
David Chinner8d280b92006-03-14 13:13:09 +11002059 */
David Chinnere8234a62006-03-14 13:23:52 +11002060
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002061#ifdef CONFIG_HOTPLUG_CPU
David Chinnere8234a62006-03-14 13:23:52 +11002062/*
2063 * hot-plug CPU notifier support.
2064 *
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002065 * We need a notifier per filesystem as we need to be able to identify
2066 * the filesystem to balance the counters out. This is achieved by
2067 * having a notifier block embedded in the xfs_mount_t and doing pointer
2068 * magic to get the mount pointer from the notifier block address.
David Chinnere8234a62006-03-14 13:23:52 +11002069 */
2070STATIC int
2071xfs_icsb_cpu_notify(
2072 struct notifier_block *nfb,
2073 unsigned long action,
2074 void *hcpu)
2075{
2076 xfs_icsb_cnts_t *cntp;
2077 xfs_mount_t *mp;
David Chinnere8234a62006-03-14 13:23:52 +11002078
2079 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
2080 cntp = (xfs_icsb_cnts_t *)
2081 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
2082 switch (action) {
2083 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002084 case CPU_UP_PREPARE_FROZEN:
David Chinnere8234a62006-03-14 13:23:52 +11002085 /* Easy Case - initialize the area and locks, and
2086 * then rebalance when online does everything else for us. */
David Chinner01e1b692006-03-14 13:29:16 +11002087 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinnere8234a62006-03-14 13:23:52 +11002088 break;
2089 case CPU_ONLINE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002090 case CPU_ONLINE_FROZEN:
David Chinner03135cf2007-02-10 18:35:15 +11002091 xfs_icsb_lock(mp);
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002092 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2093 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2094 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
David Chinner03135cf2007-02-10 18:35:15 +11002095 xfs_icsb_unlock(mp);
David Chinnere8234a62006-03-14 13:23:52 +11002096 break;
2097 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002098 case CPU_DEAD_FROZEN:
David Chinnere8234a62006-03-14 13:23:52 +11002099 /* Disable all the counters, then fold the dead cpu's
2100 * count into the total on the global superblock and
2101 * re-enable the counters. */
David Chinner03135cf2007-02-10 18:35:15 +11002102 xfs_icsb_lock(mp);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002103 spin_lock(&mp->m_sb_lock);
David Chinnere8234a62006-03-14 13:23:52 +11002104 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
2105 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
2106 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
2107
2108 mp->m_sb.sb_icount += cntp->icsb_icount;
2109 mp->m_sb.sb_ifree += cntp->icsb_ifree;
2110 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
2111
David Chinner01e1b692006-03-14 13:29:16 +11002112 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinnere8234a62006-03-14 13:23:52 +11002113
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002114 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
2115 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
2116 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002117 spin_unlock(&mp->m_sb_lock);
David Chinner03135cf2007-02-10 18:35:15 +11002118 xfs_icsb_unlock(mp);
David Chinnere8234a62006-03-14 13:23:52 +11002119 break;
2120 }
2121
2122 return NOTIFY_OK;
2123}
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002124#endif /* CONFIG_HOTPLUG_CPU */
David Chinnere8234a62006-03-14 13:23:52 +11002125
David Chinner8d280b92006-03-14 13:13:09 +11002126int
2127xfs_icsb_init_counters(
2128 xfs_mount_t *mp)
2129{
2130 xfs_icsb_cnts_t *cntp;
2131 int i;
2132
2133 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2134 if (mp->m_sb_cnts == NULL)
2135 return -ENOMEM;
2136
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002137#ifdef CONFIG_HOTPLUG_CPU
David Chinnere8234a62006-03-14 13:23:52 +11002138 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2139 mp->m_icsb_notifier.priority = 0;
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002140 register_hotcpu_notifier(&mp->m_icsb_notifier);
2141#endif /* CONFIG_HOTPLUG_CPU */
David Chinnere8234a62006-03-14 13:23:52 +11002142
David Chinner8d280b92006-03-14 13:13:09 +11002143 for_each_online_cpu(i) {
2144 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11002145 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinner8d280b92006-03-14 13:13:09 +11002146 }
David Chinner20b64282007-02-10 18:35:09 +11002147
2148 mutex_init(&mp->m_icsb_mutex);
2149
David Chinner8d280b92006-03-14 13:13:09 +11002150 /*
2151 * start with all counters disabled so that the
2152 * initial balance kicks us off correctly
2153 */
2154 mp->m_icsb_counters = -1;
2155 return 0;
2156}
2157
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11002158void
2159xfs_icsb_reinit_counters(
2160 xfs_mount_t *mp)
2161{
2162 xfs_icsb_lock(mp);
2163 /*
2164 * start with all counters disabled so that the
2165 * initial balance kicks us off correctly
2166 */
2167 mp->m_icsb_counters = -1;
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002168 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2169 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2170 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11002171 xfs_icsb_unlock(mp);
2172}
2173
Christoph Hellwigc962fb72008-05-20 15:10:52 +10002174void
David Chinner8d280b92006-03-14 13:13:09 +11002175xfs_icsb_destroy_counters(
2176 xfs_mount_t *mp)
2177{
David Chinnere8234a62006-03-14 13:23:52 +11002178 if (mp->m_sb_cnts) {
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002179 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
David Chinner8d280b92006-03-14 13:13:09 +11002180 free_percpu(mp->m_sb_cnts);
David Chinnere8234a62006-03-14 13:23:52 +11002181 }
David Chinner03135cf2007-02-10 18:35:15 +11002182 mutex_destroy(&mp->m_icsb_mutex);
David Chinner8d280b92006-03-14 13:13:09 +11002183}
2184
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00002185STATIC void
David Chinner01e1b692006-03-14 13:29:16 +11002186xfs_icsb_lock_cntr(
2187 xfs_icsb_cnts_t *icsbp)
2188{
2189 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2190 ndelay(1000);
2191 }
2192}
2193
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00002194STATIC void
David Chinner01e1b692006-03-14 13:29:16 +11002195xfs_icsb_unlock_cntr(
2196 xfs_icsb_cnts_t *icsbp)
2197{
2198 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2199}
2200
David Chinner8d280b92006-03-14 13:13:09 +11002201
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00002202STATIC void
David Chinner8d280b92006-03-14 13:13:09 +11002203xfs_icsb_lock_all_counters(
2204 xfs_mount_t *mp)
2205{
2206 xfs_icsb_cnts_t *cntp;
2207 int i;
2208
2209 for_each_online_cpu(i) {
2210 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11002211 xfs_icsb_lock_cntr(cntp);
David Chinner8d280b92006-03-14 13:13:09 +11002212 }
2213}
2214
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00002215STATIC void
David Chinner8d280b92006-03-14 13:13:09 +11002216xfs_icsb_unlock_all_counters(
2217 xfs_mount_t *mp)
2218{
2219 xfs_icsb_cnts_t *cntp;
2220 int i;
2221
2222 for_each_online_cpu(i) {
2223 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11002224 xfs_icsb_unlock_cntr(cntp);
David Chinner8d280b92006-03-14 13:13:09 +11002225 }
2226}
2227
2228STATIC void
2229xfs_icsb_count(
2230 xfs_mount_t *mp,
2231 xfs_icsb_cnts_t *cnt,
2232 int flags)
2233{
2234 xfs_icsb_cnts_t *cntp;
2235 int i;
2236
2237 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2238
2239 if (!(flags & XFS_ICSB_LAZY_COUNT))
2240 xfs_icsb_lock_all_counters(mp);
2241
2242 for_each_online_cpu(i) {
2243 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2244 cnt->icsb_icount += cntp->icsb_icount;
2245 cnt->icsb_ifree += cntp->icsb_ifree;
2246 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2247 }
2248
2249 if (!(flags & XFS_ICSB_LAZY_COUNT))
2250 xfs_icsb_unlock_all_counters(mp);
2251}
2252
2253STATIC int
2254xfs_icsb_counter_disabled(
2255 xfs_mount_t *mp,
2256 xfs_sb_field_t field)
2257{
2258 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2259 return test_bit(field, &mp->m_icsb_counters);
2260}
2261
David Chinner36fbe6e2008-04-10 12:19:56 +10002262STATIC void
David Chinner8d280b92006-03-14 13:13:09 +11002263xfs_icsb_disable_counter(
2264 xfs_mount_t *mp,
2265 xfs_sb_field_t field)
2266{
2267 xfs_icsb_cnts_t cnt;
2268
2269 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2270
David Chinner20b64282007-02-10 18:35:09 +11002271 /*
2272 * If we are already disabled, then there is nothing to do
2273 * here. We check before locking all the counters to avoid
2274 * the expensive lock operation when being called in the
2275 * slow path and the counter is already disabled. This is
2276 * safe because the only time we set or clear this state is under
2277 * the m_icsb_mutex.
2278 */
2279 if (xfs_icsb_counter_disabled(mp, field))
David Chinner36fbe6e2008-04-10 12:19:56 +10002280 return;
David Chinner20b64282007-02-10 18:35:09 +11002281
David Chinner8d280b92006-03-14 13:13:09 +11002282 xfs_icsb_lock_all_counters(mp);
2283 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2284 /* drain back to superblock */
2285
Christoph Hellwigce461932008-04-22 17:34:50 +10002286 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
David Chinner8d280b92006-03-14 13:13:09 +11002287 switch(field) {
2288 case XFS_SBS_ICOUNT:
2289 mp->m_sb.sb_icount = cnt.icsb_icount;
2290 break;
2291 case XFS_SBS_IFREE:
2292 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2293 break;
2294 case XFS_SBS_FDBLOCKS:
2295 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2296 break;
2297 default:
2298 BUG();
2299 }
2300 }
2301
2302 xfs_icsb_unlock_all_counters(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002303}
2304
2305STATIC void
2306xfs_icsb_enable_counter(
2307 xfs_mount_t *mp,
2308 xfs_sb_field_t field,
2309 uint64_t count,
2310 uint64_t resid)
2311{
2312 xfs_icsb_cnts_t *cntp;
2313 int i;
2314
2315 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2316
2317 xfs_icsb_lock_all_counters(mp);
2318 for_each_online_cpu(i) {
2319 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2320 switch (field) {
2321 case XFS_SBS_ICOUNT:
2322 cntp->icsb_icount = count + resid;
2323 break;
2324 case XFS_SBS_IFREE:
2325 cntp->icsb_ifree = count + resid;
2326 break;
2327 case XFS_SBS_FDBLOCKS:
2328 cntp->icsb_fdblocks = count + resid;
2329 break;
2330 default:
2331 BUG();
2332 break;
2333 }
2334 resid = 0;
2335 }
2336 clear_bit(field, &mp->m_icsb_counters);
2337 xfs_icsb_unlock_all_counters(mp);
2338}
2339
David Chinnerdbcabad2007-02-10 18:36:17 +11002340void
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002341xfs_icsb_sync_counters_locked(
David Chinner8d280b92006-03-14 13:13:09 +11002342 xfs_mount_t *mp,
2343 int flags)
2344{
2345 xfs_icsb_cnts_t cnt;
David Chinner8d280b92006-03-14 13:13:09 +11002346
David Chinner8d280b92006-03-14 13:13:09 +11002347 xfs_icsb_count(mp, &cnt, flags);
2348
David Chinner8d280b92006-03-14 13:13:09 +11002349 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2350 mp->m_sb.sb_icount = cnt.icsb_icount;
2351 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2352 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2353 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2354 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
David Chinner8d280b92006-03-14 13:13:09 +11002355}
2356
2357/*
2358 * Accurate update of per-cpu counters to incore superblock
2359 */
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002360void
David Chinner8d280b92006-03-14 13:13:09 +11002361xfs_icsb_sync_counters(
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002362 xfs_mount_t *mp,
2363 int flags)
David Chinner8d280b92006-03-14 13:13:09 +11002364{
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002365 spin_lock(&mp->m_sb_lock);
2366 xfs_icsb_sync_counters_locked(mp, flags);
2367 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11002368}
2369
2370/*
2371 * Balance and enable/disable counters as necessary.
2372 *
David Chinner20b64282007-02-10 18:35:09 +11002373 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2374 * chosen to be the same number as single on disk allocation chunk per CPU, and
2375 * free blocks is something far enough zero that we aren't going thrash when we
2376 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2377 * prevent looping endlessly when xfs_alloc_space asks for more than will
2378 * be distributed to a single CPU but each CPU has enough blocks to be
2379 * reenabled.
2380 *
2381 * Note that we can be called when counters are already disabled.
2382 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2383 * prevent locking every per-cpu counter needlessly.
David Chinner8d280b92006-03-14 13:13:09 +11002384 */
David Chinner20b64282007-02-10 18:35:09 +11002385
2386#define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
David Chinner4be536d2006-09-07 14:26:50 +10002387#define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
David Chinner20b64282007-02-10 18:35:09 +11002388 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
David Chinner8d280b92006-03-14 13:13:09 +11002389STATIC void
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002390xfs_icsb_balance_counter_locked(
David Chinner8d280b92006-03-14 13:13:09 +11002391 xfs_mount_t *mp,
2392 xfs_sb_field_t field,
David Chinner20b64282007-02-10 18:35:09 +11002393 int min_per_cpu)
David Chinner8d280b92006-03-14 13:13:09 +11002394{
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002395 uint64_t count, resid;
David Chinner8d280b92006-03-14 13:13:09 +11002396 int weight = num_online_cpus();
David Chinner20b64282007-02-10 18:35:09 +11002397 uint64_t min = (uint64_t)min_per_cpu;
David Chinner8d280b92006-03-14 13:13:09 +11002398
David Chinner8d280b92006-03-14 13:13:09 +11002399 /* disable counter and sync counter */
2400 xfs_icsb_disable_counter(mp, field);
2401
2402 /* update counters - first CPU gets residual*/
2403 switch (field) {
2404 case XFS_SBS_ICOUNT:
2405 count = mp->m_sb.sb_icount;
2406 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002407 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002408 return;
David Chinner8d280b92006-03-14 13:13:09 +11002409 break;
2410 case XFS_SBS_IFREE:
2411 count = mp->m_sb.sb_ifree;
2412 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002413 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002414 return;
David Chinner8d280b92006-03-14 13:13:09 +11002415 break;
2416 case XFS_SBS_FDBLOCKS:
2417 count = mp->m_sb.sb_fdblocks;
2418 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002419 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002420 return;
David Chinner8d280b92006-03-14 13:13:09 +11002421 break;
2422 default:
2423 BUG();
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002424 count = resid = 0; /* quiet, gcc */
David Chinner8d280b92006-03-14 13:13:09 +11002425 break;
2426 }
2427
2428 xfs_icsb_enable_counter(mp, field, count, resid);
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002429}
2430
2431STATIC void
2432xfs_icsb_balance_counter(
2433 xfs_mount_t *mp,
2434 xfs_sb_field_t fields,
2435 int min_per_cpu)
2436{
2437 spin_lock(&mp->m_sb_lock);
2438 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2439 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11002440}
2441
Christoph Hellwig1b040712010-09-30 02:25:56 +00002442int
David Chinner20b64282007-02-10 18:35:09 +11002443xfs_icsb_modify_counters(
David Chinner8d280b92006-03-14 13:13:09 +11002444 xfs_mount_t *mp,
2445 xfs_sb_field_t field,
David Chinner20f4ebf2007-02-10 18:36:10 +11002446 int64_t delta,
David Chinner20b64282007-02-10 18:35:09 +11002447 int rsvd)
David Chinner8d280b92006-03-14 13:13:09 +11002448{
2449 xfs_icsb_cnts_t *icsbp;
2450 long long lcounter; /* long counter for 64 bit fields */
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002451 int ret = 0;
David Chinner8d280b92006-03-14 13:13:09 +11002452
David Chinner20b64282007-02-10 18:35:09 +11002453 might_sleep();
David Chinner8d280b92006-03-14 13:13:09 +11002454again:
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002455 preempt_disable();
2456 icsbp = this_cpu_ptr(mp->m_sb_cnts);
David Chinner20b64282007-02-10 18:35:09 +11002457
2458 /*
2459 * if the counter is disabled, go to slow path
2460 */
David Chinner8d280b92006-03-14 13:13:09 +11002461 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2462 goto slow_path;
David Chinner20b64282007-02-10 18:35:09 +11002463 xfs_icsb_lock_cntr(icsbp);
2464 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2465 xfs_icsb_unlock_cntr(icsbp);
2466 goto slow_path;
2467 }
David Chinner8d280b92006-03-14 13:13:09 +11002468
2469 switch (field) {
2470 case XFS_SBS_ICOUNT:
2471 lcounter = icsbp->icsb_icount;
2472 lcounter += delta;
2473 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002474 goto balance_counter;
David Chinner8d280b92006-03-14 13:13:09 +11002475 icsbp->icsb_icount = lcounter;
2476 break;
2477
2478 case XFS_SBS_IFREE:
2479 lcounter = icsbp->icsb_ifree;
2480 lcounter += delta;
2481 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002482 goto balance_counter;
David Chinner8d280b92006-03-14 13:13:09 +11002483 icsbp->icsb_ifree = lcounter;
2484 break;
2485
2486 case XFS_SBS_FDBLOCKS:
2487 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2488
David Chinner4be536d2006-09-07 14:26:50 +10002489 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002490 lcounter += delta;
2491 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002492 goto balance_counter;
David Chinner4be536d2006-09-07 14:26:50 +10002493 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002494 break;
2495 default:
2496 BUG();
2497 break;
2498 }
David Chinner01e1b692006-03-14 13:29:16 +11002499 xfs_icsb_unlock_cntr(icsbp);
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002500 preempt_enable();
David Chinner8d280b92006-03-14 13:13:09 +11002501 return 0;
2502
David Chinner8d280b92006-03-14 13:13:09 +11002503slow_path:
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002504 preempt_enable();
David Chinner20b64282007-02-10 18:35:09 +11002505
2506 /*
2507 * serialise with a mutex so we don't burn lots of cpu on
2508 * the superblock lock. We still need to hold the superblock
2509 * lock, however, when we modify the global structures.
2510 */
David Chinner03135cf2007-02-10 18:35:15 +11002511 xfs_icsb_lock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002512
2513 /*
2514 * Now running atomically.
2515 *
2516 * If the counter is enabled, someone has beaten us to rebalancing.
2517 * Drop the lock and try again in the fast path....
2518 */
2519 if (!(xfs_icsb_counter_disabled(mp, field))) {
David Chinner03135cf2007-02-10 18:35:15 +11002520 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002521 goto again;
2522 }
2523
2524 /*
2525 * The counter is currently disabled. Because we are
2526 * running atomically here, we know a rebalance cannot
2527 * be in progress. Hence we can go straight to operating
2528 * on the global superblock. We do not call xfs_mod_incore_sb()
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002529 * here even though we need to get the m_sb_lock. Doing so
David Chinner20b64282007-02-10 18:35:09 +11002530 * will cause us to re-enter this function and deadlock.
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002531 * Hence we get the m_sb_lock ourselves and then call
David Chinner20b64282007-02-10 18:35:09 +11002532 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2533 * directly on the global counters.
2534 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002535 spin_lock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11002536 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002537 spin_unlock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11002538
2539 /*
2540 * Now that we've modified the global superblock, we
2541 * may be able to re-enable the distributed counters
2542 * (e.g. lots of space just got freed). After that
2543 * we are done.
2544 */
2545 if (ret != ENOSPC)
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002546 xfs_icsb_balance_counter(mp, field, 0);
David Chinner03135cf2007-02-10 18:35:15 +11002547 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002548 return ret;
2549
2550balance_counter:
David Chinner01e1b692006-03-14 13:29:16 +11002551 xfs_icsb_unlock_cntr(icsbp);
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002552 preempt_enable();
David Chinner8d280b92006-03-14 13:13:09 +11002553
David Chinner20b64282007-02-10 18:35:09 +11002554 /*
2555 * We may have multiple threads here if multiple per-cpu
2556 * counters run dry at the same time. This will mean we can
2557 * do more balances than strictly necessary but it is not
2558 * the common slowpath case.
2559 */
David Chinner03135cf2007-02-10 18:35:15 +11002560 xfs_icsb_lock(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002561
David Chinner20b64282007-02-10 18:35:09 +11002562 /*
2563 * running atomically.
2564 *
2565 * This will leave the counter in the correct state for future
2566 * accesses. After the rebalance, we simply try again and our retry
2567 * will either succeed through the fast path or slow path without
2568 * another balance operation being required.
2569 */
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002570 xfs_icsb_balance_counter(mp, field, delta);
David Chinner03135cf2007-02-10 18:35:15 +11002571 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002572 goto again;
David Chinner8d280b92006-03-14 13:13:09 +11002573}
2574
David Chinner8d280b92006-03-14 13:13:09 +11002575#endif