blob: aeb9d72ebf6e55102bb4b7a98dbaa60264511165 [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 Chinner8d280b92006-03-14 13:13:09 +110055STATIC int xfs_icsb_modify_counters(xfs_mount_t *, xfs_sb_field_t,
David Chinner20f4ebf2007-02-10 18:36:10 +110056 int64_t, int);
David Chinner36fbe6e2008-04-10 12:19:56 +100057STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
David Chinner8d280b92006-03-14 13:13:09 +110058
59#else
60
Christoph Hellwig45af6c62008-04-22 17:34:44 +100061#define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
62#define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
David Chinner8d280b92006-03-14 13:13:09 +110063#define xfs_icsb_modify_counters(mp, a, b, c) do { } while (0)
David Chinner8d280b92006-03-14 13:13:09 +110064
65#endif
66
Christoph Hellwig1df84c92006-01-11 15:29:52 +110067static const struct {
David Chinner8d280b92006-03-14 13:13:09 +110068 short offset;
69 short type; /* 0 = integer
70 * 1 = binary / string (no translation)
71 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070072} xfs_sb_info[] = {
73 { offsetof(xfs_sb_t, sb_magicnum), 0 },
74 { offsetof(xfs_sb_t, sb_blocksize), 0 },
75 { offsetof(xfs_sb_t, sb_dblocks), 0 },
76 { offsetof(xfs_sb_t, sb_rblocks), 0 },
77 { offsetof(xfs_sb_t, sb_rextents), 0 },
78 { offsetof(xfs_sb_t, sb_uuid), 1 },
79 { offsetof(xfs_sb_t, sb_logstart), 0 },
80 { offsetof(xfs_sb_t, sb_rootino), 0 },
81 { offsetof(xfs_sb_t, sb_rbmino), 0 },
82 { offsetof(xfs_sb_t, sb_rsumino), 0 },
83 { offsetof(xfs_sb_t, sb_rextsize), 0 },
84 { offsetof(xfs_sb_t, sb_agblocks), 0 },
85 { offsetof(xfs_sb_t, sb_agcount), 0 },
86 { offsetof(xfs_sb_t, sb_rbmblocks), 0 },
87 { offsetof(xfs_sb_t, sb_logblocks), 0 },
88 { offsetof(xfs_sb_t, sb_versionnum), 0 },
89 { offsetof(xfs_sb_t, sb_sectsize), 0 },
90 { offsetof(xfs_sb_t, sb_inodesize), 0 },
91 { offsetof(xfs_sb_t, sb_inopblock), 0 },
92 { offsetof(xfs_sb_t, sb_fname[0]), 1 },
93 { offsetof(xfs_sb_t, sb_blocklog), 0 },
94 { offsetof(xfs_sb_t, sb_sectlog), 0 },
95 { offsetof(xfs_sb_t, sb_inodelog), 0 },
96 { offsetof(xfs_sb_t, sb_inopblog), 0 },
97 { offsetof(xfs_sb_t, sb_agblklog), 0 },
98 { offsetof(xfs_sb_t, sb_rextslog), 0 },
99 { offsetof(xfs_sb_t, sb_inprogress), 0 },
100 { offsetof(xfs_sb_t, sb_imax_pct), 0 },
101 { offsetof(xfs_sb_t, sb_icount), 0 },
102 { offsetof(xfs_sb_t, sb_ifree), 0 },
103 { offsetof(xfs_sb_t, sb_fdblocks), 0 },
104 { offsetof(xfs_sb_t, sb_frextents), 0 },
105 { offsetof(xfs_sb_t, sb_uquotino), 0 },
106 { offsetof(xfs_sb_t, sb_gquotino), 0 },
107 { offsetof(xfs_sb_t, sb_qflags), 0 },
108 { offsetof(xfs_sb_t, sb_flags), 0 },
109 { offsetof(xfs_sb_t, sb_shared_vn), 0 },
110 { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
111 { offsetof(xfs_sb_t, sb_unit), 0 },
112 { offsetof(xfs_sb_t, sb_width), 0 },
113 { offsetof(xfs_sb_t, sb_dirblklog), 0 },
114 { offsetof(xfs_sb_t, sb_logsectlog), 0 },
115 { offsetof(xfs_sb_t, sb_logsectsize),0 },
116 { offsetof(xfs_sb_t, sb_logsunit), 0 },
117 { offsetof(xfs_sb_t, sb_features2), 0 },
David Chinneree1c0902008-03-06 13:45:50 +1100118 { offsetof(xfs_sb_t, sb_bad_features2), 0 },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 { sizeof(xfs_sb_t), 0 }
120};
121
Christoph Hellwig27174202009-03-30 10:21:31 +0200122static DEFINE_MUTEX(xfs_uuid_table_mutex);
123static int xfs_uuid_table_size;
124static uuid_t *xfs_uuid_table;
125
126/*
127 * See if the UUID is unique among mounted XFS filesystems.
128 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
129 */
130STATIC int
131xfs_uuid_mount(
132 struct xfs_mount *mp)
133{
134 uuid_t *uuid = &mp->m_sb.sb_uuid;
135 int hole, i;
136
137 if (mp->m_flags & XFS_MOUNT_NOUUID)
138 return 0;
139
140 if (uuid_is_nil(uuid)) {
141 cmn_err(CE_WARN,
142 "XFS: Filesystem %s has nil UUID - can't mount",
143 mp->m_fsname);
144 return XFS_ERROR(EINVAL);
145 }
146
147 mutex_lock(&xfs_uuid_table_mutex);
148 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
149 if (uuid_is_nil(&xfs_uuid_table[i])) {
150 hole = i;
151 continue;
152 }
153 if (uuid_equal(uuid, &xfs_uuid_table[i]))
154 goto out_duplicate;
155 }
156
157 if (hole < 0) {
158 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
159 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
160 xfs_uuid_table_size * sizeof(*xfs_uuid_table),
161 KM_SLEEP);
162 hole = xfs_uuid_table_size++;
163 }
164 xfs_uuid_table[hole] = *uuid;
165 mutex_unlock(&xfs_uuid_table_mutex);
166
167 return 0;
168
169 out_duplicate:
170 mutex_unlock(&xfs_uuid_table_mutex);
171 cmn_err(CE_WARN, "XFS: Filesystem %s has duplicate UUID - can't mount",
172 mp->m_fsname);
173 return XFS_ERROR(EINVAL);
174}
175
176STATIC void
177xfs_uuid_unmount(
178 struct xfs_mount *mp)
179{
180 uuid_t *uuid = &mp->m_sb.sb_uuid;
181 int i;
182
183 if (mp->m_flags & XFS_MOUNT_NOUUID)
184 return;
185
186 mutex_lock(&xfs_uuid_table_mutex);
187 for (i = 0; i < xfs_uuid_table_size; i++) {
188 if (uuid_is_nil(&xfs_uuid_table[i]))
189 continue;
190 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
191 continue;
192 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
193 break;
194 }
195 ASSERT(i < xfs_uuid_table_size);
196 mutex_unlock(&xfs_uuid_table_mutex);
197}
198
199
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200/*
Dave Chinner0fa800f2010-01-11 11:47:46 +0000201 * Reference counting access wrappers to the perag structures.
202 */
203struct xfs_perag *
204xfs_perag_get(struct xfs_mount *mp, xfs_agnumber_t agno)
205{
206 struct xfs_perag *pag;
207 int ref = 0;
208
209 spin_lock(&mp->m_perag_lock);
210 pag = radix_tree_lookup(&mp->m_perag_tree, agno);
211 if (pag) {
212 ASSERT(atomic_read(&pag->pag_ref) >= 0);
213 /* catch leaks in the positive direction during testing */
214 ASSERT(atomic_read(&pag->pag_ref) < 1000);
215 ref = atomic_inc_return(&pag->pag_ref);
216 }
217 spin_unlock(&mp->m_perag_lock);
218 trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
219 return pag;
220}
221
222void
223xfs_perag_put(struct xfs_perag *pag)
224{
225 int ref;
226
227 ASSERT(atomic_read(&pag->pag_ref) > 0);
228 ref = atomic_dec_return(&pag->pag_ref);
229 trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
230}
231
232/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 * Free up the resources associated with a mount structure. Assume that
234 * the structure was initially zeroed, so we can tell which fields got
235 * initialized.
236 */
Christoph Hellwigc962fb72008-05-20 15:10:52 +1000237STATIC void
Christoph Hellwigff4f0382008-08-13 16:50:47 +1000238xfs_free_perag(
Christoph Hellwig745f6912007-08-30 17:20:39 +1000239 xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240{
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000241 xfs_agnumber_t agno;
242 struct xfs_perag *pag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000244 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
245 spin_lock(&mp->m_perag_lock);
246 pag = radix_tree_delete(&mp->m_perag_tree, agno);
Dave Chinnere57336f2010-01-11 11:47:49 +0000247 ASSERT(pag);
Dave Chinneraed3bb92010-01-11 11:47:45 +0000248 ASSERT(atomic_read(&pag->pag_ref) == 0);
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000249 spin_unlock(&mp->m_perag_lock);
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000250 kmem_free(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252}
253
Nathan Scott4cc929e2007-05-14 18:24:02 +1000254/*
255 * Check size of device based on the (data/realtime) block count.
256 * Note: this check is used by the growfs code as well as mount.
257 */
258int
259xfs_sb_validate_fsb_count(
260 xfs_sb_t *sbp,
261 __uint64_t nblocks)
262{
263 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
264 ASSERT(sbp->sb_blocklog >= BBSHIFT);
265
266#if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
267 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
Eric Sandeen657a4cf2010-04-30 03:42:49 +0000268 return EFBIG;
Nathan Scott4cc929e2007-05-14 18:24:02 +1000269#else /* Limited by UINT_MAX of sectors */
270 if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
Eric Sandeen657a4cf2010-04-30 03:42:49 +0000271 return EFBIG;
Nathan Scott4cc929e2007-05-14 18:24:02 +1000272#endif
273 return 0;
274}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275
276/*
277 * Check the validity of the SB found.
278 */
279STATIC int
280xfs_mount_validate_sb(
281 xfs_mount_t *mp,
Nathan Scott764d1f82006-03-31 13:04:17 +1000282 xfs_sb_t *sbp,
283 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284{
285 /*
286 * If the log device and data device have the
287 * same device number, the log is internal.
288 * Consequently, the sb_logstart should be non-zero. If
289 * we have a zero sb_logstart in this case, we may be trying to mount
290 * a volume filesystem in a non-volume manner.
291 */
292 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000293 xfs_fs_mount_cmn_err(flags, "bad magic number");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 return XFS_ERROR(EWRONGFS);
295 }
296
Eric Sandeen62118702008-03-06 13:44:28 +1100297 if (!xfs_sb_good_version(sbp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000298 xfs_fs_mount_cmn_err(flags, "bad version");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 return XFS_ERROR(EWRONGFS);
300 }
301
302 if (unlikely(
303 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000304 xfs_fs_mount_cmn_err(flags,
305 "filesystem is marked as having an external log; "
306 "specify logdev on the\nmount command line.");
307 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 }
309
310 if (unlikely(
311 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000312 xfs_fs_mount_cmn_err(flags,
313 "filesystem is marked as having an internal log; "
314 "do not specify logdev on\nthe mount command line.");
315 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 }
317
318 /*
319 * More sanity checking. These were stolen directly from
320 * xfs_repair.
321 */
322 if (unlikely(
323 sbp->sb_agcount <= 0 ||
324 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
325 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
326 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
327 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
Olaf Weber2ac00af2009-04-17 16:12:45 -0500328 sbp->sb_sectsize != (1 << sbp->sb_sectlog) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
330 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
331 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
332 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
Olaf Weber2ac00af2009-04-17 16:12:45 -0500333 sbp->sb_blocksize != (1 << sbp->sb_blocklog) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
335 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
Nathan Scott9f989c92006-03-14 13:29:32 +1100336 sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
337 sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
Olaf Weber2ac00af2009-04-17 16:12:45 -0500338 sbp->sb_inodesize != (1 << sbp->sb_inodelog) ||
Nathan Scott9f989c92006-03-14 13:29:32 +1100339 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
341 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
Nathan Scotte50bd162006-04-11 15:10:45 +1000342 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000343 xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 return XFS_ERROR(EFSCORRUPTED);
345 }
346
347 /*
348 * Sanity check AG count, size fields against data size field
349 */
350 if (unlikely(
351 sbp->sb_dblocks == 0 ||
352 sbp->sb_dblocks >
353 (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
354 sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
355 sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000356 xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 return XFS_ERROR(EFSCORRUPTED);
358 }
359
Lachlan McIlroy2edbddd2008-06-27 13:34:34 +1000360 /*
361 * Until this is fixed only page-sized or smaller data blocks work.
362 */
363 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
364 xfs_fs_mount_cmn_err(flags,
365 "file system with blocksize %d bytes",
366 sbp->sb_blocksize);
367 xfs_fs_mount_cmn_err(flags,
368 "only pagesize (%ld) or less will currently work.",
369 PAGE_SIZE);
370 return XFS_ERROR(ENOSYS);
371 }
372
Christoph Hellwig1a5902c2009-03-29 19:26:46 +0200373 /*
374 * Currently only very few inode sizes are supported.
375 */
376 switch (sbp->sb_inodesize) {
377 case 256:
378 case 512:
379 case 1024:
380 case 2048:
381 break;
382 default:
383 xfs_fs_mount_cmn_err(flags,
384 "inode size of %d bytes not supported",
385 sbp->sb_inodesize);
386 return XFS_ERROR(ENOSYS);
387 }
388
Nathan Scott4cc929e2007-05-14 18:24:02 +1000389 if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
390 xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000391 xfs_fs_mount_cmn_err(flags,
392 "file system too large to be mounted on this system.");
Eric Sandeen657a4cf2010-04-30 03:42:49 +0000393 return XFS_ERROR(EFBIG);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 }
395
396 if (unlikely(sbp->sb_inprogress)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000397 xfs_fs_mount_cmn_err(flags, "file system busy");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 return XFS_ERROR(EFSCORRUPTED);
399 }
400
401 /*
Nathan Scottde206142005-05-05 13:24:13 -0700402 * Version 1 directory format has never worked on Linux.
403 */
Eric Sandeen62118702008-03-06 13:44:28 +1100404 if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000405 xfs_fs_mount_cmn_err(flags,
406 "file system using version 1 directory format");
Nathan Scottde206142005-05-05 13:24:13 -0700407 return XFS_ERROR(ENOSYS);
408 }
409
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 return 0;
411}
412
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000413int
Nathan Scottc11e2c32005-11-02 15:11:45 +1100414xfs_initialize_perag(
Nathan Scottc11e2c32005-11-02 15:11:45 +1100415 xfs_mount_t *mp,
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000416 xfs_agnumber_t agcount,
417 xfs_agnumber_t *maxagi)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418{
419 xfs_agnumber_t index, max_metadata;
Dave Chinner8b26c582010-01-11 11:47:48 +0000420 xfs_agnumber_t first_initialised = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 xfs_perag_t *pag;
422 xfs_agino_t agino;
423 xfs_ino_t ino;
424 xfs_sb_t *sbp = &mp->m_sb;
Dave Chinner8b26c582010-01-11 11:47:48 +0000425 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000427 /*
428 * Walk the current per-ag tree so we don't try to initialise AGs
429 * that already exist (growfs case). Allocate and insert all the
430 * AGs we don't find ready for initialisation.
431 */
432 for (index = 0; index < agcount; index++) {
433 pag = xfs_perag_get(mp, index);
434 if (pag) {
435 xfs_perag_put(pag);
436 continue;
437 }
Dave Chinner8b26c582010-01-11 11:47:48 +0000438 if (!first_initialised)
439 first_initialised = index;
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000440
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000441 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
442 if (!pag)
Dave Chinner8b26c582010-01-11 11:47:48 +0000443 goto out_unwind;
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000444 pag->pag_agno = index;
445 pag->pag_mount = mp;
446 rwlock_init(&pag->pag_ici_lock);
447 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
448
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000449 if (radix_tree_preload(GFP_NOFS))
Dave Chinner8b26c582010-01-11 11:47:48 +0000450 goto out_unwind;
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000451
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000452 spin_lock(&mp->m_perag_lock);
453 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
454 BUG();
455 spin_unlock(&mp->m_perag_lock);
Dave Chinner8b26c582010-01-11 11:47:48 +0000456 radix_tree_preload_end();
457 error = -EEXIST;
458 goto out_unwind;
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000459 }
460 spin_unlock(&mp->m_perag_lock);
461 radix_tree_preload_end();
462 }
463
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000464 /*
465 * If we mount with the inode64 option, or no inode overflows
466 * the legacy 32-bit address space clear the inode32 option.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 */
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000468 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
469 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000471 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
472 mp->m_flags |= XFS_MOUNT_32BITINODES;
473 else
474 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
475
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000477 /*
478 * Calculate how much should be reserved for inodes to meet
479 * the max inode percentage.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 */
481 if (mp->m_maxicount) {
482 __uint64_t icount;
483
484 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
485 do_div(icount, 100);
486 icount += sbp->sb_agblocks - 1;
Eric Sandeena749ee82005-11-02 15:13:42 +1100487 do_div(icount, sbp->sb_agblocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 max_metadata = icount;
489 } else {
490 max_metadata = agcount;
491 }
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000492
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 for (index = 0; index < agcount; index++) {
494 ino = XFS_AGINO_TO_INO(mp, index, agino);
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000495 if (ino > XFS_MAXINUMBER_32) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 index++;
497 break;
498 }
499
Dave Chinner44b56e02010-01-11 11:47:43 +0000500 pag = xfs_perag_get(mp, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 pag->pagi_inodeok = 1;
502 if (index < max_metadata)
503 pag->pagf_metadata = 1;
Dave Chinner44b56e02010-01-11 11:47:43 +0000504 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 }
506 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 for (index = 0; index < agcount; index++) {
Dave Chinner44b56e02010-01-11 11:47:43 +0000508 pag = xfs_perag_get(mp, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 pag->pagi_inodeok = 1;
Dave Chinner44b56e02010-01-11 11:47:43 +0000510 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 }
512 }
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000513
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000514 if (maxagi)
515 *maxagi = index;
516 return 0;
Dave Chinner8b26c582010-01-11 11:47:48 +0000517
518out_unwind:
519 kmem_free(pag);
520 for (; index > first_initialised; index--) {
521 pag = radix_tree_delete(&mp->m_perag_tree, index);
522 kmem_free(pag);
523 }
524 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525}
526
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000527void
528xfs_sb_from_disk(
529 xfs_sb_t *to,
530 xfs_dsb_t *from)
531{
532 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
533 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
534 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
535 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
536 to->sb_rextents = be64_to_cpu(from->sb_rextents);
537 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
538 to->sb_logstart = be64_to_cpu(from->sb_logstart);
539 to->sb_rootino = be64_to_cpu(from->sb_rootino);
540 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
541 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
542 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
543 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
544 to->sb_agcount = be32_to_cpu(from->sb_agcount);
545 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
546 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
547 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
548 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
549 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
550 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
551 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
552 to->sb_blocklog = from->sb_blocklog;
553 to->sb_sectlog = from->sb_sectlog;
554 to->sb_inodelog = from->sb_inodelog;
555 to->sb_inopblog = from->sb_inopblog;
556 to->sb_agblklog = from->sb_agblklog;
557 to->sb_rextslog = from->sb_rextslog;
558 to->sb_inprogress = from->sb_inprogress;
559 to->sb_imax_pct = from->sb_imax_pct;
560 to->sb_icount = be64_to_cpu(from->sb_icount);
561 to->sb_ifree = be64_to_cpu(from->sb_ifree);
562 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
563 to->sb_frextents = be64_to_cpu(from->sb_frextents);
564 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
565 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
566 to->sb_qflags = be16_to_cpu(from->sb_qflags);
567 to->sb_flags = from->sb_flags;
568 to->sb_shared_vn = from->sb_shared_vn;
569 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
570 to->sb_unit = be32_to_cpu(from->sb_unit);
571 to->sb_width = be32_to_cpu(from->sb_width);
572 to->sb_dirblklog = from->sb_dirblklog;
573 to->sb_logsectlog = from->sb_logsectlog;
574 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
575 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
576 to->sb_features2 = be32_to_cpu(from->sb_features2);
David Chinneree1c0902008-03-06 13:45:50 +1100577 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000578}
579
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580/*
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000581 * Copy in core superblock to ondisk one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 *
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000583 * The fields argument is mask of superblock fields to copy.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 */
585void
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000586xfs_sb_to_disk(
587 xfs_dsb_t *to,
588 xfs_sb_t *from,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 __int64_t fields)
590{
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000591 xfs_caddr_t to_ptr = (xfs_caddr_t)to;
592 xfs_caddr_t from_ptr = (xfs_caddr_t)from;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 xfs_sb_field_t f;
594 int first;
595 int size;
596
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 ASSERT(fields);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 if (!fields)
599 return;
600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 while (fields) {
602 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
603 first = xfs_sb_info[f].offset;
604 size = xfs_sb_info[f + 1].offset - first;
605
606 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
607
608 if (size == 1 || xfs_sb_info[f].type == 1) {
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000609 memcpy(to_ptr + first, from_ptr + first, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 } else {
611 switch (size) {
612 case 2:
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000613 *(__be16 *)(to_ptr + first) =
614 cpu_to_be16(*(__u16 *)(from_ptr + first));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 break;
616 case 4:
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000617 *(__be32 *)(to_ptr + first) =
618 cpu_to_be32(*(__u32 *)(from_ptr + first));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 break;
620 case 8:
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000621 *(__be64 *)(to_ptr + first) =
622 cpu_to_be64(*(__u64 *)(from_ptr + first));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 break;
624 default:
625 ASSERT(0);
626 }
627 }
628
629 fields &= ~(1LL << f);
630 }
631}
632
633/*
634 * xfs_readsb
635 *
636 * Does the initial read of the superblock.
637 */
638int
Nathan Scott764d1f82006-03-31 13:04:17 +1000639xfs_readsb(xfs_mount_t *mp, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640{
641 unsigned int sector_size;
642 unsigned int extra_flags;
643 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 int error;
645
646 ASSERT(mp->m_sb_bp == NULL);
647 ASSERT(mp->m_ddev_targp != NULL);
648
649 /*
650 * Allocate a (locked) buffer to hold the superblock.
651 * This will be kept around at all times to optimize
652 * access to the superblock.
653 */
654 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000655 extra_flags = XBF_LOCK | XBF_FS_MANAGED | XBF_MAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Christoph Hellwig6ad112b2009-11-24 18:02:23 +0000657 bp = xfs_buf_read(mp->m_ddev_targp, XFS_SB_DADDR, BTOBB(sector_size),
658 extra_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 if (!bp || XFS_BUF_ISERROR(bp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000660 xfs_fs_mount_cmn_err(flags, "SB read failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
662 goto fail;
663 }
664 ASSERT(XFS_BUF_ISBUSY(bp));
665 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
666
667 /*
668 * Initialize the mount structure from the superblock.
669 * But first do some basic consistency checking.
670 */
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000671 xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672
Nathan Scott764d1f82006-03-31 13:04:17 +1000673 error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 if (error) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000675 xfs_fs_mount_cmn_err(flags, "SB validate failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 goto fail;
677 }
678
679 /*
680 * We must be able to do sector-sized and sector-aligned IO.
681 */
682 if (sector_size > mp->m_sb.sb_sectsize) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000683 xfs_fs_mount_cmn_err(flags,
684 "device supports only %u byte sectors (not %u)",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 sector_size, mp->m_sb.sb_sectsize);
686 error = ENOSYS;
687 goto fail;
688 }
689
690 /*
691 * If device sector size is smaller than the superblock size,
692 * re-read the superblock so the buffer is correctly sized.
693 */
694 if (sector_size < mp->m_sb.sb_sectsize) {
695 XFS_BUF_UNMANAGE(bp);
696 xfs_buf_relse(bp);
697 sector_size = mp->m_sb.sb_sectsize;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +0000698 bp = xfs_buf_read(mp->m_ddev_targp, XFS_SB_DADDR,
699 BTOBB(sector_size), extra_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 if (!bp || XFS_BUF_ISERROR(bp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000701 xfs_fs_mount_cmn_err(flags, "SB re-read failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
703 goto fail;
704 }
705 ASSERT(XFS_BUF_ISBUSY(bp));
706 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
707 }
708
Lachlan McIlroy5478eea2007-02-10 18:36:29 +1100709 /* Initialize per-cpu counters */
710 xfs_icsb_reinit_counters(mp);
David Chinner8d280b92006-03-14 13:13:09 +1100711
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 mp->m_sb_bp = bp;
713 xfs_buf_relse(bp);
714 ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
715 return 0;
716
717 fail:
718 if (bp) {
719 XFS_BUF_UNMANAGE(bp);
720 xfs_buf_relse(bp);
721 }
722 return error;
723}
724
725
726/*
727 * xfs_mount_common
728 *
729 * Mount initialization code establishing various mount
730 * fields from the superblock associated with the given
731 * mount structure
732 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000733STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
735{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 mp->m_agfrotor = mp->m_agirotor = 0;
Eric Sandeen007c61c2007-10-11 17:43:56 +1000737 spin_lock_init(&mp->m_agirotor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 mp->m_maxagi = mp->m_sb.sb_agcount;
739 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
740 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
741 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
742 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
743 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 mp->m_blockmask = sbp->sb_blocksize - 1;
745 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
746 mp->m_blockwmask = mp->m_blockwsize - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
Christoph Hellwig60197e82008-10-30 17:11:19 +1100748 mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
749 mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
750 mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
751 mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
752
753 mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
754 mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
755 mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
756 mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
757
758 mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
759 mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
760 mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
761 mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762
763 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
764 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
765 sbp->sb_inopblock);
766 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
767}
David Chinner92821e22007-05-24 15:26:31 +1000768
769/*
770 * xfs_initialize_perag_data
771 *
772 * Read in each per-ag structure so we can count up the number of
773 * allocated inodes, free inodes and used filesystem blocks as this
774 * information is no longer persistent in the superblock. Once we have
775 * this information, write it into the in-core superblock structure.
776 */
777STATIC int
778xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
779{
780 xfs_agnumber_t index;
781 xfs_perag_t *pag;
782 xfs_sb_t *sbp = &mp->m_sb;
783 uint64_t ifree = 0;
784 uint64_t ialloc = 0;
785 uint64_t bfree = 0;
786 uint64_t bfreelst = 0;
787 uint64_t btree = 0;
788 int error;
David Chinner92821e22007-05-24 15:26:31 +1000789
790 for (index = 0; index < agcount; index++) {
791 /*
792 * read the agf, then the agi. This gets us
Malcolm Parsons9da096f2009-03-29 09:55:42 +0200793 * all the information we need and populates the
David Chinner92821e22007-05-24 15:26:31 +1000794 * per-ag structures for us.
795 */
796 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
797 if (error)
798 return error;
799
800 error = xfs_ialloc_pagi_init(mp, NULL, index);
801 if (error)
802 return error;
Dave Chinner44b56e02010-01-11 11:47:43 +0000803 pag = xfs_perag_get(mp, index);
David Chinner92821e22007-05-24 15:26:31 +1000804 ifree += pag->pagi_freecount;
805 ialloc += pag->pagi_count;
806 bfree += pag->pagf_freeblks;
807 bfreelst += pag->pagf_flcount;
808 btree += pag->pagf_btreeblks;
Dave Chinner44b56e02010-01-11 11:47:43 +0000809 xfs_perag_put(pag);
David Chinner92821e22007-05-24 15:26:31 +1000810 }
811 /*
812 * Overwrite incore superblock counters with just-read data
813 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +1000814 spin_lock(&mp->m_sb_lock);
David Chinner92821e22007-05-24 15:26:31 +1000815 sbp->sb_ifree = ifree;
816 sbp->sb_icount = ialloc;
817 sbp->sb_fdblocks = bfree + bfreelst + btree;
Eric Sandeen3685c2a2007-10-11 17:42:32 +1000818 spin_unlock(&mp->m_sb_lock);
David Chinner92821e22007-05-24 15:26:31 +1000819
820 /* Fixup the per-cpu counters as well. */
821 xfs_icsb_reinit_counters(mp);
822
823 return 0;
824}
825
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826/*
Eric Sandeen0771fb42007-10-12 11:03:40 +1000827 * Update alignment values based on mount options and sb values
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 */
Eric Sandeen0771fb42007-10-12 11:03:40 +1000829STATIC int
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100830xfs_update_alignment(xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 xfs_sb_t *sbp = &(mp->m_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833
Christoph Hellwig42490232008-08-13 16:49:32 +1000834 if (mp->m_dalign) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 /*
836 * If stripe unit and stripe width are not multiples
837 * of the fs blocksize turn off alignment.
838 */
839 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
840 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
841 if (mp->m_flags & XFS_MOUNT_RETERR) {
842 cmn_err(CE_WARN,
843 "XFS: alignment check 1 failed");
Eric Sandeen0771fb42007-10-12 11:03:40 +1000844 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 }
846 mp->m_dalign = mp->m_swidth = 0;
847 } else {
848 /*
849 * Convert the stripe unit and width to FSBs.
850 */
851 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
852 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
853 if (mp->m_flags & XFS_MOUNT_RETERR) {
Eric Sandeen0771fb42007-10-12 11:03:40 +1000854 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 }
856 xfs_fs_cmn_err(CE_WARN, mp,
857"stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
858 mp->m_dalign, mp->m_swidth,
859 sbp->sb_agblocks);
860
861 mp->m_dalign = 0;
862 mp->m_swidth = 0;
863 } else if (mp->m_dalign) {
864 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
865 } else {
866 if (mp->m_flags & XFS_MOUNT_RETERR) {
867 xfs_fs_cmn_err(CE_WARN, mp,
868"stripe alignment turned off: sunit(%d) less than bsize(%d)",
869 mp->m_dalign,
870 mp->m_blockmask +1);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000871 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 }
873 mp->m_swidth = 0;
874 }
875 }
876
877 /*
878 * Update superblock with new values
879 * and log changes
880 */
Eric Sandeen62118702008-03-06 13:44:28 +1100881 if (xfs_sb_version_hasdalign(sbp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 if (sbp->sb_unit != mp->m_dalign) {
883 sbp->sb_unit = mp->m_dalign;
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100884 mp->m_update_flags |= XFS_SB_UNIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 }
886 if (sbp->sb_width != mp->m_swidth) {
887 sbp->sb_width = mp->m_swidth;
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100888 mp->m_update_flags |= XFS_SB_WIDTH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 }
890 }
891 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
Eric Sandeen62118702008-03-06 13:44:28 +1100892 xfs_sb_version_hasdalign(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 mp->m_dalign = sbp->sb_unit;
894 mp->m_swidth = sbp->sb_width;
895 }
896
Eric Sandeen0771fb42007-10-12 11:03:40 +1000897 return 0;
898}
899
900/*
901 * Set the maximum inode count for this filesystem
902 */
903STATIC void
904xfs_set_maxicount(xfs_mount_t *mp)
905{
906 xfs_sb_t *sbp = &(mp->m_sb);
907 __uint64_t icount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
909 if (sbp->sb_imax_pct) {
Eric Sandeen0771fb42007-10-12 11:03:40 +1000910 /*
911 * Make sure the maximum inode count is a multiple
912 * of the units we allocate inodes in.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
915 do_div(icount, 100);
916 do_div(icount, mp->m_ialloc_blks);
917 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
918 sbp->sb_inopblog;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000919 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 mp->m_maxicount = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 }
Eric Sandeen0771fb42007-10-12 11:03:40 +1000922}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923
Eric Sandeen0771fb42007-10-12 11:03:40 +1000924/*
925 * Set the default minimum read and write sizes unless
926 * already specified in a mount option.
927 * We use smaller I/O sizes when the file system
928 * is being used for NFS service (wsync mount option).
929 */
930STATIC void
931xfs_set_rw_sizes(xfs_mount_t *mp)
932{
933 xfs_sb_t *sbp = &(mp->m_sb);
934 int readio_log, writeio_log;
935
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
937 if (mp->m_flags & XFS_MOUNT_WSYNC) {
938 readio_log = XFS_WSYNC_READIO_LOG;
939 writeio_log = XFS_WSYNC_WRITEIO_LOG;
940 } else {
941 readio_log = XFS_READIO_LOG_LARGE;
942 writeio_log = XFS_WRITEIO_LOG_LARGE;
943 }
944 } else {
945 readio_log = mp->m_readio_log;
946 writeio_log = mp->m_writeio_log;
947 }
948
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 if (sbp->sb_blocklog > readio_log) {
950 mp->m_readio_log = sbp->sb_blocklog;
951 } else {
952 mp->m_readio_log = readio_log;
953 }
954 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
955 if (sbp->sb_blocklog > writeio_log) {
956 mp->m_writeio_log = sbp->sb_blocklog;
957 } else {
958 mp->m_writeio_log = writeio_log;
959 }
960 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000961}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962
Eric Sandeen0771fb42007-10-12 11:03:40 +1000963/*
964 * Set whether we're using inode alignment.
965 */
966STATIC void
967xfs_set_inoalignment(xfs_mount_t *mp)
968{
Eric Sandeen62118702008-03-06 13:44:28 +1100969 if (xfs_sb_version_hasalign(&mp->m_sb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 mp->m_sb.sb_inoalignmt >=
971 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
972 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
973 else
974 mp->m_inoalign_mask = 0;
975 /*
976 * If we are using stripe alignment, check whether
977 * the stripe unit is a multiple of the inode alignment
978 */
979 if (mp->m_dalign && mp->m_inoalign_mask &&
980 !(mp->m_dalign & mp->m_inoalign_mask))
981 mp->m_sinoalign = mp->m_dalign;
982 else
983 mp->m_sinoalign = 0;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000984}
985
986/*
987 * Check that the data (and log if separate) are an ok size.
988 */
989STATIC int
Christoph Hellwig42490232008-08-13 16:49:32 +1000990xfs_check_sizes(xfs_mount_t *mp)
Eric Sandeen0771fb42007-10-12 11:03:40 +1000991{
992 xfs_buf_t *bp;
993 xfs_daddr_t d;
994 int error;
995
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
997 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
998 cmn_err(CE_WARN, "XFS: size check 1 failed");
Eric Sandeen657a4cf2010-04-30 03:42:49 +0000999 return XFS_ERROR(EFBIG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 }
1001 error = xfs_read_buf(mp, mp->m_ddev_targp,
1002 d - XFS_FSS_TO_BB(mp, 1),
1003 XFS_FSS_TO_BB(mp, 1), 0, &bp);
1004 if (!error) {
1005 xfs_buf_relse(bp);
1006 } else {
1007 cmn_err(CE_WARN, "XFS: size check 2 failed");
Eric Sandeen0771fb42007-10-12 11:03:40 +10001008 if (error == ENOSPC)
Eric Sandeen657a4cf2010-04-30 03:42:49 +00001009 error = XFS_ERROR(EFBIG);
Eric Sandeen0771fb42007-10-12 11:03:40 +10001010 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 }
1012
Christoph Hellwig42490232008-08-13 16:49:32 +10001013 if (mp->m_logdev_targp != mp->m_ddev_targp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
1015 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
1016 cmn_err(CE_WARN, "XFS: size check 3 failed");
Eric Sandeen657a4cf2010-04-30 03:42:49 +00001017 return XFS_ERROR(EFBIG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 }
1019 error = xfs_read_buf(mp, mp->m_logdev_targp,
1020 d - XFS_FSB_TO_BB(mp, 1),
1021 XFS_FSB_TO_BB(mp, 1), 0, &bp);
1022 if (!error) {
1023 xfs_buf_relse(bp);
1024 } else {
1025 cmn_err(CE_WARN, "XFS: size check 3 failed");
Eric Sandeen0771fb42007-10-12 11:03:40 +10001026 if (error == ENOSPC)
Eric Sandeen657a4cf2010-04-30 03:42:49 +00001027 error = XFS_ERROR(EFBIG);
Eric Sandeen0771fb42007-10-12 11:03:40 +10001028 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 }
1030 }
Eric Sandeen0771fb42007-10-12 11:03:40 +10001031 return 0;
1032}
1033
1034/*
Christoph Hellwig7d095252009-06-08 15:33:32 +02001035 * Clear the quotaflags in memory and in the superblock.
1036 */
1037int
1038xfs_mount_reset_sbqflags(
1039 struct xfs_mount *mp)
1040{
1041 int error;
1042 struct xfs_trans *tp;
1043
1044 mp->m_qflags = 0;
1045
1046 /*
1047 * It is OK to look at sb_qflags here in mount path,
1048 * without m_sb_lock.
1049 */
1050 if (mp->m_sb.sb_qflags == 0)
1051 return 0;
1052 spin_lock(&mp->m_sb_lock);
1053 mp->m_sb.sb_qflags = 0;
1054 spin_unlock(&mp->m_sb_lock);
1055
1056 /*
1057 * If the fs is readonly, let the incore superblock run
1058 * with quotas off but don't flush the update out to disk
1059 */
1060 if (mp->m_flags & XFS_MOUNT_RDONLY)
1061 return 0;
1062
1063#ifdef QUOTADEBUG
1064 xfs_fs_cmn_err(CE_NOTE, mp, "Writing superblock quota changes");
1065#endif
1066
1067 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1068 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1069 XFS_DEFAULT_LOG_COUNT);
1070 if (error) {
1071 xfs_trans_cancel(tp, 0);
1072 xfs_fs_cmn_err(CE_ALERT, mp,
1073 "xfs_mount_reset_sbqflags: Superblock update failed!");
1074 return error;
1075 }
1076
1077 xfs_mod_sb(tp, XFS_SB_QFLAGS);
1078 return xfs_trans_commit(tp, 0);
1079}
1080
Eric Sandeend5db0f92010-02-05 22:59:53 +00001081__uint64_t
1082xfs_default_resblks(xfs_mount_t *mp)
1083{
1084 __uint64_t resblks;
1085
1086 /*
Dave Chinner8babd8a2010-03-04 01:46:25 +00001087 * We default to 5% or 8192 fsbs of space reserved, whichever is
1088 * smaller. This is intended to cover concurrent allocation
1089 * transactions when we initially hit enospc. These each require a 4
1090 * block reservation. Hence by default we cover roughly 2000 concurrent
1091 * allocation reservations.
Eric Sandeend5db0f92010-02-05 22:59:53 +00001092 */
1093 resblks = mp->m_sb.sb_dblocks;
1094 do_div(resblks, 20);
Dave Chinner8babd8a2010-03-04 01:46:25 +00001095 resblks = min_t(__uint64_t, resblks, 8192);
Eric Sandeend5db0f92010-02-05 22:59:53 +00001096 return resblks;
1097}
1098
Christoph Hellwig7d095252009-06-08 15:33:32 +02001099/*
Eric Sandeen0771fb42007-10-12 11:03:40 +10001100 * This function does the following on an initial mount of a file system:
1101 * - reads the superblock from disk and init the mount struct
1102 * - if we're a 32-bit kernel, do a size check on the superblock
1103 * so we don't mount terabyte filesystems
1104 * - init mount struct realtime fields
1105 * - allocate inode hash table for fs
1106 * - init directory manager
1107 * - perform recovery and init the log manager
1108 */
1109int
1110xfs_mountfs(
Christoph Hellwig42490232008-08-13 16:49:32 +10001111 xfs_mount_t *mp)
Eric Sandeen0771fb42007-10-12 11:03:40 +10001112{
1113 xfs_sb_t *sbp = &(mp->m_sb);
1114 xfs_inode_t *rip;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001115 __uint64_t resblks;
Christoph Hellwig7d095252009-06-08 15:33:32 +02001116 uint quotamount = 0;
1117 uint quotaflags = 0;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001118 int error = 0;
1119
Eric Sandeen0771fb42007-10-12 11:03:40 +10001120 xfs_mount_common(mp, sbp);
1121
1122 /*
Eric Sandeene6957ea2008-04-10 12:19:34 +10001123 * Check for a mismatched features2 values. Older kernels
1124 * read & wrote into the wrong sb offset for sb_features2
1125 * on some platforms due to xfs_sb_t not being 64bit size aligned
1126 * when sb_features2 was added, which made older superblock
1127 * reading/writing routines swap it as a 64-bit value.
David Chinneree1c0902008-03-06 13:45:50 +11001128 *
Eric Sandeene6957ea2008-04-10 12:19:34 +10001129 * For backwards compatibility, we make both slots equal.
1130 *
1131 * If we detect a mismatched field, we OR the set bits into the
1132 * existing features2 field in case it has already been modified; we
1133 * don't want to lose any features. We then update the bad location
1134 * with the ORed value so that older kernels will see any features2
1135 * flags, and mark the two fields as needing updates once the
1136 * transaction subsystem is online.
David Chinneree1c0902008-03-06 13:45:50 +11001137 */
Eric Sandeene6957ea2008-04-10 12:19:34 +10001138 if (xfs_sb_has_mismatched_features2(sbp)) {
David Chinneree1c0902008-03-06 13:45:50 +11001139 cmn_err(CE_WARN,
1140 "XFS: correcting sb_features alignment problem");
1141 sbp->sb_features2 |= sbp->sb_bad_features2;
Eric Sandeene6957ea2008-04-10 12:19:34 +10001142 sbp->sb_bad_features2 = sbp->sb_features2;
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001143 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
Eric Sandeene6957ea2008-04-10 12:19:34 +10001144
1145 /*
1146 * Re-check for ATTR2 in case it was found in bad_features2
1147 * slot.
1148 */
Tim Shimmin7c12f292008-04-30 18:15:28 +10001149 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1150 !(mp->m_flags & XFS_MOUNT_NOATTR2))
Eric Sandeene6957ea2008-04-10 12:19:34 +10001151 mp->m_flags |= XFS_MOUNT_ATTR2;
Tim Shimmin7c12f292008-04-30 18:15:28 +10001152 }
Eric Sandeene6957ea2008-04-10 12:19:34 +10001153
Tim Shimmin7c12f292008-04-30 18:15:28 +10001154 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1155 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1156 xfs_sb_version_removeattr2(&mp->m_sb);
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001157 mp->m_update_flags |= XFS_SB_FEATURES2;
Tim Shimmin7c12f292008-04-30 18:15:28 +10001158
1159 /* update sb_versionnum for the clearing of the morebits */
1160 if (!sbp->sb_features2)
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001161 mp->m_update_flags |= XFS_SB_VERSIONNUM;
David Chinneree1c0902008-03-06 13:45:50 +11001162 }
1163
1164 /*
Eric Sandeen0771fb42007-10-12 11:03:40 +10001165 * Check if sb_agblocks is aligned at stripe boundary
1166 * If sb_agblocks is NOT aligned turn off m_dalign since
1167 * allocator alignment is within an ag, therefore ag has
1168 * to be aligned at stripe boundary.
1169 */
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001170 error = xfs_update_alignment(mp);
Eric Sandeen0771fb42007-10-12 11:03:40 +10001171 if (error)
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001172 goto out;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001173
1174 xfs_alloc_compute_maxlevels(mp);
1175 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
1176 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
1177 xfs_ialloc_compute_maxlevels(mp);
1178
1179 xfs_set_maxicount(mp);
1180
1181 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
1182
Christoph Hellwig27174202009-03-30 10:21:31 +02001183 error = xfs_uuid_mount(mp);
1184 if (error)
1185 goto out;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001186
1187 /*
1188 * Set the minimum read and write sizes
1189 */
1190 xfs_set_rw_sizes(mp);
1191
1192 /*
1193 * Set the inode cluster size.
1194 * This may still be overridden by the file system
1195 * block size if it is larger than the chosen cluster size.
1196 */
1197 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1198
1199 /*
1200 * Set inode alignment fields
1201 */
1202 xfs_set_inoalignment(mp);
1203
1204 /*
1205 * Check that the data (and log if separate) are an ok size.
1206 */
Christoph Hellwig42490232008-08-13 16:49:32 +10001207 error = xfs_check_sizes(mp);
Eric Sandeen0771fb42007-10-12 11:03:40 +10001208 if (error)
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001209 goto out_remove_uuid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210
1211 /*
1212 * Initialize realtime fields in the mount structure
1213 */
Eric Sandeen0771fb42007-10-12 11:03:40 +10001214 error = xfs_rtmount_init(mp);
1215 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 cmn_err(CE_WARN, "XFS: RT mount failed");
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001217 goto out_remove_uuid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 }
1219
1220 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 * Copies the low order bits of the timestamp and the randomly
1222 * set "sequence" number out of a UUID.
1223 */
1224 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1225
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 mp->m_dmevmask = 0; /* not persistent; set after each mount */
1227
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001228 xfs_dir_mount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229
1230 /*
1231 * Initialize the attribute manager's entries.
1232 */
1233 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1234
1235 /*
1236 * Initialize the precomputed transaction reservations values.
1237 */
1238 xfs_trans_init(mp);
1239
1240 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 * Allocate and initialize the per-ag data.
1242 */
Dave Chinner1c1c6eb2010-01-11 11:47:44 +00001243 spin_lock_init(&mp->m_perag_lock);
Dave Chinner9b98b6f2010-05-27 01:58:13 +00001244 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
Dave Chinner1c1c6eb2010-01-11 11:47:44 +00001245 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
1246 if (error) {
1247 cmn_err(CE_WARN, "XFS: Failed per-ag init: %d", error);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001248 goto out_remove_uuid;
Dave Chinner1c1c6eb2010-01-11 11:47:44 +00001249 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001251 if (!sbp->sb_logblocks) {
1252 cmn_err(CE_WARN, "XFS: no log defined");
1253 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
1254 error = XFS_ERROR(EFSCORRUPTED);
1255 goto out_free_perag;
1256 }
1257
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 /*
1259 * log's mount-time initialization. Perform 1st part recovery if needed
1260 */
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001261 error = xfs_log_mount(mp, mp->m_logdev_targp,
1262 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1263 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1264 if (error) {
1265 cmn_err(CE_WARN, "XFS: log mount failed");
1266 goto out_free_perag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 }
1268
1269 /*
David Chinner92821e22007-05-24 15:26:31 +10001270 * Now the log is mounted, we know if it was an unclean shutdown or
1271 * not. If it was, with the first phase of recovery has completed, we
1272 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1273 * but they are recovered transactionally in the second recovery phase
1274 * later.
1275 *
1276 * Hence we can safely re-initialise incore superblock counters from
1277 * the per-ag data. These may not be correct if the filesystem was not
1278 * cleanly unmounted, so we need to wait for recovery to finish before
1279 * doing this.
1280 *
1281 * If the filesystem was cleanly unmounted, then we can trust the
1282 * values in the superblock to be correct and we don't need to do
1283 * anything here.
1284 *
1285 * If we are currently making the filesystem, the initialisation will
1286 * fail as the perag data is in an undefined state.
1287 */
David Chinner92821e22007-05-24 15:26:31 +10001288 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1289 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1290 !mp->m_sb.sb_inprogress) {
1291 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001292 if (error)
1293 goto out_free_perag;
David Chinner92821e22007-05-24 15:26:31 +10001294 }
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001295
David Chinner92821e22007-05-24 15:26:31 +10001296 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 * Get and sanity-check the root inode.
1298 * Save the pointer to it in the mount structure.
1299 */
Dave Chinner7b6259e2010-06-24 11:35:17 +10001300 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 if (error) {
1302 cmn_err(CE_WARN, "XFS: failed to read root inode");
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001303 goto out_log_dealloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 }
1305
1306 ASSERT(rip != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307
1308 if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
1309 cmn_err(CE_WARN, "XFS: corrupted root inode");
Nathan Scottb6574522006-06-09 15:29:40 +10001310 cmn_err(CE_WARN, "Device %s - root %llu is not a directory",
1311 XFS_BUFTARG_NAME(mp->m_ddev_targp),
1312 (unsigned long long)rip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1314 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1315 mp);
1316 error = XFS_ERROR(EFSCORRUPTED);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001317 goto out_rele_rip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 }
1319 mp->m_rootip = rip; /* save it */
1320
1321 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1322
1323 /*
1324 * Initialize realtime inode pointers in the mount structure
1325 */
Eric Sandeen0771fb42007-10-12 11:03:40 +10001326 error = xfs_rtmount_inodes(mp);
1327 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 /*
1329 * Free up the root inode.
1330 */
1331 cmn_err(CE_WARN, "XFS: failed to read RT inodes");
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001332 goto out_rele_rip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 }
1334
1335 /*
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001336 * If this is a read-only mount defer the superblock updates until
1337 * the next remount into writeable mode. Otherwise we would never
1338 * perform the update e.g. for the root filesystem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 */
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001340 if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1341 error = xfs_mount_log_sb(mp, mp->m_update_flags);
David Chinnere5720ee2008-04-10 12:21:18 +10001342 if (error) {
1343 cmn_err(CE_WARN, "XFS: failed to write sb changes");
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001344 goto out_rtunmount;
David Chinnere5720ee2008-04-10 12:21:18 +10001345 }
1346 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
1348 /*
1349 * Initialise the XFS quota management subsystem for this mount
1350 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001351 if (XFS_IS_QUOTA_RUNNING(mp)) {
1352 error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
1353 if (error)
1354 goto out_rtunmount;
1355 } else {
1356 ASSERT(!XFS_IS_QUOTA_ON(mp));
1357
1358 /*
1359 * If a file system had quotas running earlier, but decided to
1360 * mount without -o uquota/pquota/gquota options, revoke the
1361 * quotachecked license.
1362 */
1363 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
1364 cmn_err(CE_NOTE,
1365 "XFS: resetting qflags for filesystem %s",
1366 mp->m_fsname);
1367
1368 error = xfs_mount_reset_sbqflags(mp);
1369 if (error)
1370 return error;
1371 }
1372 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373
1374 /*
1375 * Finish recovering the file system. This part needed to be
1376 * delayed until after the root and real-time bitmap inodes
1377 * were consistently read in.
1378 */
Christoph Hellwig42490232008-08-13 16:49:32 +10001379 error = xfs_log_mount_finish(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 if (error) {
1381 cmn_err(CE_WARN, "XFS: log mount finish failed");
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001382 goto out_rtunmount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 }
1384
1385 /*
1386 * Complete the quota initialisation, post-log-replay component.
1387 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001388 if (quotamount) {
1389 ASSERT(mp->m_qflags == 0);
1390 mp->m_qflags = quotaflags;
1391
1392 xfs_qm_mount_quotas(mp);
1393 }
1394
David Chinner84e1e992007-06-18 16:50:27 +10001395 /*
1396 * Now we are mounted, reserve a small amount of unused space for
1397 * privileged transactions. This is needed so that transaction
1398 * space required for critical operations can dip into this pool
1399 * when at ENOSPC. This is needed for operations like create with
1400 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1401 * are not allowed to use this reserved space.
Dave Chinner8babd8a2010-03-04 01:46:25 +00001402 *
1403 * This may drive us straight to ENOSPC on mount, but that implies
1404 * we were already there on the last unmount. Warn if this occurs.
David Chinner84e1e992007-06-18 16:50:27 +10001405 */
Eric Sandeend5db0f92010-02-05 22:59:53 +00001406 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
1407 resblks = xfs_default_resblks(mp);
1408 error = xfs_reserve_blocks(mp, &resblks, NULL);
1409 if (error)
1410 cmn_err(CE_WARN, "XFS: Unable to allocate reserve "
1411 "blocks. Continuing without a reserve pool.");
1412 }
David Chinner84e1e992007-06-18 16:50:27 +10001413
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 return 0;
1415
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001416 out_rtunmount:
1417 xfs_rtunmount_inodes(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001418 out_rele_rip:
Christoph Hellwig43355092008-03-27 18:01:08 +11001419 IRELE(rip);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001420 out_log_dealloc:
Christoph Hellwig21b699c2009-03-16 08:19:29 +01001421 xfs_log_unmount(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001422 out_free_perag:
Christoph Hellwigff4f0382008-08-13 16:50:47 +10001423 xfs_free_perag(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001424 out_remove_uuid:
Christoph Hellwig27174202009-03-30 10:21:31 +02001425 xfs_uuid_unmount(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001426 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 return error;
1428}
1429
1430/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 * This flushes out the inodes,dquots and the superblock, unmounts the
1432 * log and makes sure that incore structures are freed.
1433 */
Christoph Hellwig41b5c2e2008-08-13 16:49:57 +10001434void
1435xfs_unmountfs(
1436 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437{
Christoph Hellwig41b5c2e2008-08-13 16:49:57 +10001438 __uint64_t resblks;
1439 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440
Christoph Hellwig7d095252009-06-08 15:33:32 +02001441 xfs_qm_unmount_quotas(mp);
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001442 xfs_rtunmount_inodes(mp);
Christoph Hellwig77508ec2008-08-13 16:49:04 +10001443 IRELE(mp->m_rootip);
1444
David Chinner641c56f2007-06-18 16:50:17 +10001445 /*
1446 * We can potentially deadlock here if we have an inode cluster
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001447 * that has been freed has its buffer still pinned in memory because
David Chinner641c56f2007-06-18 16:50:17 +10001448 * the transaction is still sitting in a iclog. The stale inodes
1449 * on that buffer will have their flush locks held until the
1450 * transaction hits the disk and the callbacks run. the inode
1451 * flush takes the flush lock unconditionally and with nothing to
1452 * push out the iclog we will never get that unlocked. hence we
1453 * need to force the log first.
1454 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00001455 xfs_log_force(mp, XFS_LOG_SYNC);
Dave Chinnerc8543632010-02-06 12:39:36 +11001456
1457 /*
1458 * Do a delwri reclaim pass first so that as many dirty inodes are
1459 * queued up for IO as possible. Then flush the buffers before making
1460 * a synchronous path to catch all the remaining inodes are reclaimed.
1461 * This makes the reclaim process as quick as possible by avoiding
1462 * synchronous writeout and blocking on inodes already in the delwri
1463 * state as much as possible.
1464 */
1465 xfs_reclaim_inodes(mp, 0);
1466 XFS_bflush(mp->m_ddev_targp);
1467 xfs_reclaim_inodes(mp, SYNC_WAIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468
Christoph Hellwig7d095252009-06-08 15:33:32 +02001469 xfs_qm_unmount(mp);
Lachlan McIlroya357a122008-10-30 16:53:25 +11001470
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 /*
1472 * Flush out the log synchronously so that we know for sure
1473 * that nothing is pinned. This is important because bflush()
1474 * will skip pinned buffers.
1475 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00001476 xfs_log_force(mp, XFS_LOG_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477
1478 xfs_binval(mp->m_ddev_targp);
1479 if (mp->m_rtdev_targp) {
1480 xfs_binval(mp->m_rtdev_targp);
1481 }
1482
David Chinner84e1e992007-06-18 16:50:27 +10001483 /*
1484 * Unreserve any blocks we have so that when we unmount we don't account
1485 * the reserved free space as used. This is really only necessary for
1486 * lazy superblock counting because it trusts the incore superblock
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001487 * counters to be absolutely correct on clean unmount.
David Chinner84e1e992007-06-18 16:50:27 +10001488 *
1489 * We don't bother correcting this elsewhere for lazy superblock
1490 * counting because on mount of an unclean filesystem we reconstruct the
1491 * correct counter value and this is irrelevant.
1492 *
1493 * For non-lazy counter filesystems, this doesn't matter at all because
1494 * we only every apply deltas to the superblock and hence the incore
1495 * value does not matter....
1496 */
1497 resblks = 0;
David Chinner714082b2008-04-10 12:20:03 +10001498 error = xfs_reserve_blocks(mp, &resblks, NULL);
1499 if (error)
1500 cmn_err(CE_WARN, "XFS: Unable to free reserved block pool. "
1501 "Freespace may not be correct on next mount.");
1502
David Chinnere5720ee2008-04-10 12:21:18 +10001503 error = xfs_log_sbcount(mp, 1);
1504 if (error)
1505 cmn_err(CE_WARN, "XFS: Unable to update superblock counters. "
1506 "Freespace may not be correct on next mount.");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 xfs_unmountfs_writesb(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 xfs_unmountfs_wait(mp); /* wait for async bufs */
Christoph Hellwig21b699c2009-03-16 08:19:29 +01001509 xfs_log_unmount_write(mp);
1510 xfs_log_unmount(mp);
Christoph Hellwig27174202009-03-30 10:21:31 +02001511 xfs_uuid_unmount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512
Christoph Hellwig1550d0b2008-08-13 16:17:37 +10001513#if defined(DEBUG)
Christoph Hellwig0ce4cfd2007-08-30 17:20:53 +10001514 xfs_errortag_clearall(mp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515#endif
Christoph Hellwigff4f0382008-08-13 16:50:47 +10001516 xfs_free_perag(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517}
1518
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10001519STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520xfs_unmountfs_wait(xfs_mount_t *mp)
1521{
1522 if (mp->m_logdev_targp != mp->m_ddev_targp)
1523 xfs_wait_buftarg(mp->m_logdev_targp);
1524 if (mp->m_rtdev_targp)
1525 xfs_wait_buftarg(mp->m_rtdev_targp);
1526 xfs_wait_buftarg(mp->m_ddev_targp);
1527}
1528
1529int
David Chinner92821e22007-05-24 15:26:31 +10001530xfs_fs_writable(xfs_mount_t *mp)
1531{
Christoph Hellwigb267ce92007-08-30 17:21:30 +10001532 return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001533 (mp->m_flags & XFS_MOUNT_RDONLY));
David Chinner92821e22007-05-24 15:26:31 +10001534}
1535
1536/*
1537 * xfs_log_sbcount
1538 *
1539 * Called either periodically to keep the on disk superblock values
1540 * roughly up to date or from unmount to make sure the values are
1541 * correct on a clean unmount.
1542 *
1543 * Note this code can be called during the process of freezing, so
1544 * we may need to use the transaction allocator which does not not
1545 * block when the transaction subsystem is in its frozen state.
1546 */
1547int
1548xfs_log_sbcount(
1549 xfs_mount_t *mp,
1550 uint sync)
1551{
1552 xfs_trans_t *tp;
1553 int error;
1554
1555 if (!xfs_fs_writable(mp))
1556 return 0;
1557
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10001558 xfs_icsb_sync_counters(mp, 0);
David Chinner92821e22007-05-24 15:26:31 +10001559
1560 /*
1561 * we don't need to do this if we are updating the superblock
1562 * counters on every modification.
1563 */
1564 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1565 return 0;
1566
Christoph Hellwig80641dc2009-10-19 04:00:03 +00001567 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
David Chinner92821e22007-05-24 15:26:31 +10001568 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1569 XFS_DEFAULT_LOG_COUNT);
1570 if (error) {
1571 xfs_trans_cancel(tp, 0);
1572 return error;
1573 }
1574
1575 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1576 if (sync)
1577 xfs_trans_set_sync(tp);
David Chinnere5720ee2008-04-10 12:21:18 +10001578 error = xfs_trans_commit(tp, 0);
1579 return error;
David Chinner92821e22007-05-24 15:26:31 +10001580}
1581
1582int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583xfs_unmountfs_writesb(xfs_mount_t *mp)
1584{
1585 xfs_buf_t *sbp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 int error = 0;
1587
1588 /*
1589 * skip superblock write if fs is read-only, or
1590 * if we are doing a forced umount.
1591 */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001592 if (!((mp->m_flags & XFS_MOUNT_RDONLY) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 XFS_FORCED_SHUTDOWN(mp))) {
David Chinner8d280b92006-03-14 13:13:09 +11001594
David Chinner92821e22007-05-24 15:26:31 +10001595 sbp = xfs_getsb(mp, 0);
David Chinner8d280b92006-03-14 13:13:09 +11001596
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 XFS_BUF_UNDONE(sbp);
1598 XFS_BUF_UNREAD(sbp);
1599 XFS_BUF_UNDELAYWRITE(sbp);
1600 XFS_BUF_WRITE(sbp);
1601 XFS_BUF_UNASYNC(sbp);
1602 ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1603 xfsbdstrat(mp, sbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 error = xfs_iowait(sbp);
1605 if (error)
1606 xfs_ioerror_alert("xfs_unmountfs_writesb",
1607 mp, sbp, XFS_BUF_ADDR(sbp));
David Chinner92821e22007-05-24 15:26:31 +10001608 xfs_buf_relse(sbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001610 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611}
1612
1613/*
1614 * xfs_mod_sb() can be used to copy arbitrary changes to the
1615 * in-core superblock into the superblock buffer to be logged.
1616 * It does not provide the higher level of locking that is
1617 * needed to protect the in-core superblock from concurrent
1618 * access.
1619 */
1620void
1621xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1622{
1623 xfs_buf_t *bp;
1624 int first;
1625 int last;
1626 xfs_mount_t *mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 xfs_sb_field_t f;
1628
1629 ASSERT(fields);
1630 if (!fields)
1631 return;
1632 mp = tp->t_mountp;
1633 bp = xfs_trans_getsb(tp, mp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 first = sizeof(xfs_sb_t);
1635 last = 0;
1636
1637 /* translate/copy */
1638
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +10001639 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640
1641 /* find modified range */
Dave Chinner587aa0f2010-01-20 12:04:53 +11001642 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1643 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1644 last = xfs_sb_info[f + 1].offset - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645
1646 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1647 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1648 first = xfs_sb_info[f].offset;
1649
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 xfs_trans_log_buf(tp, bp, first, last);
1651}
Yingping Lud210a282006-06-09 14:55:18 +10001652
Yingping Lud210a282006-06-09 14:55:18 +10001653
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654/*
1655 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1656 * a delta to a specified field in the in-core superblock. Simply
1657 * switch on the field indicated and apply the delta to that field.
1658 * Fields are not allowed to dip below zero, so if the delta would
1659 * do this do not apply it and return EINVAL.
1660 *
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001661 * The m_sb_lock must be held when this routine is called.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05001663STATIC int
David Chinner20f4ebf2007-02-10 18:36:10 +11001664xfs_mod_incore_sb_unlocked(
1665 xfs_mount_t *mp,
1666 xfs_sb_field_t field,
1667 int64_t delta,
1668 int rsvd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669{
1670 int scounter; /* short counter for 32 bit fields */
1671 long long lcounter; /* long counter for 64 bit fields */
1672 long long res_used, rem;
1673
1674 /*
1675 * With the in-core superblock spin lock held, switch
1676 * on the indicated field. Apply the delta to the
1677 * proper field. If the fields value would dip below
1678 * 0, then do not apply the delta and return EINVAL.
1679 */
1680 switch (field) {
1681 case XFS_SBS_ICOUNT:
1682 lcounter = (long long)mp->m_sb.sb_icount;
1683 lcounter += delta;
1684 if (lcounter < 0) {
1685 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001686 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 }
1688 mp->m_sb.sb_icount = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001689 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 case XFS_SBS_IFREE:
1691 lcounter = (long long)mp->m_sb.sb_ifree;
1692 lcounter += delta;
1693 if (lcounter < 0) {
1694 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001695 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 }
1697 mp->m_sb.sb_ifree = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001698 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 case XFS_SBS_FDBLOCKS:
David Chinner4be536d2006-09-07 14:26:50 +10001700 lcounter = (long long)
1701 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1703
1704 if (delta > 0) { /* Putting blocks back */
1705 if (res_used > delta) {
1706 mp->m_resblks_avail += delta;
1707 } else {
1708 rem = delta - res_used;
1709 mp->m_resblks_avail = mp->m_resblks;
1710 lcounter += rem;
1711 }
1712 } else { /* Taking blocks away */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 lcounter += delta;
Dave Chinner8babd8a2010-03-04 01:46:25 +00001714 if (lcounter >= 0) {
1715 mp->m_sb.sb_fdblocks = lcounter +
1716 XFS_ALLOC_SET_ASIDE(mp);
1717 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 }
Dave Chinner8babd8a2010-03-04 01:46:25 +00001719
1720 /*
1721 * We are out of blocks, use any available reserved
1722 * blocks if were allowed to.
1723 */
1724 if (!rsvd)
1725 return XFS_ERROR(ENOSPC);
1726
1727 lcounter = (long long)mp->m_resblks_avail + delta;
1728 if (lcounter >= 0) {
1729 mp->m_resblks_avail = lcounter;
1730 return 0;
1731 }
1732 printk_once(KERN_WARNING
1733 "Filesystem \"%s\": reserve blocks depleted! "
1734 "Consider increasing reserve pool size.",
1735 mp->m_fsname);
1736 return XFS_ERROR(ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 }
1738
David Chinner4be536d2006-09-07 14:26:50 +10001739 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
Jesper Juhl014c2542006-01-15 02:37:08 +01001740 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 case XFS_SBS_FREXTENTS:
1742 lcounter = (long long)mp->m_sb.sb_frextents;
1743 lcounter += delta;
1744 if (lcounter < 0) {
Jesper Juhl014c2542006-01-15 02:37:08 +01001745 return XFS_ERROR(ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 }
1747 mp->m_sb.sb_frextents = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001748 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 case XFS_SBS_DBLOCKS:
1750 lcounter = (long long)mp->m_sb.sb_dblocks;
1751 lcounter += delta;
1752 if (lcounter < 0) {
1753 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001754 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 }
1756 mp->m_sb.sb_dblocks = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001757 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 case XFS_SBS_AGCOUNT:
1759 scounter = mp->m_sb.sb_agcount;
1760 scounter += delta;
1761 if (scounter < 0) {
1762 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001763 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 }
1765 mp->m_sb.sb_agcount = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001766 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 case XFS_SBS_IMAX_PCT:
1768 scounter = mp->m_sb.sb_imax_pct;
1769 scounter += delta;
1770 if (scounter < 0) {
1771 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001772 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 }
1774 mp->m_sb.sb_imax_pct = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001775 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 case XFS_SBS_REXTSIZE:
1777 scounter = mp->m_sb.sb_rextsize;
1778 scounter += delta;
1779 if (scounter < 0) {
1780 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001781 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 }
1783 mp->m_sb.sb_rextsize = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001784 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 case XFS_SBS_RBMBLOCKS:
1786 scounter = mp->m_sb.sb_rbmblocks;
1787 scounter += delta;
1788 if (scounter < 0) {
1789 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001790 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 }
1792 mp->m_sb.sb_rbmblocks = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001793 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 case XFS_SBS_RBLOCKS:
1795 lcounter = (long long)mp->m_sb.sb_rblocks;
1796 lcounter += delta;
1797 if (lcounter < 0) {
1798 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001799 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 }
1801 mp->m_sb.sb_rblocks = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001802 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 case XFS_SBS_REXTENTS:
1804 lcounter = (long long)mp->m_sb.sb_rextents;
1805 lcounter += delta;
1806 if (lcounter < 0) {
1807 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001808 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 }
1810 mp->m_sb.sb_rextents = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001811 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 case XFS_SBS_REXTSLOG:
1813 scounter = mp->m_sb.sb_rextslog;
1814 scounter += delta;
1815 if (scounter < 0) {
1816 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001817 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 }
1819 mp->m_sb.sb_rextslog = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001820 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 default:
1822 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001823 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 }
1825}
1826
1827/*
1828 * xfs_mod_incore_sb() is used to change a field in the in-core
1829 * superblock structure by the specified delta. This modification
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001830 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 * routine to do the work.
1832 */
1833int
David Chinner20f4ebf2007-02-10 18:36:10 +11001834xfs_mod_incore_sb(
1835 xfs_mount_t *mp,
1836 xfs_sb_field_t field,
1837 int64_t delta,
1838 int rsvd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 int status;
1841
David Chinner8d280b92006-03-14 13:13:09 +11001842 /* check for per-cpu counters */
1843 switch (field) {
1844#ifdef HAVE_PERCPU_SB
1845 case XFS_SBS_ICOUNT:
1846 case XFS_SBS_IFREE:
1847 case XFS_SBS_FDBLOCKS:
1848 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1849 status = xfs_icsb_modify_counters(mp, field,
1850 delta, rsvd);
1851 break;
1852 }
1853 /* FALLTHROUGH */
1854#endif
1855 default:
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001856 spin_lock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001857 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001858 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001859 break;
1860 }
1861
Jesper Juhl014c2542006-01-15 02:37:08 +01001862 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863}
1864
1865/*
1866 * xfs_mod_incore_sb_batch() is used to change more than one field
1867 * in the in-core superblock structure at a time. This modification
1868 * is protected by a lock internal to this module. The fields and
1869 * changes to those fields are specified in the array of xfs_mod_sb
1870 * structures passed in.
1871 *
1872 * Either all of the specified deltas will be applied or none of
1873 * them will. If any modified field dips below 0, then all modifications
1874 * will be backed out and EINVAL will be returned.
1875 */
1876int
1877xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1878{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 int status=0;
1880 xfs_mod_sb_t *msbp;
1881
1882 /*
1883 * Loop through the array of mod structures and apply each
1884 * individually. If any fail, then back out all those
1885 * which have already been applied. Do all of this within
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001886 * the scope of the m_sb_lock so that all of the changes will
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 * be atomic.
1888 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001889 spin_lock(&mp->m_sb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 msbp = &msb[0];
1891 for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1892 /*
1893 * Apply the delta at index n. If it fails, break
1894 * from the loop so we'll fall into the undo loop
1895 * below.
1896 */
David Chinner8d280b92006-03-14 13:13:09 +11001897 switch (msbp->msb_field) {
1898#ifdef HAVE_PERCPU_SB
1899 case XFS_SBS_ICOUNT:
1900 case XFS_SBS_IFREE:
1901 case XFS_SBS_FDBLOCKS:
1902 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001903 spin_unlock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11001904 status = xfs_icsb_modify_counters(mp,
David Chinner8d280b92006-03-14 13:13:09 +11001905 msbp->msb_field,
1906 msbp->msb_delta, rsvd);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001907 spin_lock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001908 break;
1909 }
1910 /* FALLTHROUGH */
1911#endif
1912 default:
1913 status = xfs_mod_incore_sb_unlocked(mp,
1914 msbp->msb_field,
1915 msbp->msb_delta, rsvd);
1916 break;
1917 }
1918
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 if (status != 0) {
1920 break;
1921 }
1922 }
1923
1924 /*
1925 * If we didn't complete the loop above, then back out
1926 * any changes made to the superblock. If you add code
1927 * between the loop above and here, make sure that you
1928 * preserve the value of status. Loop back until
1929 * we step below the beginning of the array. Make sure
1930 * we don't touch anything back there.
1931 */
1932 if (status != 0) {
1933 msbp--;
1934 while (msbp >= msb) {
David Chinner8d280b92006-03-14 13:13:09 +11001935 switch (msbp->msb_field) {
1936#ifdef HAVE_PERCPU_SB
1937 case XFS_SBS_ICOUNT:
1938 case XFS_SBS_IFREE:
1939 case XFS_SBS_FDBLOCKS:
1940 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001941 spin_unlock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11001942 status = xfs_icsb_modify_counters(mp,
David Chinner8d280b92006-03-14 13:13:09 +11001943 msbp->msb_field,
1944 -(msbp->msb_delta),
1945 rsvd);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001946 spin_lock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001947 break;
1948 }
1949 /* FALLTHROUGH */
1950#endif
1951 default:
1952 status = xfs_mod_incore_sb_unlocked(mp,
1953 msbp->msb_field,
1954 -(msbp->msb_delta),
1955 rsvd);
1956 break;
1957 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 ASSERT(status == 0);
1959 msbp--;
1960 }
1961 }
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001962 spin_unlock(&mp->m_sb_lock);
Jesper Juhl014c2542006-01-15 02:37:08 +01001963 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964}
1965
1966/*
1967 * xfs_getsb() is called to obtain the buffer for the superblock.
1968 * The buffer is returned locked and read in from disk.
1969 * The buffer should be released with a call to xfs_brelse().
1970 *
1971 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1972 * the superblock buffer if it can be locked without sleeping.
1973 * If it can't then we'll return NULL.
1974 */
1975xfs_buf_t *
1976xfs_getsb(
1977 xfs_mount_t *mp,
1978 int flags)
1979{
1980 xfs_buf_t *bp;
1981
1982 ASSERT(mp->m_sb_bp != NULL);
1983 bp = mp->m_sb_bp;
Christoph Hellwig0cadda12010-01-19 09:56:44 +00001984 if (flags & XBF_TRYLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 if (!XFS_BUF_CPSEMA(bp)) {
1986 return NULL;
1987 }
1988 } else {
1989 XFS_BUF_PSEMA(bp, PRIBIO);
1990 }
1991 XFS_BUF_HOLD(bp);
1992 ASSERT(XFS_BUF_ISDONE(bp));
Jesper Juhl014c2542006-01-15 02:37:08 +01001993 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994}
1995
1996/*
1997 * Used to free the superblock along various error paths.
1998 */
1999void
2000xfs_freesb(
2001 xfs_mount_t *mp)
2002{
2003 xfs_buf_t *bp;
2004
2005 /*
2006 * Use xfs_getsb() so that the buffer will be locked
2007 * when we call xfs_buf_relse().
2008 */
2009 bp = xfs_getsb(mp, 0);
2010 XFS_BUF_UNMANAGE(bp);
2011 xfs_buf_relse(bp);
2012 mp->m_sb_bp = NULL;
2013}
2014
2015/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 * Used to log changes to the superblock unit and width fields which could
Eric Sandeene6957ea2008-04-10 12:19:34 +10002017 * be altered by the mount options, as well as any potential sb_features2
2018 * fixup. Only the first superblock is updated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 */
Christoph Hellwig7884bc82009-01-19 02:04:07 +01002020int
David Chinneree1c0902008-03-06 13:45:50 +11002021xfs_mount_log_sb(
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 xfs_mount_t *mp,
2023 __int64_t fields)
2024{
2025 xfs_trans_t *tp;
David Chinnere5720ee2008-04-10 12:21:18 +10002026 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027
David Chinneree1c0902008-03-06 13:45:50 +11002028 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
David Chinner4b166de2008-05-20 11:30:27 +10002029 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
2030 XFS_SB_VERSIONNUM));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031
2032 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
David Chinnere5720ee2008-04-10 12:21:18 +10002033 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
2034 XFS_DEFAULT_LOG_COUNT);
2035 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 xfs_trans_cancel(tp, 0);
David Chinnere5720ee2008-04-10 12:21:18 +10002037 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 }
2039 xfs_mod_sb(tp, fields);
David Chinnere5720ee2008-04-10 12:21:18 +10002040 error = xfs_trans_commit(tp, 0);
2041 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042}
David Chinner8d280b92006-03-14 13:13:09 +11002043
Christoph Hellwigdda35b82010-02-15 09:44:46 +00002044/*
2045 * If the underlying (data/log/rt) device is readonly, there are some
2046 * operations that cannot proceed.
2047 */
2048int
2049xfs_dev_is_read_only(
2050 struct xfs_mount *mp,
2051 char *message)
2052{
2053 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
2054 xfs_readonly_buftarg(mp->m_logdev_targp) ||
2055 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
2056 cmn_err(CE_NOTE,
2057 "XFS: %s required on read-only device.", message);
2058 cmn_err(CE_NOTE,
2059 "XFS: write access unavailable, cannot proceed.");
2060 return EROFS;
2061 }
2062 return 0;
2063}
David Chinner8d280b92006-03-14 13:13:09 +11002064
2065#ifdef HAVE_PERCPU_SB
2066/*
2067 * Per-cpu incore superblock counters
2068 *
2069 * Simple concept, difficult implementation
2070 *
2071 * Basically, replace the incore superblock counters with a distributed per cpu
2072 * counter for contended fields (e.g. free block count).
2073 *
2074 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
2075 * hence needs to be accurately read when we are running low on space. Hence
2076 * there is a method to enable and disable the per-cpu counters based on how
2077 * much "stuff" is available in them.
2078 *
2079 * Basically, a counter is enabled if there is enough free resource to justify
2080 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
2081 * ENOSPC), then we disable the counters to synchronise all callers and
2082 * re-distribute the available resources.
2083 *
2084 * If, once we redistributed the available resources, we still get a failure,
2085 * we disable the per-cpu counter and go through the slow path.
2086 *
2087 * The slow path is the current xfs_mod_incore_sb() function. This means that
Malcolm Parsons9da096f2009-03-29 09:55:42 +02002088 * when we disable a per-cpu counter, we need to drain its resources back to
David Chinner8d280b92006-03-14 13:13:09 +11002089 * the global superblock. We do this after disabling the counter to prevent
2090 * more threads from queueing up on the counter.
2091 *
2092 * Essentially, this means that we still need a lock in the fast path to enable
2093 * synchronisation between the global counters and the per-cpu counters. This
2094 * is not a problem because the lock will be local to a CPU almost all the time
2095 * and have little contention except when we get to ENOSPC conditions.
2096 *
2097 * Basically, this lock becomes a barrier that enables us to lock out the fast
2098 * path while we do things like enabling and disabling counters and
2099 * synchronising the counters.
2100 *
2101 * Locking rules:
2102 *
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002103 * 1. m_sb_lock before picking up per-cpu locks
David Chinner8d280b92006-03-14 13:13:09 +11002104 * 2. per-cpu locks always picked up via for_each_online_cpu() order
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002105 * 3. accurate counter sync requires m_sb_lock + per cpu locks
David Chinner8d280b92006-03-14 13:13:09 +11002106 * 4. modifying per-cpu counters requires holding per-cpu lock
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002107 * 5. modifying global counters requires holding m_sb_lock
2108 * 6. enabling or disabling a counter requires holding the m_sb_lock
David Chinner8d280b92006-03-14 13:13:09 +11002109 * and _none_ of the per-cpu locks.
2110 *
2111 * Disabled counters are only ever re-enabled by a balance operation
2112 * that results in more free resources per CPU than a given threshold.
2113 * To ensure counters don't remain disabled, they are rebalanced when
2114 * the global resource goes above a higher threshold (i.e. some hysteresis
2115 * is present to prevent thrashing).
David Chinner8d280b92006-03-14 13:13:09 +11002116 */
David Chinnere8234a62006-03-14 13:23:52 +11002117
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002118#ifdef CONFIG_HOTPLUG_CPU
David Chinnere8234a62006-03-14 13:23:52 +11002119/*
2120 * hot-plug CPU notifier support.
2121 *
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002122 * We need a notifier per filesystem as we need to be able to identify
2123 * the filesystem to balance the counters out. This is achieved by
2124 * having a notifier block embedded in the xfs_mount_t and doing pointer
2125 * magic to get the mount pointer from the notifier block address.
David Chinnere8234a62006-03-14 13:23:52 +11002126 */
2127STATIC int
2128xfs_icsb_cpu_notify(
2129 struct notifier_block *nfb,
2130 unsigned long action,
2131 void *hcpu)
2132{
2133 xfs_icsb_cnts_t *cntp;
2134 xfs_mount_t *mp;
David Chinnere8234a62006-03-14 13:23:52 +11002135
2136 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
2137 cntp = (xfs_icsb_cnts_t *)
2138 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
2139 switch (action) {
2140 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002141 case CPU_UP_PREPARE_FROZEN:
David Chinnere8234a62006-03-14 13:23:52 +11002142 /* Easy Case - initialize the area and locks, and
2143 * then rebalance when online does everything else for us. */
David Chinner01e1b692006-03-14 13:29:16 +11002144 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinnere8234a62006-03-14 13:23:52 +11002145 break;
2146 case CPU_ONLINE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002147 case CPU_ONLINE_FROZEN:
David Chinner03135cf2007-02-10 18:35:15 +11002148 xfs_icsb_lock(mp);
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002149 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2150 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2151 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
David Chinner03135cf2007-02-10 18:35:15 +11002152 xfs_icsb_unlock(mp);
David Chinnere8234a62006-03-14 13:23:52 +11002153 break;
2154 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002155 case CPU_DEAD_FROZEN:
David Chinnere8234a62006-03-14 13:23:52 +11002156 /* Disable all the counters, then fold the dead cpu's
2157 * count into the total on the global superblock and
2158 * re-enable the counters. */
David Chinner03135cf2007-02-10 18:35:15 +11002159 xfs_icsb_lock(mp);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002160 spin_lock(&mp->m_sb_lock);
David Chinnere8234a62006-03-14 13:23:52 +11002161 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
2162 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
2163 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
2164
2165 mp->m_sb.sb_icount += cntp->icsb_icount;
2166 mp->m_sb.sb_ifree += cntp->icsb_ifree;
2167 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
2168
David Chinner01e1b692006-03-14 13:29:16 +11002169 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinnere8234a62006-03-14 13:23:52 +11002170
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002171 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
2172 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
2173 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002174 spin_unlock(&mp->m_sb_lock);
David Chinner03135cf2007-02-10 18:35:15 +11002175 xfs_icsb_unlock(mp);
David Chinnere8234a62006-03-14 13:23:52 +11002176 break;
2177 }
2178
2179 return NOTIFY_OK;
2180}
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002181#endif /* CONFIG_HOTPLUG_CPU */
David Chinnere8234a62006-03-14 13:23:52 +11002182
David Chinner8d280b92006-03-14 13:13:09 +11002183int
2184xfs_icsb_init_counters(
2185 xfs_mount_t *mp)
2186{
2187 xfs_icsb_cnts_t *cntp;
2188 int i;
2189
2190 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2191 if (mp->m_sb_cnts == NULL)
2192 return -ENOMEM;
2193
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002194#ifdef CONFIG_HOTPLUG_CPU
David Chinnere8234a62006-03-14 13:23:52 +11002195 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2196 mp->m_icsb_notifier.priority = 0;
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002197 register_hotcpu_notifier(&mp->m_icsb_notifier);
2198#endif /* CONFIG_HOTPLUG_CPU */
David Chinnere8234a62006-03-14 13:23:52 +11002199
David Chinner8d280b92006-03-14 13:13:09 +11002200 for_each_online_cpu(i) {
2201 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11002202 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinner8d280b92006-03-14 13:13:09 +11002203 }
David Chinner20b64282007-02-10 18:35:09 +11002204
2205 mutex_init(&mp->m_icsb_mutex);
2206
David Chinner8d280b92006-03-14 13:13:09 +11002207 /*
2208 * start with all counters disabled so that the
2209 * initial balance kicks us off correctly
2210 */
2211 mp->m_icsb_counters = -1;
2212 return 0;
2213}
2214
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11002215void
2216xfs_icsb_reinit_counters(
2217 xfs_mount_t *mp)
2218{
2219 xfs_icsb_lock(mp);
2220 /*
2221 * start with all counters disabled so that the
2222 * initial balance kicks us off correctly
2223 */
2224 mp->m_icsb_counters = -1;
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002225 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2226 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2227 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11002228 xfs_icsb_unlock(mp);
2229}
2230
Christoph Hellwigc962fb72008-05-20 15:10:52 +10002231void
David Chinner8d280b92006-03-14 13:13:09 +11002232xfs_icsb_destroy_counters(
2233 xfs_mount_t *mp)
2234{
David Chinnere8234a62006-03-14 13:23:52 +11002235 if (mp->m_sb_cnts) {
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002236 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
David Chinner8d280b92006-03-14 13:13:09 +11002237 free_percpu(mp->m_sb_cnts);
David Chinnere8234a62006-03-14 13:23:52 +11002238 }
David Chinner03135cf2007-02-10 18:35:15 +11002239 mutex_destroy(&mp->m_icsb_mutex);
David Chinner8d280b92006-03-14 13:13:09 +11002240}
2241
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00002242STATIC void
David Chinner01e1b692006-03-14 13:29:16 +11002243xfs_icsb_lock_cntr(
2244 xfs_icsb_cnts_t *icsbp)
2245{
2246 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2247 ndelay(1000);
2248 }
2249}
2250
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00002251STATIC void
David Chinner01e1b692006-03-14 13:29:16 +11002252xfs_icsb_unlock_cntr(
2253 xfs_icsb_cnts_t *icsbp)
2254{
2255 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2256}
2257
David Chinner8d280b92006-03-14 13:13:09 +11002258
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00002259STATIC void
David Chinner8d280b92006-03-14 13:13:09 +11002260xfs_icsb_lock_all_counters(
2261 xfs_mount_t *mp)
2262{
2263 xfs_icsb_cnts_t *cntp;
2264 int i;
2265
2266 for_each_online_cpu(i) {
2267 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11002268 xfs_icsb_lock_cntr(cntp);
David Chinner8d280b92006-03-14 13:13:09 +11002269 }
2270}
2271
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00002272STATIC void
David Chinner8d280b92006-03-14 13:13:09 +11002273xfs_icsb_unlock_all_counters(
2274 xfs_mount_t *mp)
2275{
2276 xfs_icsb_cnts_t *cntp;
2277 int i;
2278
2279 for_each_online_cpu(i) {
2280 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11002281 xfs_icsb_unlock_cntr(cntp);
David Chinner8d280b92006-03-14 13:13:09 +11002282 }
2283}
2284
2285STATIC void
2286xfs_icsb_count(
2287 xfs_mount_t *mp,
2288 xfs_icsb_cnts_t *cnt,
2289 int flags)
2290{
2291 xfs_icsb_cnts_t *cntp;
2292 int i;
2293
2294 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2295
2296 if (!(flags & XFS_ICSB_LAZY_COUNT))
2297 xfs_icsb_lock_all_counters(mp);
2298
2299 for_each_online_cpu(i) {
2300 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2301 cnt->icsb_icount += cntp->icsb_icount;
2302 cnt->icsb_ifree += cntp->icsb_ifree;
2303 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2304 }
2305
2306 if (!(flags & XFS_ICSB_LAZY_COUNT))
2307 xfs_icsb_unlock_all_counters(mp);
2308}
2309
2310STATIC int
2311xfs_icsb_counter_disabled(
2312 xfs_mount_t *mp,
2313 xfs_sb_field_t field)
2314{
2315 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2316 return test_bit(field, &mp->m_icsb_counters);
2317}
2318
David Chinner36fbe6e2008-04-10 12:19:56 +10002319STATIC void
David Chinner8d280b92006-03-14 13:13:09 +11002320xfs_icsb_disable_counter(
2321 xfs_mount_t *mp,
2322 xfs_sb_field_t field)
2323{
2324 xfs_icsb_cnts_t cnt;
2325
2326 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2327
David Chinner20b64282007-02-10 18:35:09 +11002328 /*
2329 * If we are already disabled, then there is nothing to do
2330 * here. We check before locking all the counters to avoid
2331 * the expensive lock operation when being called in the
2332 * slow path and the counter is already disabled. This is
2333 * safe because the only time we set or clear this state is under
2334 * the m_icsb_mutex.
2335 */
2336 if (xfs_icsb_counter_disabled(mp, field))
David Chinner36fbe6e2008-04-10 12:19:56 +10002337 return;
David Chinner20b64282007-02-10 18:35:09 +11002338
David Chinner8d280b92006-03-14 13:13:09 +11002339 xfs_icsb_lock_all_counters(mp);
2340 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2341 /* drain back to superblock */
2342
Christoph Hellwigce461932008-04-22 17:34:50 +10002343 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
David Chinner8d280b92006-03-14 13:13:09 +11002344 switch(field) {
2345 case XFS_SBS_ICOUNT:
2346 mp->m_sb.sb_icount = cnt.icsb_icount;
2347 break;
2348 case XFS_SBS_IFREE:
2349 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2350 break;
2351 case XFS_SBS_FDBLOCKS:
2352 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2353 break;
2354 default:
2355 BUG();
2356 }
2357 }
2358
2359 xfs_icsb_unlock_all_counters(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002360}
2361
2362STATIC void
2363xfs_icsb_enable_counter(
2364 xfs_mount_t *mp,
2365 xfs_sb_field_t field,
2366 uint64_t count,
2367 uint64_t resid)
2368{
2369 xfs_icsb_cnts_t *cntp;
2370 int i;
2371
2372 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2373
2374 xfs_icsb_lock_all_counters(mp);
2375 for_each_online_cpu(i) {
2376 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2377 switch (field) {
2378 case XFS_SBS_ICOUNT:
2379 cntp->icsb_icount = count + resid;
2380 break;
2381 case XFS_SBS_IFREE:
2382 cntp->icsb_ifree = count + resid;
2383 break;
2384 case XFS_SBS_FDBLOCKS:
2385 cntp->icsb_fdblocks = count + resid;
2386 break;
2387 default:
2388 BUG();
2389 break;
2390 }
2391 resid = 0;
2392 }
2393 clear_bit(field, &mp->m_icsb_counters);
2394 xfs_icsb_unlock_all_counters(mp);
2395}
2396
David Chinnerdbcabad2007-02-10 18:36:17 +11002397void
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002398xfs_icsb_sync_counters_locked(
David Chinner8d280b92006-03-14 13:13:09 +11002399 xfs_mount_t *mp,
2400 int flags)
2401{
2402 xfs_icsb_cnts_t cnt;
David Chinner8d280b92006-03-14 13:13:09 +11002403
David Chinner8d280b92006-03-14 13:13:09 +11002404 xfs_icsb_count(mp, &cnt, flags);
2405
David Chinner8d280b92006-03-14 13:13:09 +11002406 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2407 mp->m_sb.sb_icount = cnt.icsb_icount;
2408 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2409 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2410 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2411 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
David Chinner8d280b92006-03-14 13:13:09 +11002412}
2413
2414/*
2415 * Accurate update of per-cpu counters to incore superblock
2416 */
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002417void
David Chinner8d280b92006-03-14 13:13:09 +11002418xfs_icsb_sync_counters(
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002419 xfs_mount_t *mp,
2420 int flags)
David Chinner8d280b92006-03-14 13:13:09 +11002421{
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002422 spin_lock(&mp->m_sb_lock);
2423 xfs_icsb_sync_counters_locked(mp, flags);
2424 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11002425}
2426
2427/*
2428 * Balance and enable/disable counters as necessary.
2429 *
David Chinner20b64282007-02-10 18:35:09 +11002430 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2431 * chosen to be the same number as single on disk allocation chunk per CPU, and
2432 * free blocks is something far enough zero that we aren't going thrash when we
2433 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2434 * prevent looping endlessly when xfs_alloc_space asks for more than will
2435 * be distributed to a single CPU but each CPU has enough blocks to be
2436 * reenabled.
2437 *
2438 * Note that we can be called when counters are already disabled.
2439 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2440 * prevent locking every per-cpu counter needlessly.
David Chinner8d280b92006-03-14 13:13:09 +11002441 */
David Chinner20b64282007-02-10 18:35:09 +11002442
2443#define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
David Chinner4be536d2006-09-07 14:26:50 +10002444#define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
David Chinner20b64282007-02-10 18:35:09 +11002445 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
David Chinner8d280b92006-03-14 13:13:09 +11002446STATIC void
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002447xfs_icsb_balance_counter_locked(
David Chinner8d280b92006-03-14 13:13:09 +11002448 xfs_mount_t *mp,
2449 xfs_sb_field_t field,
David Chinner20b64282007-02-10 18:35:09 +11002450 int min_per_cpu)
David Chinner8d280b92006-03-14 13:13:09 +11002451{
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002452 uint64_t count, resid;
David Chinner8d280b92006-03-14 13:13:09 +11002453 int weight = num_online_cpus();
David Chinner20b64282007-02-10 18:35:09 +11002454 uint64_t min = (uint64_t)min_per_cpu;
David Chinner8d280b92006-03-14 13:13:09 +11002455
David Chinner8d280b92006-03-14 13:13:09 +11002456 /* disable counter and sync counter */
2457 xfs_icsb_disable_counter(mp, field);
2458
2459 /* update counters - first CPU gets residual*/
2460 switch (field) {
2461 case XFS_SBS_ICOUNT:
2462 count = mp->m_sb.sb_icount;
2463 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002464 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002465 return;
David Chinner8d280b92006-03-14 13:13:09 +11002466 break;
2467 case XFS_SBS_IFREE:
2468 count = mp->m_sb.sb_ifree;
2469 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002470 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002471 return;
David Chinner8d280b92006-03-14 13:13:09 +11002472 break;
2473 case XFS_SBS_FDBLOCKS:
2474 count = mp->m_sb.sb_fdblocks;
2475 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002476 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002477 return;
David Chinner8d280b92006-03-14 13:13:09 +11002478 break;
2479 default:
2480 BUG();
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002481 count = resid = 0; /* quiet, gcc */
David Chinner8d280b92006-03-14 13:13:09 +11002482 break;
2483 }
2484
2485 xfs_icsb_enable_counter(mp, field, count, resid);
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002486}
2487
2488STATIC void
2489xfs_icsb_balance_counter(
2490 xfs_mount_t *mp,
2491 xfs_sb_field_t fields,
2492 int min_per_cpu)
2493{
2494 spin_lock(&mp->m_sb_lock);
2495 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2496 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11002497}
2498
David Chinnera8272ce2007-11-23 16:28:09 +11002499STATIC int
David Chinner20b64282007-02-10 18:35:09 +11002500xfs_icsb_modify_counters(
David Chinner8d280b92006-03-14 13:13:09 +11002501 xfs_mount_t *mp,
2502 xfs_sb_field_t field,
David Chinner20f4ebf2007-02-10 18:36:10 +11002503 int64_t delta,
David Chinner20b64282007-02-10 18:35:09 +11002504 int rsvd)
David Chinner8d280b92006-03-14 13:13:09 +11002505{
2506 xfs_icsb_cnts_t *icsbp;
2507 long long lcounter; /* long counter for 64 bit fields */
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002508 int ret = 0;
David Chinner8d280b92006-03-14 13:13:09 +11002509
David Chinner20b64282007-02-10 18:35:09 +11002510 might_sleep();
David Chinner8d280b92006-03-14 13:13:09 +11002511again:
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002512 preempt_disable();
2513 icsbp = this_cpu_ptr(mp->m_sb_cnts);
David Chinner20b64282007-02-10 18:35:09 +11002514
2515 /*
2516 * if the counter is disabled, go to slow path
2517 */
David Chinner8d280b92006-03-14 13:13:09 +11002518 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2519 goto slow_path;
David Chinner20b64282007-02-10 18:35:09 +11002520 xfs_icsb_lock_cntr(icsbp);
2521 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2522 xfs_icsb_unlock_cntr(icsbp);
2523 goto slow_path;
2524 }
David Chinner8d280b92006-03-14 13:13:09 +11002525
2526 switch (field) {
2527 case XFS_SBS_ICOUNT:
2528 lcounter = icsbp->icsb_icount;
2529 lcounter += delta;
2530 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002531 goto balance_counter;
David Chinner8d280b92006-03-14 13:13:09 +11002532 icsbp->icsb_icount = lcounter;
2533 break;
2534
2535 case XFS_SBS_IFREE:
2536 lcounter = icsbp->icsb_ifree;
2537 lcounter += delta;
2538 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002539 goto balance_counter;
David Chinner8d280b92006-03-14 13:13:09 +11002540 icsbp->icsb_ifree = lcounter;
2541 break;
2542
2543 case XFS_SBS_FDBLOCKS:
2544 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2545
David Chinner4be536d2006-09-07 14:26:50 +10002546 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002547 lcounter += delta;
2548 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002549 goto balance_counter;
David Chinner4be536d2006-09-07 14:26:50 +10002550 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002551 break;
2552 default:
2553 BUG();
2554 break;
2555 }
David Chinner01e1b692006-03-14 13:29:16 +11002556 xfs_icsb_unlock_cntr(icsbp);
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002557 preempt_enable();
David Chinner8d280b92006-03-14 13:13:09 +11002558 return 0;
2559
David Chinner8d280b92006-03-14 13:13:09 +11002560slow_path:
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002561 preempt_enable();
David Chinner20b64282007-02-10 18:35:09 +11002562
2563 /*
2564 * serialise with a mutex so we don't burn lots of cpu on
2565 * the superblock lock. We still need to hold the superblock
2566 * lock, however, when we modify the global structures.
2567 */
David Chinner03135cf2007-02-10 18:35:15 +11002568 xfs_icsb_lock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002569
2570 /*
2571 * Now running atomically.
2572 *
2573 * If the counter is enabled, someone has beaten us to rebalancing.
2574 * Drop the lock and try again in the fast path....
2575 */
2576 if (!(xfs_icsb_counter_disabled(mp, field))) {
David Chinner03135cf2007-02-10 18:35:15 +11002577 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002578 goto again;
2579 }
2580
2581 /*
2582 * The counter is currently disabled. Because we are
2583 * running atomically here, we know a rebalance cannot
2584 * be in progress. Hence we can go straight to operating
2585 * on the global superblock. We do not call xfs_mod_incore_sb()
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002586 * here even though we need to get the m_sb_lock. Doing so
David Chinner20b64282007-02-10 18:35:09 +11002587 * will cause us to re-enter this function and deadlock.
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002588 * Hence we get the m_sb_lock ourselves and then call
David Chinner20b64282007-02-10 18:35:09 +11002589 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2590 * directly on the global counters.
2591 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002592 spin_lock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11002593 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002594 spin_unlock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11002595
2596 /*
2597 * Now that we've modified the global superblock, we
2598 * may be able to re-enable the distributed counters
2599 * (e.g. lots of space just got freed). After that
2600 * we are done.
2601 */
2602 if (ret != ENOSPC)
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002603 xfs_icsb_balance_counter(mp, field, 0);
David Chinner03135cf2007-02-10 18:35:15 +11002604 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002605 return ret;
2606
2607balance_counter:
David Chinner01e1b692006-03-14 13:29:16 +11002608 xfs_icsb_unlock_cntr(icsbp);
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002609 preempt_enable();
David Chinner8d280b92006-03-14 13:13:09 +11002610
David Chinner20b64282007-02-10 18:35:09 +11002611 /*
2612 * We may have multiple threads here if multiple per-cpu
2613 * counters run dry at the same time. This will mean we can
2614 * do more balances than strictly necessary but it is not
2615 * the common slowpath case.
2616 */
David Chinner03135cf2007-02-10 18:35:15 +11002617 xfs_icsb_lock(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002618
David Chinner20b64282007-02-10 18:35:09 +11002619 /*
2620 * running atomically.
2621 *
2622 * This will leave the counter in the correct state for future
2623 * accesses. After the rebalance, we simply try again and our retry
2624 * will either succeed through the fast path or slow path without
2625 * another balance operation being required.
2626 */
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002627 xfs_icsb_balance_counter(mp, field, delta);
David Chinner03135cf2007-02-10 18:35:15 +11002628 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002629 goto again;
David Chinner8d280b92006-03-14 13:13:09 +11002630}
2631
David Chinner8d280b92006-03-14 13:13:09 +11002632#endif