blob: 69f62d8b2816c52389f4947c9d86b3958ccf57a6 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_dir2.h"
28#include "xfs_dmapi.h"
29#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110031#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_dir2_sf.h"
Nathan Scotta844f452005-11-02 14:38:42 +110034#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_dinode.h"
36#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110037#include "xfs_btree.h"
38#include "xfs_ialloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_alloc.h"
40#include "xfs_rtalloc.h"
41#include "xfs_bmap.h"
42#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include "xfs_rw.h"
44#include "xfs_quota.h"
45#include "xfs_fsops.h"
Christoph Hellwig43355092008-03-27 18:01:08 +110046#include "xfs_utils.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000047#include "xfs_trace.h"
48
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Christoph Hellwigba0f32d2005-06-21 15:36:52 +100050STATIC void xfs_unmountfs_wait(xfs_mount_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
David Chinner8d280b92006-03-14 13:13:09 +110052
53#ifdef HAVE_PERCPU_SB
David Chinner20f4ebf2007-02-10 18:36:10 +110054STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
Christoph Hellwig45af6c62008-04-22 17:34:44 +100055 int);
56STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t,
57 int);
David Chinner8d280b92006-03-14 13:13:09 +110058STATIC int xfs_icsb_modify_counters(xfs_mount_t *, xfs_sb_field_t,
David Chinner20f4ebf2007-02-10 18:36:10 +110059 int64_t, int);
David Chinner36fbe6e2008-04-10 12:19:56 +100060STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
David Chinner8d280b92006-03-14 13:13:09 +110061
62#else
63
Christoph Hellwig45af6c62008-04-22 17:34:44 +100064#define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
65#define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
David Chinner8d280b92006-03-14 13:13:09 +110066#define xfs_icsb_modify_counters(mp, a, b, c) do { } while (0)
David Chinner8d280b92006-03-14 13:13:09 +110067
68#endif
69
Christoph Hellwig1df84c92006-01-11 15:29:52 +110070static const struct {
David Chinner8d280b92006-03-14 13:13:09 +110071 short offset;
72 short type; /* 0 = integer
73 * 1 = binary / string (no translation)
74 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070075} xfs_sb_info[] = {
76 { offsetof(xfs_sb_t, sb_magicnum), 0 },
77 { offsetof(xfs_sb_t, sb_blocksize), 0 },
78 { offsetof(xfs_sb_t, sb_dblocks), 0 },
79 { offsetof(xfs_sb_t, sb_rblocks), 0 },
80 { offsetof(xfs_sb_t, sb_rextents), 0 },
81 { offsetof(xfs_sb_t, sb_uuid), 1 },
82 { offsetof(xfs_sb_t, sb_logstart), 0 },
83 { offsetof(xfs_sb_t, sb_rootino), 0 },
84 { offsetof(xfs_sb_t, sb_rbmino), 0 },
85 { offsetof(xfs_sb_t, sb_rsumino), 0 },
86 { offsetof(xfs_sb_t, sb_rextsize), 0 },
87 { offsetof(xfs_sb_t, sb_agblocks), 0 },
88 { offsetof(xfs_sb_t, sb_agcount), 0 },
89 { offsetof(xfs_sb_t, sb_rbmblocks), 0 },
90 { offsetof(xfs_sb_t, sb_logblocks), 0 },
91 { offsetof(xfs_sb_t, sb_versionnum), 0 },
92 { offsetof(xfs_sb_t, sb_sectsize), 0 },
93 { offsetof(xfs_sb_t, sb_inodesize), 0 },
94 { offsetof(xfs_sb_t, sb_inopblock), 0 },
95 { offsetof(xfs_sb_t, sb_fname[0]), 1 },
96 { offsetof(xfs_sb_t, sb_blocklog), 0 },
97 { offsetof(xfs_sb_t, sb_sectlog), 0 },
98 { offsetof(xfs_sb_t, sb_inodelog), 0 },
99 { offsetof(xfs_sb_t, sb_inopblog), 0 },
100 { offsetof(xfs_sb_t, sb_agblklog), 0 },
101 { offsetof(xfs_sb_t, sb_rextslog), 0 },
102 { offsetof(xfs_sb_t, sb_inprogress), 0 },
103 { offsetof(xfs_sb_t, sb_imax_pct), 0 },
104 { offsetof(xfs_sb_t, sb_icount), 0 },
105 { offsetof(xfs_sb_t, sb_ifree), 0 },
106 { offsetof(xfs_sb_t, sb_fdblocks), 0 },
107 { offsetof(xfs_sb_t, sb_frextents), 0 },
108 { offsetof(xfs_sb_t, sb_uquotino), 0 },
109 { offsetof(xfs_sb_t, sb_gquotino), 0 },
110 { offsetof(xfs_sb_t, sb_qflags), 0 },
111 { offsetof(xfs_sb_t, sb_flags), 0 },
112 { offsetof(xfs_sb_t, sb_shared_vn), 0 },
113 { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
114 { offsetof(xfs_sb_t, sb_unit), 0 },
115 { offsetof(xfs_sb_t, sb_width), 0 },
116 { offsetof(xfs_sb_t, sb_dirblklog), 0 },
117 { offsetof(xfs_sb_t, sb_logsectlog), 0 },
118 { offsetof(xfs_sb_t, sb_logsectsize),0 },
119 { offsetof(xfs_sb_t, sb_logsunit), 0 },
120 { offsetof(xfs_sb_t, sb_features2), 0 },
David Chinneree1c0902008-03-06 13:45:50 +1100121 { offsetof(xfs_sb_t, sb_bad_features2), 0 },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 { sizeof(xfs_sb_t), 0 }
123};
124
Christoph Hellwig27174202009-03-30 10:21:31 +0200125static DEFINE_MUTEX(xfs_uuid_table_mutex);
126static int xfs_uuid_table_size;
127static uuid_t *xfs_uuid_table;
128
129/*
130 * See if the UUID is unique among mounted XFS filesystems.
131 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
132 */
133STATIC int
134xfs_uuid_mount(
135 struct xfs_mount *mp)
136{
137 uuid_t *uuid = &mp->m_sb.sb_uuid;
138 int hole, i;
139
140 if (mp->m_flags & XFS_MOUNT_NOUUID)
141 return 0;
142
143 if (uuid_is_nil(uuid)) {
144 cmn_err(CE_WARN,
145 "XFS: Filesystem %s has nil UUID - can't mount",
146 mp->m_fsname);
147 return XFS_ERROR(EINVAL);
148 }
149
150 mutex_lock(&xfs_uuid_table_mutex);
151 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
152 if (uuid_is_nil(&xfs_uuid_table[i])) {
153 hole = i;
154 continue;
155 }
156 if (uuid_equal(uuid, &xfs_uuid_table[i]))
157 goto out_duplicate;
158 }
159
160 if (hole < 0) {
161 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
162 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
163 xfs_uuid_table_size * sizeof(*xfs_uuid_table),
164 KM_SLEEP);
165 hole = xfs_uuid_table_size++;
166 }
167 xfs_uuid_table[hole] = *uuid;
168 mutex_unlock(&xfs_uuid_table_mutex);
169
170 return 0;
171
172 out_duplicate:
173 mutex_unlock(&xfs_uuid_table_mutex);
174 cmn_err(CE_WARN, "XFS: Filesystem %s has duplicate UUID - can't mount",
175 mp->m_fsname);
176 return XFS_ERROR(EINVAL);
177}
178
179STATIC void
180xfs_uuid_unmount(
181 struct xfs_mount *mp)
182{
183 uuid_t *uuid = &mp->m_sb.sb_uuid;
184 int i;
185
186 if (mp->m_flags & XFS_MOUNT_NOUUID)
187 return;
188
189 mutex_lock(&xfs_uuid_table_mutex);
190 for (i = 0; i < xfs_uuid_table_size; i++) {
191 if (uuid_is_nil(&xfs_uuid_table[i]))
192 continue;
193 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
194 continue;
195 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
196 break;
197 }
198 ASSERT(i < xfs_uuid_table_size);
199 mutex_unlock(&xfs_uuid_table_mutex);
200}
201
202
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203/*
Dave Chinner0fa800f2010-01-11 11:47:46 +0000204 * Reference counting access wrappers to the perag structures.
205 */
206struct xfs_perag *
207xfs_perag_get(struct xfs_mount *mp, xfs_agnumber_t agno)
208{
209 struct xfs_perag *pag;
210 int ref = 0;
211
212 spin_lock(&mp->m_perag_lock);
213 pag = radix_tree_lookup(&mp->m_perag_tree, agno);
214 if (pag) {
215 ASSERT(atomic_read(&pag->pag_ref) >= 0);
216 /* catch leaks in the positive direction during testing */
217 ASSERT(atomic_read(&pag->pag_ref) < 1000);
218 ref = atomic_inc_return(&pag->pag_ref);
219 }
220 spin_unlock(&mp->m_perag_lock);
221 trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
222 return pag;
223}
224
225void
226xfs_perag_put(struct xfs_perag *pag)
227{
228 int ref;
229
230 ASSERT(atomic_read(&pag->pag_ref) > 0);
231 ref = atomic_dec_return(&pag->pag_ref);
232 trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
233}
234
235/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 * Free up the resources associated with a mount structure. Assume that
237 * the structure was initially zeroed, so we can tell which fields got
238 * initialized.
239 */
Christoph Hellwigc962fb72008-05-20 15:10:52 +1000240STATIC void
Christoph Hellwigff4f0382008-08-13 16:50:47 +1000241xfs_free_perag(
Christoph Hellwig745f6912007-08-30 17:20:39 +1000242 xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243{
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000244 xfs_agnumber_t agno;
245 struct xfs_perag *pag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000247 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
248 spin_lock(&mp->m_perag_lock);
249 pag = radix_tree_delete(&mp->m_perag_tree, agno);
Dave Chinnere57336f2010-01-11 11:47:49 +0000250 ASSERT(pag);
Dave Chinneraed3bb92010-01-11 11:47:45 +0000251 ASSERT(atomic_read(&pag->pag_ref) == 0);
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000252 spin_unlock(&mp->m_perag_lock);
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000253 kmem_free(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255}
256
Nathan Scott4cc929e2007-05-14 18:24:02 +1000257/*
258 * Check size of device based on the (data/realtime) block count.
259 * Note: this check is used by the growfs code as well as mount.
260 */
261int
262xfs_sb_validate_fsb_count(
263 xfs_sb_t *sbp,
264 __uint64_t nblocks)
265{
266 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
267 ASSERT(sbp->sb_blocklog >= BBSHIFT);
268
269#if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
270 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
Eric Sandeen657a4cf2010-04-30 03:42:49 +0000271 return EFBIG;
Nathan Scott4cc929e2007-05-14 18:24:02 +1000272#else /* Limited by UINT_MAX of sectors */
273 if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
Eric Sandeen657a4cf2010-04-30 03:42:49 +0000274 return EFBIG;
Nathan Scott4cc929e2007-05-14 18:24:02 +1000275#endif
276 return 0;
277}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
279/*
280 * Check the validity of the SB found.
281 */
282STATIC int
283xfs_mount_validate_sb(
284 xfs_mount_t *mp,
Nathan Scott764d1f82006-03-31 13:04:17 +1000285 xfs_sb_t *sbp,
286 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287{
288 /*
289 * If the log device and data device have the
290 * same device number, the log is internal.
291 * Consequently, the sb_logstart should be non-zero. If
292 * we have a zero sb_logstart in this case, we may be trying to mount
293 * a volume filesystem in a non-volume manner.
294 */
295 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000296 xfs_fs_mount_cmn_err(flags, "bad magic number");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 return XFS_ERROR(EWRONGFS);
298 }
299
Eric Sandeen62118702008-03-06 13:44:28 +1100300 if (!xfs_sb_good_version(sbp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000301 xfs_fs_mount_cmn_err(flags, "bad version");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 return XFS_ERROR(EWRONGFS);
303 }
304
305 if (unlikely(
306 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000307 xfs_fs_mount_cmn_err(flags,
308 "filesystem is marked as having an external log; "
309 "specify logdev on the\nmount command line.");
310 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 }
312
313 if (unlikely(
314 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000315 xfs_fs_mount_cmn_err(flags,
316 "filesystem is marked as having an internal log; "
317 "do not specify logdev on\nthe mount command line.");
318 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 }
320
321 /*
322 * More sanity checking. These were stolen directly from
323 * xfs_repair.
324 */
325 if (unlikely(
326 sbp->sb_agcount <= 0 ||
327 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
328 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
329 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
330 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
Olaf Weber2ac00af2009-04-17 16:12:45 -0500331 sbp->sb_sectsize != (1 << sbp->sb_sectlog) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
333 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
334 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
335 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
Olaf Weber2ac00af2009-04-17 16:12:45 -0500336 sbp->sb_blocksize != (1 << sbp->sb_blocklog) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
338 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
Nathan Scott9f989c92006-03-14 13:29:32 +1100339 sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
340 sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
Olaf Weber2ac00af2009-04-17 16:12:45 -0500341 sbp->sb_inodesize != (1 << sbp->sb_inodelog) ||
Nathan Scott9f989c92006-03-14 13:29:32 +1100342 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
344 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
Nathan Scotte50bd162006-04-11 15:10:45 +1000345 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000346 xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 return XFS_ERROR(EFSCORRUPTED);
348 }
349
350 /*
351 * Sanity check AG count, size fields against data size field
352 */
353 if (unlikely(
354 sbp->sb_dblocks == 0 ||
355 sbp->sb_dblocks >
356 (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
357 sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
358 sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000359 xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 return XFS_ERROR(EFSCORRUPTED);
361 }
362
Lachlan McIlroy2edbddd2008-06-27 13:34:34 +1000363 /*
364 * Until this is fixed only page-sized or smaller data blocks work.
365 */
366 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
367 xfs_fs_mount_cmn_err(flags,
368 "file system with blocksize %d bytes",
369 sbp->sb_blocksize);
370 xfs_fs_mount_cmn_err(flags,
371 "only pagesize (%ld) or less will currently work.",
372 PAGE_SIZE);
373 return XFS_ERROR(ENOSYS);
374 }
375
Christoph Hellwig1a5902c2009-03-29 19:26:46 +0200376 /*
377 * Currently only very few inode sizes are supported.
378 */
379 switch (sbp->sb_inodesize) {
380 case 256:
381 case 512:
382 case 1024:
383 case 2048:
384 break;
385 default:
386 xfs_fs_mount_cmn_err(flags,
387 "inode size of %d bytes not supported",
388 sbp->sb_inodesize);
389 return XFS_ERROR(ENOSYS);
390 }
391
Nathan Scott4cc929e2007-05-14 18:24:02 +1000392 if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
393 xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000394 xfs_fs_mount_cmn_err(flags,
395 "file system too large to be mounted on this system.");
Eric Sandeen657a4cf2010-04-30 03:42:49 +0000396 return XFS_ERROR(EFBIG);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 }
398
399 if (unlikely(sbp->sb_inprogress)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000400 xfs_fs_mount_cmn_err(flags, "file system busy");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 return XFS_ERROR(EFSCORRUPTED);
402 }
403
404 /*
Nathan Scottde206142005-05-05 13:24:13 -0700405 * Version 1 directory format has never worked on Linux.
406 */
Eric Sandeen62118702008-03-06 13:44:28 +1100407 if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000408 xfs_fs_mount_cmn_err(flags,
409 "file system using version 1 directory format");
Nathan Scottde206142005-05-05 13:24:13 -0700410 return XFS_ERROR(ENOSYS);
411 }
412
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 return 0;
414}
415
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000416int
Nathan Scottc11e2c32005-11-02 15:11:45 +1100417xfs_initialize_perag(
Nathan Scottc11e2c32005-11-02 15:11:45 +1100418 xfs_mount_t *mp,
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000419 xfs_agnumber_t agcount,
420 xfs_agnumber_t *maxagi)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421{
422 xfs_agnumber_t index, max_metadata;
Dave Chinner8b26c582010-01-11 11:47:48 +0000423 xfs_agnumber_t first_initialised = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 xfs_perag_t *pag;
425 xfs_agino_t agino;
426 xfs_ino_t ino;
427 xfs_sb_t *sbp = &mp->m_sb;
Dave Chinner8b26c582010-01-11 11:47:48 +0000428 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000430 /*
431 * Walk the current per-ag tree so we don't try to initialise AGs
432 * that already exist (growfs case). Allocate and insert all the
433 * AGs we don't find ready for initialisation.
434 */
435 for (index = 0; index < agcount; index++) {
436 pag = xfs_perag_get(mp, index);
437 if (pag) {
438 xfs_perag_put(pag);
439 continue;
440 }
Dave Chinner8b26c582010-01-11 11:47:48 +0000441 if (!first_initialised)
442 first_initialised = index;
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000443
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000444 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
445 if (!pag)
Dave Chinner8b26c582010-01-11 11:47:48 +0000446 goto out_unwind;
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000447 pag->pag_agno = index;
448 pag->pag_mount = mp;
449 rwlock_init(&pag->pag_ici_lock);
450 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
451
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000452 if (radix_tree_preload(GFP_NOFS))
Dave Chinner8b26c582010-01-11 11:47:48 +0000453 goto out_unwind;
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000454
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000455 spin_lock(&mp->m_perag_lock);
456 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
457 BUG();
458 spin_unlock(&mp->m_perag_lock);
Dave Chinner8b26c582010-01-11 11:47:48 +0000459 radix_tree_preload_end();
460 error = -EEXIST;
461 goto out_unwind;
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000462 }
463 spin_unlock(&mp->m_perag_lock);
464 radix_tree_preload_end();
465 }
466
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000467 /*
468 * If we mount with the inode64 option, or no inode overflows
469 * the legacy 32-bit address space clear the inode32 option.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 */
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000471 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
472 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000474 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
475 mp->m_flags |= XFS_MOUNT_32BITINODES;
476 else
477 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
478
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000480 /*
481 * Calculate how much should be reserved for inodes to meet
482 * the max inode percentage.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 */
484 if (mp->m_maxicount) {
485 __uint64_t icount;
486
487 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
488 do_div(icount, 100);
489 icount += sbp->sb_agblocks - 1;
Eric Sandeena749ee82005-11-02 15:13:42 +1100490 do_div(icount, sbp->sb_agblocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 max_metadata = icount;
492 } else {
493 max_metadata = agcount;
494 }
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000495
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 for (index = 0; index < agcount; index++) {
497 ino = XFS_AGINO_TO_INO(mp, index, agino);
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000498 if (ino > XFS_MAXINUMBER_32) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 index++;
500 break;
501 }
502
Dave Chinner44b56e02010-01-11 11:47:43 +0000503 pag = xfs_perag_get(mp, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 pag->pagi_inodeok = 1;
505 if (index < max_metadata)
506 pag->pagf_metadata = 1;
Dave Chinner44b56e02010-01-11 11:47:43 +0000507 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 }
509 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 for (index = 0; index < agcount; index++) {
Dave Chinner44b56e02010-01-11 11:47:43 +0000511 pag = xfs_perag_get(mp, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 pag->pagi_inodeok = 1;
Dave Chinner44b56e02010-01-11 11:47:43 +0000513 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 }
515 }
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000516
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000517 if (maxagi)
518 *maxagi = index;
519 return 0;
Dave Chinner8b26c582010-01-11 11:47:48 +0000520
521out_unwind:
522 kmem_free(pag);
523 for (; index > first_initialised; index--) {
524 pag = radix_tree_delete(&mp->m_perag_tree, index);
525 kmem_free(pag);
526 }
527 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528}
529
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000530void
531xfs_sb_from_disk(
532 xfs_sb_t *to,
533 xfs_dsb_t *from)
534{
535 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
536 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
537 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
538 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
539 to->sb_rextents = be64_to_cpu(from->sb_rextents);
540 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
541 to->sb_logstart = be64_to_cpu(from->sb_logstart);
542 to->sb_rootino = be64_to_cpu(from->sb_rootino);
543 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
544 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
545 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
546 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
547 to->sb_agcount = be32_to_cpu(from->sb_agcount);
548 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
549 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
550 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
551 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
552 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
553 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
554 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
555 to->sb_blocklog = from->sb_blocklog;
556 to->sb_sectlog = from->sb_sectlog;
557 to->sb_inodelog = from->sb_inodelog;
558 to->sb_inopblog = from->sb_inopblog;
559 to->sb_agblklog = from->sb_agblklog;
560 to->sb_rextslog = from->sb_rextslog;
561 to->sb_inprogress = from->sb_inprogress;
562 to->sb_imax_pct = from->sb_imax_pct;
563 to->sb_icount = be64_to_cpu(from->sb_icount);
564 to->sb_ifree = be64_to_cpu(from->sb_ifree);
565 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
566 to->sb_frextents = be64_to_cpu(from->sb_frextents);
567 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
568 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
569 to->sb_qflags = be16_to_cpu(from->sb_qflags);
570 to->sb_flags = from->sb_flags;
571 to->sb_shared_vn = from->sb_shared_vn;
572 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
573 to->sb_unit = be32_to_cpu(from->sb_unit);
574 to->sb_width = be32_to_cpu(from->sb_width);
575 to->sb_dirblklog = from->sb_dirblklog;
576 to->sb_logsectlog = from->sb_logsectlog;
577 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
578 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
579 to->sb_features2 = be32_to_cpu(from->sb_features2);
David Chinneree1c0902008-03-06 13:45:50 +1100580 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000581}
582
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583/*
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000584 * Copy in core superblock to ondisk one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 *
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000586 * The fields argument is mask of superblock fields to copy.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 */
588void
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000589xfs_sb_to_disk(
590 xfs_dsb_t *to,
591 xfs_sb_t *from,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 __int64_t fields)
593{
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000594 xfs_caddr_t to_ptr = (xfs_caddr_t)to;
595 xfs_caddr_t from_ptr = (xfs_caddr_t)from;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 xfs_sb_field_t f;
597 int first;
598 int size;
599
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 ASSERT(fields);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 if (!fields)
602 return;
603
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 while (fields) {
605 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
606 first = xfs_sb_info[f].offset;
607 size = xfs_sb_info[f + 1].offset - first;
608
609 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
610
611 if (size == 1 || xfs_sb_info[f].type == 1) {
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000612 memcpy(to_ptr + first, from_ptr + first, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 } else {
614 switch (size) {
615 case 2:
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000616 *(__be16 *)(to_ptr + first) =
617 cpu_to_be16(*(__u16 *)(from_ptr + first));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 break;
619 case 4:
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000620 *(__be32 *)(to_ptr + first) =
621 cpu_to_be32(*(__u32 *)(from_ptr + first));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 break;
623 case 8:
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000624 *(__be64 *)(to_ptr + first) =
625 cpu_to_be64(*(__u64 *)(from_ptr + first));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 break;
627 default:
628 ASSERT(0);
629 }
630 }
631
632 fields &= ~(1LL << f);
633 }
634}
635
636/*
637 * xfs_readsb
638 *
639 * Does the initial read of the superblock.
640 */
641int
Nathan Scott764d1f82006-03-31 13:04:17 +1000642xfs_readsb(xfs_mount_t *mp, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643{
644 unsigned int sector_size;
645 unsigned int extra_flags;
646 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 int error;
648
649 ASSERT(mp->m_sb_bp == NULL);
650 ASSERT(mp->m_ddev_targp != NULL);
651
652 /*
653 * Allocate a (locked) buffer to hold the superblock.
654 * This will be kept around at all times to optimize
655 * access to the superblock.
656 */
657 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000658 extra_flags = XBF_LOCK | XBF_FS_MANAGED | XBF_MAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
Christoph Hellwig6ad112b2009-11-24 18:02:23 +0000660 bp = xfs_buf_read(mp->m_ddev_targp, XFS_SB_DADDR, BTOBB(sector_size),
661 extra_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 if (!bp || XFS_BUF_ISERROR(bp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000663 xfs_fs_mount_cmn_err(flags, "SB read failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
665 goto fail;
666 }
667 ASSERT(XFS_BUF_ISBUSY(bp));
668 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
669
670 /*
671 * Initialize the mount structure from the superblock.
672 * But first do some basic consistency checking.
673 */
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000674 xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675
Nathan Scott764d1f82006-03-31 13:04:17 +1000676 error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 if (error) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000678 xfs_fs_mount_cmn_err(flags, "SB validate failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 goto fail;
680 }
681
682 /*
683 * We must be able to do sector-sized and sector-aligned IO.
684 */
685 if (sector_size > mp->m_sb.sb_sectsize) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000686 xfs_fs_mount_cmn_err(flags,
687 "device supports only %u byte sectors (not %u)",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 sector_size, mp->m_sb.sb_sectsize);
689 error = ENOSYS;
690 goto fail;
691 }
692
693 /*
694 * If device sector size is smaller than the superblock size,
695 * re-read the superblock so the buffer is correctly sized.
696 */
697 if (sector_size < mp->m_sb.sb_sectsize) {
698 XFS_BUF_UNMANAGE(bp);
699 xfs_buf_relse(bp);
700 sector_size = mp->m_sb.sb_sectsize;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +0000701 bp = xfs_buf_read(mp->m_ddev_targp, XFS_SB_DADDR,
702 BTOBB(sector_size), extra_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 if (!bp || XFS_BUF_ISERROR(bp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000704 xfs_fs_mount_cmn_err(flags, "SB re-read failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
706 goto fail;
707 }
708 ASSERT(XFS_BUF_ISBUSY(bp));
709 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
710 }
711
Lachlan McIlroy5478eea2007-02-10 18:36:29 +1100712 /* Initialize per-cpu counters */
713 xfs_icsb_reinit_counters(mp);
David Chinner8d280b92006-03-14 13:13:09 +1100714
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 mp->m_sb_bp = bp;
716 xfs_buf_relse(bp);
717 ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
718 return 0;
719
720 fail:
721 if (bp) {
722 XFS_BUF_UNMANAGE(bp);
723 xfs_buf_relse(bp);
724 }
725 return error;
726}
727
728
729/*
730 * xfs_mount_common
731 *
732 * Mount initialization code establishing various mount
733 * fields from the superblock associated with the given
734 * mount structure
735 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000736STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
738{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 mp->m_agfrotor = mp->m_agirotor = 0;
Eric Sandeen007c61c2007-10-11 17:43:56 +1000740 spin_lock_init(&mp->m_agirotor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 mp->m_maxagi = mp->m_sb.sb_agcount;
742 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
743 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
744 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
745 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
746 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 mp->m_blockmask = sbp->sb_blocksize - 1;
748 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
749 mp->m_blockwmask = mp->m_blockwsize - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
Christoph Hellwig60197e82008-10-30 17:11:19 +1100751 mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
752 mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
753 mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
754 mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
755
756 mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
757 mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
758 mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
759 mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
760
761 mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
762 mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
763 mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
764 mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
766 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
767 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
768 sbp->sb_inopblock);
769 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
770}
David Chinner92821e22007-05-24 15:26:31 +1000771
772/*
773 * xfs_initialize_perag_data
774 *
775 * Read in each per-ag structure so we can count up the number of
776 * allocated inodes, free inodes and used filesystem blocks as this
777 * information is no longer persistent in the superblock. Once we have
778 * this information, write it into the in-core superblock structure.
779 */
780STATIC int
781xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
782{
783 xfs_agnumber_t index;
784 xfs_perag_t *pag;
785 xfs_sb_t *sbp = &mp->m_sb;
786 uint64_t ifree = 0;
787 uint64_t ialloc = 0;
788 uint64_t bfree = 0;
789 uint64_t bfreelst = 0;
790 uint64_t btree = 0;
791 int error;
David Chinner92821e22007-05-24 15:26:31 +1000792
793 for (index = 0; index < agcount; index++) {
794 /*
795 * read the agf, then the agi. This gets us
Malcolm Parsons9da096f2009-03-29 09:55:42 +0200796 * all the information we need and populates the
David Chinner92821e22007-05-24 15:26:31 +1000797 * per-ag structures for us.
798 */
799 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
800 if (error)
801 return error;
802
803 error = xfs_ialloc_pagi_init(mp, NULL, index);
804 if (error)
805 return error;
Dave Chinner44b56e02010-01-11 11:47:43 +0000806 pag = xfs_perag_get(mp, index);
David Chinner92821e22007-05-24 15:26:31 +1000807 ifree += pag->pagi_freecount;
808 ialloc += pag->pagi_count;
809 bfree += pag->pagf_freeblks;
810 bfreelst += pag->pagf_flcount;
811 btree += pag->pagf_btreeblks;
Dave Chinner44b56e02010-01-11 11:47:43 +0000812 xfs_perag_put(pag);
David Chinner92821e22007-05-24 15:26:31 +1000813 }
814 /*
815 * Overwrite incore superblock counters with just-read data
816 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +1000817 spin_lock(&mp->m_sb_lock);
David Chinner92821e22007-05-24 15:26:31 +1000818 sbp->sb_ifree = ifree;
819 sbp->sb_icount = ialloc;
820 sbp->sb_fdblocks = bfree + bfreelst + btree;
Eric Sandeen3685c2a2007-10-11 17:42:32 +1000821 spin_unlock(&mp->m_sb_lock);
David Chinner92821e22007-05-24 15:26:31 +1000822
823 /* Fixup the per-cpu counters as well. */
824 xfs_icsb_reinit_counters(mp);
825
826 return 0;
827}
828
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829/*
Eric Sandeen0771fb42007-10-12 11:03:40 +1000830 * Update alignment values based on mount options and sb values
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 */
Eric Sandeen0771fb42007-10-12 11:03:40 +1000832STATIC int
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100833xfs_update_alignment(xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 xfs_sb_t *sbp = &(mp->m_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836
Christoph Hellwig42490232008-08-13 16:49:32 +1000837 if (mp->m_dalign) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 /*
839 * If stripe unit and stripe width are not multiples
840 * of the fs blocksize turn off alignment.
841 */
842 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
843 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
844 if (mp->m_flags & XFS_MOUNT_RETERR) {
845 cmn_err(CE_WARN,
846 "XFS: alignment check 1 failed");
Eric Sandeen0771fb42007-10-12 11:03:40 +1000847 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 }
849 mp->m_dalign = mp->m_swidth = 0;
850 } else {
851 /*
852 * Convert the stripe unit and width to FSBs.
853 */
854 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
855 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
856 if (mp->m_flags & XFS_MOUNT_RETERR) {
Eric Sandeen0771fb42007-10-12 11:03:40 +1000857 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 }
859 xfs_fs_cmn_err(CE_WARN, mp,
860"stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
861 mp->m_dalign, mp->m_swidth,
862 sbp->sb_agblocks);
863
864 mp->m_dalign = 0;
865 mp->m_swidth = 0;
866 } else if (mp->m_dalign) {
867 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
868 } else {
869 if (mp->m_flags & XFS_MOUNT_RETERR) {
870 xfs_fs_cmn_err(CE_WARN, mp,
871"stripe alignment turned off: sunit(%d) less than bsize(%d)",
872 mp->m_dalign,
873 mp->m_blockmask +1);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000874 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 }
876 mp->m_swidth = 0;
877 }
878 }
879
880 /*
881 * Update superblock with new values
882 * and log changes
883 */
Eric Sandeen62118702008-03-06 13:44:28 +1100884 if (xfs_sb_version_hasdalign(sbp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 if (sbp->sb_unit != mp->m_dalign) {
886 sbp->sb_unit = mp->m_dalign;
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100887 mp->m_update_flags |= XFS_SB_UNIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 }
889 if (sbp->sb_width != mp->m_swidth) {
890 sbp->sb_width = mp->m_swidth;
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100891 mp->m_update_flags |= XFS_SB_WIDTH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 }
893 }
894 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
Eric Sandeen62118702008-03-06 13:44:28 +1100895 xfs_sb_version_hasdalign(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 mp->m_dalign = sbp->sb_unit;
897 mp->m_swidth = sbp->sb_width;
898 }
899
Eric Sandeen0771fb42007-10-12 11:03:40 +1000900 return 0;
901}
902
903/*
904 * Set the maximum inode count for this filesystem
905 */
906STATIC void
907xfs_set_maxicount(xfs_mount_t *mp)
908{
909 xfs_sb_t *sbp = &(mp->m_sb);
910 __uint64_t icount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
912 if (sbp->sb_imax_pct) {
Eric Sandeen0771fb42007-10-12 11:03:40 +1000913 /*
914 * Make sure the maximum inode count is a multiple
915 * of the units we allocate inodes in.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
918 do_div(icount, 100);
919 do_div(icount, mp->m_ialloc_blks);
920 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
921 sbp->sb_inopblog;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000922 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 mp->m_maxicount = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 }
Eric Sandeen0771fb42007-10-12 11:03:40 +1000925}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926
Eric Sandeen0771fb42007-10-12 11:03:40 +1000927/*
928 * Set the default minimum read and write sizes unless
929 * already specified in a mount option.
930 * We use smaller I/O sizes when the file system
931 * is being used for NFS service (wsync mount option).
932 */
933STATIC void
934xfs_set_rw_sizes(xfs_mount_t *mp)
935{
936 xfs_sb_t *sbp = &(mp->m_sb);
937 int readio_log, writeio_log;
938
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
940 if (mp->m_flags & XFS_MOUNT_WSYNC) {
941 readio_log = XFS_WSYNC_READIO_LOG;
942 writeio_log = XFS_WSYNC_WRITEIO_LOG;
943 } else {
944 readio_log = XFS_READIO_LOG_LARGE;
945 writeio_log = XFS_WRITEIO_LOG_LARGE;
946 }
947 } else {
948 readio_log = mp->m_readio_log;
949 writeio_log = mp->m_writeio_log;
950 }
951
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 if (sbp->sb_blocklog > readio_log) {
953 mp->m_readio_log = sbp->sb_blocklog;
954 } else {
955 mp->m_readio_log = readio_log;
956 }
957 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
958 if (sbp->sb_blocklog > writeio_log) {
959 mp->m_writeio_log = sbp->sb_blocklog;
960 } else {
961 mp->m_writeio_log = writeio_log;
962 }
963 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000964}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
Eric Sandeen0771fb42007-10-12 11:03:40 +1000966/*
967 * Set whether we're using inode alignment.
968 */
969STATIC void
970xfs_set_inoalignment(xfs_mount_t *mp)
971{
Eric Sandeen62118702008-03-06 13:44:28 +1100972 if (xfs_sb_version_hasalign(&mp->m_sb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 mp->m_sb.sb_inoalignmt >=
974 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
975 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
976 else
977 mp->m_inoalign_mask = 0;
978 /*
979 * If we are using stripe alignment, check whether
980 * the stripe unit is a multiple of the inode alignment
981 */
982 if (mp->m_dalign && mp->m_inoalign_mask &&
983 !(mp->m_dalign & mp->m_inoalign_mask))
984 mp->m_sinoalign = mp->m_dalign;
985 else
986 mp->m_sinoalign = 0;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000987}
988
989/*
990 * Check that the data (and log if separate) are an ok size.
991 */
992STATIC int
Christoph Hellwig42490232008-08-13 16:49:32 +1000993xfs_check_sizes(xfs_mount_t *mp)
Eric Sandeen0771fb42007-10-12 11:03:40 +1000994{
995 xfs_buf_t *bp;
996 xfs_daddr_t d;
997 int error;
998
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
1000 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
1001 cmn_err(CE_WARN, "XFS: size check 1 failed");
Eric Sandeen657a4cf2010-04-30 03:42:49 +00001002 return XFS_ERROR(EFBIG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 }
1004 error = xfs_read_buf(mp, mp->m_ddev_targp,
1005 d - XFS_FSS_TO_BB(mp, 1),
1006 XFS_FSS_TO_BB(mp, 1), 0, &bp);
1007 if (!error) {
1008 xfs_buf_relse(bp);
1009 } else {
1010 cmn_err(CE_WARN, "XFS: size check 2 failed");
Eric Sandeen0771fb42007-10-12 11:03:40 +10001011 if (error == ENOSPC)
Eric Sandeen657a4cf2010-04-30 03:42:49 +00001012 error = XFS_ERROR(EFBIG);
Eric Sandeen0771fb42007-10-12 11:03:40 +10001013 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 }
1015
Christoph Hellwig42490232008-08-13 16:49:32 +10001016 if (mp->m_logdev_targp != mp->m_ddev_targp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
1018 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
1019 cmn_err(CE_WARN, "XFS: size check 3 failed");
Eric Sandeen657a4cf2010-04-30 03:42:49 +00001020 return XFS_ERROR(EFBIG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 }
1022 error = xfs_read_buf(mp, mp->m_logdev_targp,
1023 d - XFS_FSB_TO_BB(mp, 1),
1024 XFS_FSB_TO_BB(mp, 1), 0, &bp);
1025 if (!error) {
1026 xfs_buf_relse(bp);
1027 } else {
1028 cmn_err(CE_WARN, "XFS: size check 3 failed");
Eric Sandeen0771fb42007-10-12 11:03:40 +10001029 if (error == ENOSPC)
Eric Sandeen657a4cf2010-04-30 03:42:49 +00001030 error = XFS_ERROR(EFBIG);
Eric Sandeen0771fb42007-10-12 11:03:40 +10001031 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 }
1033 }
Eric Sandeen0771fb42007-10-12 11:03:40 +10001034 return 0;
1035}
1036
1037/*
Christoph Hellwig7d095252009-06-08 15:33:32 +02001038 * Clear the quotaflags in memory and in the superblock.
1039 */
1040int
1041xfs_mount_reset_sbqflags(
1042 struct xfs_mount *mp)
1043{
1044 int error;
1045 struct xfs_trans *tp;
1046
1047 mp->m_qflags = 0;
1048
1049 /*
1050 * It is OK to look at sb_qflags here in mount path,
1051 * without m_sb_lock.
1052 */
1053 if (mp->m_sb.sb_qflags == 0)
1054 return 0;
1055 spin_lock(&mp->m_sb_lock);
1056 mp->m_sb.sb_qflags = 0;
1057 spin_unlock(&mp->m_sb_lock);
1058
1059 /*
1060 * If the fs is readonly, let the incore superblock run
1061 * with quotas off but don't flush the update out to disk
1062 */
1063 if (mp->m_flags & XFS_MOUNT_RDONLY)
1064 return 0;
1065
1066#ifdef QUOTADEBUG
1067 xfs_fs_cmn_err(CE_NOTE, mp, "Writing superblock quota changes");
1068#endif
1069
1070 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1071 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1072 XFS_DEFAULT_LOG_COUNT);
1073 if (error) {
1074 xfs_trans_cancel(tp, 0);
1075 xfs_fs_cmn_err(CE_ALERT, mp,
1076 "xfs_mount_reset_sbqflags: Superblock update failed!");
1077 return error;
1078 }
1079
1080 xfs_mod_sb(tp, XFS_SB_QFLAGS);
1081 return xfs_trans_commit(tp, 0);
1082}
1083
Eric Sandeend5db0f92010-02-05 22:59:53 +00001084__uint64_t
1085xfs_default_resblks(xfs_mount_t *mp)
1086{
1087 __uint64_t resblks;
1088
1089 /*
Dave Chinner8babd8a2010-03-04 01:46:25 +00001090 * We default to 5% or 8192 fsbs of space reserved, whichever is
1091 * smaller. This is intended to cover concurrent allocation
1092 * transactions when we initially hit enospc. These each require a 4
1093 * block reservation. Hence by default we cover roughly 2000 concurrent
1094 * allocation reservations.
Eric Sandeend5db0f92010-02-05 22:59:53 +00001095 */
1096 resblks = mp->m_sb.sb_dblocks;
1097 do_div(resblks, 20);
Dave Chinner8babd8a2010-03-04 01:46:25 +00001098 resblks = min_t(__uint64_t, resblks, 8192);
Eric Sandeend5db0f92010-02-05 22:59:53 +00001099 return resblks;
1100}
1101
Christoph Hellwig7d095252009-06-08 15:33:32 +02001102/*
Eric Sandeen0771fb42007-10-12 11:03:40 +10001103 * This function does the following on an initial mount of a file system:
1104 * - reads the superblock from disk and init the mount struct
1105 * - if we're a 32-bit kernel, do a size check on the superblock
1106 * so we don't mount terabyte filesystems
1107 * - init mount struct realtime fields
1108 * - allocate inode hash table for fs
1109 * - init directory manager
1110 * - perform recovery and init the log manager
1111 */
1112int
1113xfs_mountfs(
Christoph Hellwig42490232008-08-13 16:49:32 +10001114 xfs_mount_t *mp)
Eric Sandeen0771fb42007-10-12 11:03:40 +10001115{
1116 xfs_sb_t *sbp = &(mp->m_sb);
1117 xfs_inode_t *rip;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001118 __uint64_t resblks;
Christoph Hellwig7d095252009-06-08 15:33:32 +02001119 uint quotamount = 0;
1120 uint quotaflags = 0;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001121 int error = 0;
1122
Eric Sandeen0771fb42007-10-12 11:03:40 +10001123 xfs_mount_common(mp, sbp);
1124
1125 /*
Eric Sandeene6957ea2008-04-10 12:19:34 +10001126 * Check for a mismatched features2 values. Older kernels
1127 * read & wrote into the wrong sb offset for sb_features2
1128 * on some platforms due to xfs_sb_t not being 64bit size aligned
1129 * when sb_features2 was added, which made older superblock
1130 * reading/writing routines swap it as a 64-bit value.
David Chinneree1c0902008-03-06 13:45:50 +11001131 *
Eric Sandeene6957ea2008-04-10 12:19:34 +10001132 * For backwards compatibility, we make both slots equal.
1133 *
1134 * If we detect a mismatched field, we OR the set bits into the
1135 * existing features2 field in case it has already been modified; we
1136 * don't want to lose any features. We then update the bad location
1137 * with the ORed value so that older kernels will see any features2
1138 * flags, and mark the two fields as needing updates once the
1139 * transaction subsystem is online.
David Chinneree1c0902008-03-06 13:45:50 +11001140 */
Eric Sandeene6957ea2008-04-10 12:19:34 +10001141 if (xfs_sb_has_mismatched_features2(sbp)) {
David Chinneree1c0902008-03-06 13:45:50 +11001142 cmn_err(CE_WARN,
1143 "XFS: correcting sb_features alignment problem");
1144 sbp->sb_features2 |= sbp->sb_bad_features2;
Eric Sandeene6957ea2008-04-10 12:19:34 +10001145 sbp->sb_bad_features2 = sbp->sb_features2;
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001146 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
Eric Sandeene6957ea2008-04-10 12:19:34 +10001147
1148 /*
1149 * Re-check for ATTR2 in case it was found in bad_features2
1150 * slot.
1151 */
Tim Shimmin7c12f292008-04-30 18:15:28 +10001152 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1153 !(mp->m_flags & XFS_MOUNT_NOATTR2))
Eric Sandeene6957ea2008-04-10 12:19:34 +10001154 mp->m_flags |= XFS_MOUNT_ATTR2;
Tim Shimmin7c12f292008-04-30 18:15:28 +10001155 }
Eric Sandeene6957ea2008-04-10 12:19:34 +10001156
Tim Shimmin7c12f292008-04-30 18:15:28 +10001157 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1158 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1159 xfs_sb_version_removeattr2(&mp->m_sb);
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001160 mp->m_update_flags |= XFS_SB_FEATURES2;
Tim Shimmin7c12f292008-04-30 18:15:28 +10001161
1162 /* update sb_versionnum for the clearing of the morebits */
1163 if (!sbp->sb_features2)
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001164 mp->m_update_flags |= XFS_SB_VERSIONNUM;
David Chinneree1c0902008-03-06 13:45:50 +11001165 }
1166
1167 /*
Eric Sandeen0771fb42007-10-12 11:03:40 +10001168 * Check if sb_agblocks is aligned at stripe boundary
1169 * If sb_agblocks is NOT aligned turn off m_dalign since
1170 * allocator alignment is within an ag, therefore ag has
1171 * to be aligned at stripe boundary.
1172 */
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001173 error = xfs_update_alignment(mp);
Eric Sandeen0771fb42007-10-12 11:03:40 +10001174 if (error)
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001175 goto out;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001176
1177 xfs_alloc_compute_maxlevels(mp);
1178 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
1179 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
1180 xfs_ialloc_compute_maxlevels(mp);
1181
1182 xfs_set_maxicount(mp);
1183
1184 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
1185
Christoph Hellwig27174202009-03-30 10:21:31 +02001186 error = xfs_uuid_mount(mp);
1187 if (error)
1188 goto out;
Eric Sandeen0771fb42007-10-12 11:03:40 +10001189
1190 /*
1191 * Set the minimum read and write sizes
1192 */
1193 xfs_set_rw_sizes(mp);
1194
1195 /*
1196 * Set the inode cluster size.
1197 * This may still be overridden by the file system
1198 * block size if it is larger than the chosen cluster size.
1199 */
1200 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1201
1202 /*
1203 * Set inode alignment fields
1204 */
1205 xfs_set_inoalignment(mp);
1206
1207 /*
1208 * Check that the data (and log if separate) are an ok size.
1209 */
Christoph Hellwig42490232008-08-13 16:49:32 +10001210 error = xfs_check_sizes(mp);
Eric Sandeen0771fb42007-10-12 11:03:40 +10001211 if (error)
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001212 goto out_remove_uuid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213
1214 /*
1215 * Initialize realtime fields in the mount structure
1216 */
Eric Sandeen0771fb42007-10-12 11:03:40 +10001217 error = xfs_rtmount_init(mp);
1218 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 cmn_err(CE_WARN, "XFS: RT mount failed");
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001220 goto out_remove_uuid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 }
1222
1223 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 * Copies the low order bits of the timestamp and the randomly
1225 * set "sequence" number out of a UUID.
1226 */
1227 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1228
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 mp->m_dmevmask = 0; /* not persistent; set after each mount */
1230
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001231 xfs_dir_mount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
1233 /*
1234 * Initialize the attribute manager's entries.
1235 */
1236 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1237
1238 /*
1239 * Initialize the precomputed transaction reservations values.
1240 */
1241 xfs_trans_init(mp);
1242
1243 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 * Allocate and initialize the per-ag data.
1245 */
Dave Chinner1c1c6eb2010-01-11 11:47:44 +00001246 spin_lock_init(&mp->m_perag_lock);
Dave Chinner9b98b6f2010-05-27 01:58:13 +00001247 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
Dave Chinner1c1c6eb2010-01-11 11:47:44 +00001248 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
1249 if (error) {
1250 cmn_err(CE_WARN, "XFS: Failed per-ag init: %d", error);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001251 goto out_remove_uuid;
Dave Chinner1c1c6eb2010-01-11 11:47:44 +00001252 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001254 if (!sbp->sb_logblocks) {
1255 cmn_err(CE_WARN, "XFS: no log defined");
1256 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
1257 error = XFS_ERROR(EFSCORRUPTED);
1258 goto out_free_perag;
1259 }
1260
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 /*
1262 * log's mount-time initialization. Perform 1st part recovery if needed
1263 */
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001264 error = xfs_log_mount(mp, mp->m_logdev_targp,
1265 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1266 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1267 if (error) {
1268 cmn_err(CE_WARN, "XFS: log mount failed");
1269 goto out_free_perag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 }
1271
1272 /*
David Chinner92821e22007-05-24 15:26:31 +10001273 * Now the log is mounted, we know if it was an unclean shutdown or
1274 * not. If it was, with the first phase of recovery has completed, we
1275 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1276 * but they are recovered transactionally in the second recovery phase
1277 * later.
1278 *
1279 * Hence we can safely re-initialise incore superblock counters from
1280 * the per-ag data. These may not be correct if the filesystem was not
1281 * cleanly unmounted, so we need to wait for recovery to finish before
1282 * doing this.
1283 *
1284 * If the filesystem was cleanly unmounted, then we can trust the
1285 * values in the superblock to be correct and we don't need to do
1286 * anything here.
1287 *
1288 * If we are currently making the filesystem, the initialisation will
1289 * fail as the perag data is in an undefined state.
1290 */
David Chinner92821e22007-05-24 15:26:31 +10001291 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1292 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1293 !mp->m_sb.sb_inprogress) {
1294 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001295 if (error)
1296 goto out_free_perag;
David Chinner92821e22007-05-24 15:26:31 +10001297 }
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001298
David Chinner92821e22007-05-24 15:26:31 +10001299 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 * Get and sanity-check the root inode.
1301 * Save the pointer to it in the mount structure.
1302 */
Dave Chinner7b6259e2010-06-24 11:35:17 +10001303 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 if (error) {
1305 cmn_err(CE_WARN, "XFS: failed to read root inode");
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001306 goto out_log_dealloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 }
1308
1309 ASSERT(rip != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310
1311 if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
1312 cmn_err(CE_WARN, "XFS: corrupted root inode");
Nathan Scottb6574522006-06-09 15:29:40 +10001313 cmn_err(CE_WARN, "Device %s - root %llu is not a directory",
1314 XFS_BUFTARG_NAME(mp->m_ddev_targp),
1315 (unsigned long long)rip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1317 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1318 mp);
1319 error = XFS_ERROR(EFSCORRUPTED);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001320 goto out_rele_rip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 }
1322 mp->m_rootip = rip; /* save it */
1323
1324 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1325
1326 /*
1327 * Initialize realtime inode pointers in the mount structure
1328 */
Eric Sandeen0771fb42007-10-12 11:03:40 +10001329 error = xfs_rtmount_inodes(mp);
1330 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 /*
1332 * Free up the root inode.
1333 */
1334 cmn_err(CE_WARN, "XFS: failed to read RT inodes");
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001335 goto out_rele_rip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 }
1337
1338 /*
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001339 * If this is a read-only mount defer the superblock updates until
1340 * the next remount into writeable mode. Otherwise we would never
1341 * perform the update e.g. for the root filesystem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 */
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001343 if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1344 error = xfs_mount_log_sb(mp, mp->m_update_flags);
David Chinnere5720ee2008-04-10 12:21:18 +10001345 if (error) {
1346 cmn_err(CE_WARN, "XFS: failed to write sb changes");
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001347 goto out_rtunmount;
David Chinnere5720ee2008-04-10 12:21:18 +10001348 }
1349 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
1351 /*
1352 * Initialise the XFS quota management subsystem for this mount
1353 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001354 if (XFS_IS_QUOTA_RUNNING(mp)) {
1355 error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
1356 if (error)
1357 goto out_rtunmount;
1358 } else {
1359 ASSERT(!XFS_IS_QUOTA_ON(mp));
1360
1361 /*
1362 * If a file system had quotas running earlier, but decided to
1363 * mount without -o uquota/pquota/gquota options, revoke the
1364 * quotachecked license.
1365 */
1366 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
1367 cmn_err(CE_NOTE,
1368 "XFS: resetting qflags for filesystem %s",
1369 mp->m_fsname);
1370
1371 error = xfs_mount_reset_sbqflags(mp);
1372 if (error)
1373 return error;
1374 }
1375 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376
1377 /*
1378 * Finish recovering the file system. This part needed to be
1379 * delayed until after the root and real-time bitmap inodes
1380 * were consistently read in.
1381 */
Christoph Hellwig42490232008-08-13 16:49:32 +10001382 error = xfs_log_mount_finish(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 if (error) {
1384 cmn_err(CE_WARN, "XFS: log mount finish failed");
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001385 goto out_rtunmount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 }
1387
1388 /*
1389 * Complete the quota initialisation, post-log-replay component.
1390 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001391 if (quotamount) {
1392 ASSERT(mp->m_qflags == 0);
1393 mp->m_qflags = quotaflags;
1394
1395 xfs_qm_mount_quotas(mp);
1396 }
1397
David Chinner84e1e992007-06-18 16:50:27 +10001398 /*
1399 * Now we are mounted, reserve a small amount of unused space for
1400 * privileged transactions. This is needed so that transaction
1401 * space required for critical operations can dip into this pool
1402 * when at ENOSPC. This is needed for operations like create with
1403 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1404 * are not allowed to use this reserved space.
Dave Chinner8babd8a2010-03-04 01:46:25 +00001405 *
1406 * This may drive us straight to ENOSPC on mount, but that implies
1407 * we were already there on the last unmount. Warn if this occurs.
David Chinner84e1e992007-06-18 16:50:27 +10001408 */
Eric Sandeend5db0f92010-02-05 22:59:53 +00001409 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
1410 resblks = xfs_default_resblks(mp);
1411 error = xfs_reserve_blocks(mp, &resblks, NULL);
1412 if (error)
1413 cmn_err(CE_WARN, "XFS: Unable to allocate reserve "
1414 "blocks. Continuing without a reserve pool.");
1415 }
David Chinner84e1e992007-06-18 16:50:27 +10001416
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 return 0;
1418
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001419 out_rtunmount:
1420 xfs_rtunmount_inodes(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001421 out_rele_rip:
Christoph Hellwig43355092008-03-27 18:01:08 +11001422 IRELE(rip);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001423 out_log_dealloc:
Christoph Hellwig21b699c2009-03-16 08:19:29 +01001424 xfs_log_unmount(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001425 out_free_perag:
Christoph Hellwigff4f0382008-08-13 16:50:47 +10001426 xfs_free_perag(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001427 out_remove_uuid:
Christoph Hellwig27174202009-03-30 10:21:31 +02001428 xfs_uuid_unmount(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +01001429 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 return error;
1431}
1432
1433/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 * This flushes out the inodes,dquots and the superblock, unmounts the
1435 * log and makes sure that incore structures are freed.
1436 */
Christoph Hellwig41b5c2e2008-08-13 16:49:57 +10001437void
1438xfs_unmountfs(
1439 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440{
Christoph Hellwig41b5c2e2008-08-13 16:49:57 +10001441 __uint64_t resblks;
1442 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443
Christoph Hellwig7d095252009-06-08 15:33:32 +02001444 xfs_qm_unmount_quotas(mp);
Christoph Hellwigb93b6e42009-02-04 09:33:58 +01001445 xfs_rtunmount_inodes(mp);
Christoph Hellwig77508ec2008-08-13 16:49:04 +10001446 IRELE(mp->m_rootip);
1447
David Chinner641c56f2007-06-18 16:50:17 +10001448 /*
1449 * We can potentially deadlock here if we have an inode cluster
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001450 * that has been freed has its buffer still pinned in memory because
David Chinner641c56f2007-06-18 16:50:17 +10001451 * the transaction is still sitting in a iclog. The stale inodes
1452 * on that buffer will have their flush locks held until the
1453 * transaction hits the disk and the callbacks run. the inode
1454 * flush takes the flush lock unconditionally and with nothing to
1455 * push out the iclog we will never get that unlocked. hence we
1456 * need to force the log first.
1457 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00001458 xfs_log_force(mp, XFS_LOG_SYNC);
Dave Chinnerc8543632010-02-06 12:39:36 +11001459
1460 /*
1461 * Do a delwri reclaim pass first so that as many dirty inodes are
1462 * queued up for IO as possible. Then flush the buffers before making
1463 * a synchronous path to catch all the remaining inodes are reclaimed.
1464 * This makes the reclaim process as quick as possible by avoiding
1465 * synchronous writeout and blocking on inodes already in the delwri
1466 * state as much as possible.
1467 */
1468 xfs_reclaim_inodes(mp, 0);
1469 XFS_bflush(mp->m_ddev_targp);
1470 xfs_reclaim_inodes(mp, SYNC_WAIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471
Christoph Hellwig7d095252009-06-08 15:33:32 +02001472 xfs_qm_unmount(mp);
Lachlan McIlroya357a122008-10-30 16:53:25 +11001473
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 /*
1475 * Flush out the log synchronously so that we know for sure
1476 * that nothing is pinned. This is important because bflush()
1477 * will skip pinned buffers.
1478 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00001479 xfs_log_force(mp, XFS_LOG_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480
1481 xfs_binval(mp->m_ddev_targp);
1482 if (mp->m_rtdev_targp) {
1483 xfs_binval(mp->m_rtdev_targp);
1484 }
1485
David Chinner84e1e992007-06-18 16:50:27 +10001486 /*
1487 * Unreserve any blocks we have so that when we unmount we don't account
1488 * the reserved free space as used. This is really only necessary for
1489 * lazy superblock counting because it trusts the incore superblock
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001490 * counters to be absolutely correct on clean unmount.
David Chinner84e1e992007-06-18 16:50:27 +10001491 *
1492 * We don't bother correcting this elsewhere for lazy superblock
1493 * counting because on mount of an unclean filesystem we reconstruct the
1494 * correct counter value and this is irrelevant.
1495 *
1496 * For non-lazy counter filesystems, this doesn't matter at all because
1497 * we only every apply deltas to the superblock and hence the incore
1498 * value does not matter....
1499 */
1500 resblks = 0;
David Chinner714082b2008-04-10 12:20:03 +10001501 error = xfs_reserve_blocks(mp, &resblks, NULL);
1502 if (error)
1503 cmn_err(CE_WARN, "XFS: Unable to free reserved block pool. "
1504 "Freespace may not be correct on next mount.");
1505
David Chinnere5720ee2008-04-10 12:21:18 +10001506 error = xfs_log_sbcount(mp, 1);
1507 if (error)
1508 cmn_err(CE_WARN, "XFS: Unable to update superblock counters. "
1509 "Freespace may not be correct on next mount.");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 xfs_unmountfs_writesb(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 xfs_unmountfs_wait(mp); /* wait for async bufs */
Christoph Hellwig21b699c2009-03-16 08:19:29 +01001512 xfs_log_unmount_write(mp);
1513 xfs_log_unmount(mp);
Christoph Hellwig27174202009-03-30 10:21:31 +02001514 xfs_uuid_unmount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515
Christoph Hellwig1550d0b2008-08-13 16:17:37 +10001516#if defined(DEBUG)
Christoph Hellwig0ce4cfd2007-08-30 17:20:53 +10001517 xfs_errortag_clearall(mp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518#endif
Christoph Hellwigff4f0382008-08-13 16:50:47 +10001519 xfs_free_perag(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520}
1521
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10001522STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523xfs_unmountfs_wait(xfs_mount_t *mp)
1524{
1525 if (mp->m_logdev_targp != mp->m_ddev_targp)
1526 xfs_wait_buftarg(mp->m_logdev_targp);
1527 if (mp->m_rtdev_targp)
1528 xfs_wait_buftarg(mp->m_rtdev_targp);
1529 xfs_wait_buftarg(mp->m_ddev_targp);
1530}
1531
1532int
David Chinner92821e22007-05-24 15:26:31 +10001533xfs_fs_writable(xfs_mount_t *mp)
1534{
Christoph Hellwigb267ce92007-08-30 17:21:30 +10001535 return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001536 (mp->m_flags & XFS_MOUNT_RDONLY));
David Chinner92821e22007-05-24 15:26:31 +10001537}
1538
1539/*
1540 * xfs_log_sbcount
1541 *
1542 * Called either periodically to keep the on disk superblock values
1543 * roughly up to date or from unmount to make sure the values are
1544 * correct on a clean unmount.
1545 *
1546 * Note this code can be called during the process of freezing, so
1547 * we may need to use the transaction allocator which does not not
1548 * block when the transaction subsystem is in its frozen state.
1549 */
1550int
1551xfs_log_sbcount(
1552 xfs_mount_t *mp,
1553 uint sync)
1554{
1555 xfs_trans_t *tp;
1556 int error;
1557
1558 if (!xfs_fs_writable(mp))
1559 return 0;
1560
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10001561 xfs_icsb_sync_counters(mp, 0);
David Chinner92821e22007-05-24 15:26:31 +10001562
1563 /*
1564 * we don't need to do this if we are updating the superblock
1565 * counters on every modification.
1566 */
1567 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1568 return 0;
1569
Christoph Hellwig80641dc2009-10-19 04:00:03 +00001570 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
David Chinner92821e22007-05-24 15:26:31 +10001571 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1572 XFS_DEFAULT_LOG_COUNT);
1573 if (error) {
1574 xfs_trans_cancel(tp, 0);
1575 return error;
1576 }
1577
1578 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1579 if (sync)
1580 xfs_trans_set_sync(tp);
David Chinnere5720ee2008-04-10 12:21:18 +10001581 error = xfs_trans_commit(tp, 0);
1582 return error;
David Chinner92821e22007-05-24 15:26:31 +10001583}
1584
1585int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586xfs_unmountfs_writesb(xfs_mount_t *mp)
1587{
1588 xfs_buf_t *sbp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 int error = 0;
1590
1591 /*
1592 * skip superblock write if fs is read-only, or
1593 * if we are doing a forced umount.
1594 */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001595 if (!((mp->m_flags & XFS_MOUNT_RDONLY) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 XFS_FORCED_SHUTDOWN(mp))) {
David Chinner8d280b92006-03-14 13:13:09 +11001597
David Chinner92821e22007-05-24 15:26:31 +10001598 sbp = xfs_getsb(mp, 0);
David Chinner8d280b92006-03-14 13:13:09 +11001599
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 XFS_BUF_UNDONE(sbp);
1601 XFS_BUF_UNREAD(sbp);
1602 XFS_BUF_UNDELAYWRITE(sbp);
1603 XFS_BUF_WRITE(sbp);
1604 XFS_BUF_UNASYNC(sbp);
1605 ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1606 xfsbdstrat(mp, sbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 error = xfs_iowait(sbp);
1608 if (error)
1609 xfs_ioerror_alert("xfs_unmountfs_writesb",
1610 mp, sbp, XFS_BUF_ADDR(sbp));
David Chinner92821e22007-05-24 15:26:31 +10001611 xfs_buf_relse(sbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001613 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614}
1615
1616/*
1617 * xfs_mod_sb() can be used to copy arbitrary changes to the
1618 * in-core superblock into the superblock buffer to be logged.
1619 * It does not provide the higher level of locking that is
1620 * needed to protect the in-core superblock from concurrent
1621 * access.
1622 */
1623void
1624xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1625{
1626 xfs_buf_t *bp;
1627 int first;
1628 int last;
1629 xfs_mount_t *mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 xfs_sb_field_t f;
1631
1632 ASSERT(fields);
1633 if (!fields)
1634 return;
1635 mp = tp->t_mountp;
1636 bp = xfs_trans_getsb(tp, mp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 first = sizeof(xfs_sb_t);
1638 last = 0;
1639
1640 /* translate/copy */
1641
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +10001642 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643
1644 /* find modified range */
Dave Chinner587aa0f2010-01-20 12:04:53 +11001645 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1646 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1647 last = xfs_sb_info[f + 1].offset - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648
1649 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1650 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1651 first = xfs_sb_info[f].offset;
1652
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 xfs_trans_log_buf(tp, bp, first, last);
1654}
Yingping Lud210a282006-06-09 14:55:18 +10001655
Yingping Lud210a282006-06-09 14:55:18 +10001656
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657/*
1658 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1659 * a delta to a specified field in the in-core superblock. Simply
1660 * switch on the field indicated and apply the delta to that field.
1661 * Fields are not allowed to dip below zero, so if the delta would
1662 * do this do not apply it and return EINVAL.
1663 *
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001664 * The m_sb_lock must be held when this routine is called.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05001666STATIC int
David Chinner20f4ebf2007-02-10 18:36:10 +11001667xfs_mod_incore_sb_unlocked(
1668 xfs_mount_t *mp,
1669 xfs_sb_field_t field,
1670 int64_t delta,
1671 int rsvd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672{
1673 int scounter; /* short counter for 32 bit fields */
1674 long long lcounter; /* long counter for 64 bit fields */
1675 long long res_used, rem;
1676
1677 /*
1678 * With the in-core superblock spin lock held, switch
1679 * on the indicated field. Apply the delta to the
1680 * proper field. If the fields value would dip below
1681 * 0, then do not apply the delta and return EINVAL.
1682 */
1683 switch (field) {
1684 case XFS_SBS_ICOUNT:
1685 lcounter = (long long)mp->m_sb.sb_icount;
1686 lcounter += delta;
1687 if (lcounter < 0) {
1688 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001689 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 }
1691 mp->m_sb.sb_icount = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001692 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 case XFS_SBS_IFREE:
1694 lcounter = (long long)mp->m_sb.sb_ifree;
1695 lcounter += delta;
1696 if (lcounter < 0) {
1697 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001698 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 }
1700 mp->m_sb.sb_ifree = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001701 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 case XFS_SBS_FDBLOCKS:
David Chinner4be536d2006-09-07 14:26:50 +10001703 lcounter = (long long)
1704 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1706
1707 if (delta > 0) { /* Putting blocks back */
1708 if (res_used > delta) {
1709 mp->m_resblks_avail += delta;
1710 } else {
1711 rem = delta - res_used;
1712 mp->m_resblks_avail = mp->m_resblks;
1713 lcounter += rem;
1714 }
1715 } else { /* Taking blocks away */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 lcounter += delta;
Dave Chinner8babd8a2010-03-04 01:46:25 +00001717 if (lcounter >= 0) {
1718 mp->m_sb.sb_fdblocks = lcounter +
1719 XFS_ALLOC_SET_ASIDE(mp);
1720 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 }
Dave Chinner8babd8a2010-03-04 01:46:25 +00001722
1723 /*
1724 * We are out of blocks, use any available reserved
1725 * blocks if were allowed to.
1726 */
1727 if (!rsvd)
1728 return XFS_ERROR(ENOSPC);
1729
1730 lcounter = (long long)mp->m_resblks_avail + delta;
1731 if (lcounter >= 0) {
1732 mp->m_resblks_avail = lcounter;
1733 return 0;
1734 }
1735 printk_once(KERN_WARNING
1736 "Filesystem \"%s\": reserve blocks depleted! "
1737 "Consider increasing reserve pool size.",
1738 mp->m_fsname);
1739 return XFS_ERROR(ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 }
1741
David Chinner4be536d2006-09-07 14:26:50 +10001742 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
Jesper Juhl014c2542006-01-15 02:37:08 +01001743 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 case XFS_SBS_FREXTENTS:
1745 lcounter = (long long)mp->m_sb.sb_frextents;
1746 lcounter += delta;
1747 if (lcounter < 0) {
Jesper Juhl014c2542006-01-15 02:37:08 +01001748 return XFS_ERROR(ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 }
1750 mp->m_sb.sb_frextents = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001751 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 case XFS_SBS_DBLOCKS:
1753 lcounter = (long long)mp->m_sb.sb_dblocks;
1754 lcounter += delta;
1755 if (lcounter < 0) {
1756 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001757 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 }
1759 mp->m_sb.sb_dblocks = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001760 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 case XFS_SBS_AGCOUNT:
1762 scounter = mp->m_sb.sb_agcount;
1763 scounter += delta;
1764 if (scounter < 0) {
1765 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001766 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 }
1768 mp->m_sb.sb_agcount = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001769 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 case XFS_SBS_IMAX_PCT:
1771 scounter = mp->m_sb.sb_imax_pct;
1772 scounter += delta;
1773 if (scounter < 0) {
1774 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001775 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 }
1777 mp->m_sb.sb_imax_pct = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001778 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 case XFS_SBS_REXTSIZE:
1780 scounter = mp->m_sb.sb_rextsize;
1781 scounter += delta;
1782 if (scounter < 0) {
1783 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001784 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 }
1786 mp->m_sb.sb_rextsize = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001787 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 case XFS_SBS_RBMBLOCKS:
1789 scounter = mp->m_sb.sb_rbmblocks;
1790 scounter += delta;
1791 if (scounter < 0) {
1792 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001793 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 }
1795 mp->m_sb.sb_rbmblocks = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001796 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 case XFS_SBS_RBLOCKS:
1798 lcounter = (long long)mp->m_sb.sb_rblocks;
1799 lcounter += delta;
1800 if (lcounter < 0) {
1801 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001802 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 }
1804 mp->m_sb.sb_rblocks = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001805 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 case XFS_SBS_REXTENTS:
1807 lcounter = (long long)mp->m_sb.sb_rextents;
1808 lcounter += delta;
1809 if (lcounter < 0) {
1810 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001811 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 }
1813 mp->m_sb.sb_rextents = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001814 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 case XFS_SBS_REXTSLOG:
1816 scounter = mp->m_sb.sb_rextslog;
1817 scounter += delta;
1818 if (scounter < 0) {
1819 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001820 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 }
1822 mp->m_sb.sb_rextslog = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001823 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 default:
1825 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001826 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 }
1828}
1829
1830/*
1831 * xfs_mod_incore_sb() is used to change a field in the in-core
1832 * superblock structure by the specified delta. This modification
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001833 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 * routine to do the work.
1835 */
1836int
David Chinner20f4ebf2007-02-10 18:36:10 +11001837xfs_mod_incore_sb(
1838 xfs_mount_t *mp,
1839 xfs_sb_field_t field,
1840 int64_t delta,
1841 int rsvd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 int status;
1844
David Chinner8d280b92006-03-14 13:13:09 +11001845 /* check for per-cpu counters */
1846 switch (field) {
1847#ifdef HAVE_PERCPU_SB
1848 case XFS_SBS_ICOUNT:
1849 case XFS_SBS_IFREE:
1850 case XFS_SBS_FDBLOCKS:
1851 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1852 status = xfs_icsb_modify_counters(mp, field,
1853 delta, rsvd);
1854 break;
1855 }
1856 /* FALLTHROUGH */
1857#endif
1858 default:
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001859 spin_lock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001860 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001861 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001862 break;
1863 }
1864
Jesper Juhl014c2542006-01-15 02:37:08 +01001865 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866}
1867
1868/*
1869 * xfs_mod_incore_sb_batch() is used to change more than one field
1870 * in the in-core superblock structure at a time. This modification
1871 * is protected by a lock internal to this module. The fields and
1872 * changes to those fields are specified in the array of xfs_mod_sb
1873 * structures passed in.
1874 *
1875 * Either all of the specified deltas will be applied or none of
1876 * them will. If any modified field dips below 0, then all modifications
1877 * will be backed out and EINVAL will be returned.
1878 */
1879int
1880xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1881{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 int status=0;
1883 xfs_mod_sb_t *msbp;
1884
1885 /*
1886 * Loop through the array of mod structures and apply each
1887 * individually. If any fail, then back out all those
1888 * which have already been applied. Do all of this within
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001889 * the scope of the m_sb_lock so that all of the changes will
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 * be atomic.
1891 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001892 spin_lock(&mp->m_sb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 msbp = &msb[0];
1894 for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1895 /*
1896 * Apply the delta at index n. If it fails, break
1897 * from the loop so we'll fall into the undo loop
1898 * below.
1899 */
David Chinner8d280b92006-03-14 13:13:09 +11001900 switch (msbp->msb_field) {
1901#ifdef HAVE_PERCPU_SB
1902 case XFS_SBS_ICOUNT:
1903 case XFS_SBS_IFREE:
1904 case XFS_SBS_FDBLOCKS:
1905 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001906 spin_unlock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11001907 status = xfs_icsb_modify_counters(mp,
David Chinner8d280b92006-03-14 13:13:09 +11001908 msbp->msb_field,
1909 msbp->msb_delta, rsvd);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001910 spin_lock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001911 break;
1912 }
1913 /* FALLTHROUGH */
1914#endif
1915 default:
1916 status = xfs_mod_incore_sb_unlocked(mp,
1917 msbp->msb_field,
1918 msbp->msb_delta, rsvd);
1919 break;
1920 }
1921
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 if (status != 0) {
1923 break;
1924 }
1925 }
1926
1927 /*
1928 * If we didn't complete the loop above, then back out
1929 * any changes made to the superblock. If you add code
1930 * between the loop above and here, make sure that you
1931 * preserve the value of status. Loop back until
1932 * we step below the beginning of the array. Make sure
1933 * we don't touch anything back there.
1934 */
1935 if (status != 0) {
1936 msbp--;
1937 while (msbp >= msb) {
David Chinner8d280b92006-03-14 13:13:09 +11001938 switch (msbp->msb_field) {
1939#ifdef HAVE_PERCPU_SB
1940 case XFS_SBS_ICOUNT:
1941 case XFS_SBS_IFREE:
1942 case XFS_SBS_FDBLOCKS:
1943 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001944 spin_unlock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11001945 status = xfs_icsb_modify_counters(mp,
David Chinner8d280b92006-03-14 13:13:09 +11001946 msbp->msb_field,
1947 -(msbp->msb_delta),
1948 rsvd);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001949 spin_lock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001950 break;
1951 }
1952 /* FALLTHROUGH */
1953#endif
1954 default:
1955 status = xfs_mod_incore_sb_unlocked(mp,
1956 msbp->msb_field,
1957 -(msbp->msb_delta),
1958 rsvd);
1959 break;
1960 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 ASSERT(status == 0);
1962 msbp--;
1963 }
1964 }
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001965 spin_unlock(&mp->m_sb_lock);
Jesper Juhl014c2542006-01-15 02:37:08 +01001966 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967}
1968
1969/*
1970 * xfs_getsb() is called to obtain the buffer for the superblock.
1971 * The buffer is returned locked and read in from disk.
1972 * The buffer should be released with a call to xfs_brelse().
1973 *
1974 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1975 * the superblock buffer if it can be locked without sleeping.
1976 * If it can't then we'll return NULL.
1977 */
1978xfs_buf_t *
1979xfs_getsb(
1980 xfs_mount_t *mp,
1981 int flags)
1982{
1983 xfs_buf_t *bp;
1984
1985 ASSERT(mp->m_sb_bp != NULL);
1986 bp = mp->m_sb_bp;
Christoph Hellwig0cadda12010-01-19 09:56:44 +00001987 if (flags & XBF_TRYLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 if (!XFS_BUF_CPSEMA(bp)) {
1989 return NULL;
1990 }
1991 } else {
1992 XFS_BUF_PSEMA(bp, PRIBIO);
1993 }
1994 XFS_BUF_HOLD(bp);
1995 ASSERT(XFS_BUF_ISDONE(bp));
Jesper Juhl014c2542006-01-15 02:37:08 +01001996 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997}
1998
1999/*
2000 * Used to free the superblock along various error paths.
2001 */
2002void
2003xfs_freesb(
2004 xfs_mount_t *mp)
2005{
2006 xfs_buf_t *bp;
2007
2008 /*
2009 * Use xfs_getsb() so that the buffer will be locked
2010 * when we call xfs_buf_relse().
2011 */
2012 bp = xfs_getsb(mp, 0);
2013 XFS_BUF_UNMANAGE(bp);
2014 xfs_buf_relse(bp);
2015 mp->m_sb_bp = NULL;
2016}
2017
2018/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 * Used to log changes to the superblock unit and width fields which could
Eric Sandeene6957ea2008-04-10 12:19:34 +10002020 * be altered by the mount options, as well as any potential sb_features2
2021 * fixup. Only the first superblock is updated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 */
Christoph Hellwig7884bc82009-01-19 02:04:07 +01002023int
David Chinneree1c0902008-03-06 13:45:50 +11002024xfs_mount_log_sb(
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 xfs_mount_t *mp,
2026 __int64_t fields)
2027{
2028 xfs_trans_t *tp;
David Chinnere5720ee2008-04-10 12:21:18 +10002029 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030
David Chinneree1c0902008-03-06 13:45:50 +11002031 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
David Chinner4b166de2008-05-20 11:30:27 +10002032 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
2033 XFS_SB_VERSIONNUM));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034
2035 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
David Chinnere5720ee2008-04-10 12:21:18 +10002036 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
2037 XFS_DEFAULT_LOG_COUNT);
2038 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 xfs_trans_cancel(tp, 0);
David Chinnere5720ee2008-04-10 12:21:18 +10002040 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 }
2042 xfs_mod_sb(tp, fields);
David Chinnere5720ee2008-04-10 12:21:18 +10002043 error = xfs_trans_commit(tp, 0);
2044 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045}
David Chinner8d280b92006-03-14 13:13:09 +11002046
Christoph Hellwigdda35b82010-02-15 09:44:46 +00002047/*
2048 * If the underlying (data/log/rt) device is readonly, there are some
2049 * operations that cannot proceed.
2050 */
2051int
2052xfs_dev_is_read_only(
2053 struct xfs_mount *mp,
2054 char *message)
2055{
2056 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
2057 xfs_readonly_buftarg(mp->m_logdev_targp) ||
2058 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
2059 cmn_err(CE_NOTE,
2060 "XFS: %s required on read-only device.", message);
2061 cmn_err(CE_NOTE,
2062 "XFS: write access unavailable, cannot proceed.");
2063 return EROFS;
2064 }
2065 return 0;
2066}
David Chinner8d280b92006-03-14 13:13:09 +11002067
2068#ifdef HAVE_PERCPU_SB
2069/*
2070 * Per-cpu incore superblock counters
2071 *
2072 * Simple concept, difficult implementation
2073 *
2074 * Basically, replace the incore superblock counters with a distributed per cpu
2075 * counter for contended fields (e.g. free block count).
2076 *
2077 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
2078 * hence needs to be accurately read when we are running low on space. Hence
2079 * there is a method to enable and disable the per-cpu counters based on how
2080 * much "stuff" is available in them.
2081 *
2082 * Basically, a counter is enabled if there is enough free resource to justify
2083 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
2084 * ENOSPC), then we disable the counters to synchronise all callers and
2085 * re-distribute the available resources.
2086 *
2087 * If, once we redistributed the available resources, we still get a failure,
2088 * we disable the per-cpu counter and go through the slow path.
2089 *
2090 * The slow path is the current xfs_mod_incore_sb() function. This means that
Malcolm Parsons9da096f2009-03-29 09:55:42 +02002091 * when we disable a per-cpu counter, we need to drain its resources back to
David Chinner8d280b92006-03-14 13:13:09 +11002092 * the global superblock. We do this after disabling the counter to prevent
2093 * more threads from queueing up on the counter.
2094 *
2095 * Essentially, this means that we still need a lock in the fast path to enable
2096 * synchronisation between the global counters and the per-cpu counters. This
2097 * is not a problem because the lock will be local to a CPU almost all the time
2098 * and have little contention except when we get to ENOSPC conditions.
2099 *
2100 * Basically, this lock becomes a barrier that enables us to lock out the fast
2101 * path while we do things like enabling and disabling counters and
2102 * synchronising the counters.
2103 *
2104 * Locking rules:
2105 *
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002106 * 1. m_sb_lock before picking up per-cpu locks
David Chinner8d280b92006-03-14 13:13:09 +11002107 * 2. per-cpu locks always picked up via for_each_online_cpu() order
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002108 * 3. accurate counter sync requires m_sb_lock + per cpu locks
David Chinner8d280b92006-03-14 13:13:09 +11002109 * 4. modifying per-cpu counters requires holding per-cpu lock
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002110 * 5. modifying global counters requires holding m_sb_lock
2111 * 6. enabling or disabling a counter requires holding the m_sb_lock
David Chinner8d280b92006-03-14 13:13:09 +11002112 * and _none_ of the per-cpu locks.
2113 *
2114 * Disabled counters are only ever re-enabled by a balance operation
2115 * that results in more free resources per CPU than a given threshold.
2116 * To ensure counters don't remain disabled, they are rebalanced when
2117 * the global resource goes above a higher threshold (i.e. some hysteresis
2118 * is present to prevent thrashing).
David Chinner8d280b92006-03-14 13:13:09 +11002119 */
David Chinnere8234a62006-03-14 13:23:52 +11002120
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002121#ifdef CONFIG_HOTPLUG_CPU
David Chinnere8234a62006-03-14 13:23:52 +11002122/*
2123 * hot-plug CPU notifier support.
2124 *
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002125 * We need a notifier per filesystem as we need to be able to identify
2126 * the filesystem to balance the counters out. This is achieved by
2127 * having a notifier block embedded in the xfs_mount_t and doing pointer
2128 * magic to get the mount pointer from the notifier block address.
David Chinnere8234a62006-03-14 13:23:52 +11002129 */
2130STATIC int
2131xfs_icsb_cpu_notify(
2132 struct notifier_block *nfb,
2133 unsigned long action,
2134 void *hcpu)
2135{
2136 xfs_icsb_cnts_t *cntp;
2137 xfs_mount_t *mp;
David Chinnere8234a62006-03-14 13:23:52 +11002138
2139 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
2140 cntp = (xfs_icsb_cnts_t *)
2141 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
2142 switch (action) {
2143 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002144 case CPU_UP_PREPARE_FROZEN:
David Chinnere8234a62006-03-14 13:23:52 +11002145 /* Easy Case - initialize the area and locks, and
2146 * then rebalance when online does everything else for us. */
David Chinner01e1b692006-03-14 13:29:16 +11002147 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinnere8234a62006-03-14 13:23:52 +11002148 break;
2149 case CPU_ONLINE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002150 case CPU_ONLINE_FROZEN:
David Chinner03135cf2007-02-10 18:35:15 +11002151 xfs_icsb_lock(mp);
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002152 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2153 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2154 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
David Chinner03135cf2007-02-10 18:35:15 +11002155 xfs_icsb_unlock(mp);
David Chinnere8234a62006-03-14 13:23:52 +11002156 break;
2157 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002158 case CPU_DEAD_FROZEN:
David Chinnere8234a62006-03-14 13:23:52 +11002159 /* Disable all the counters, then fold the dead cpu's
2160 * count into the total on the global superblock and
2161 * re-enable the counters. */
David Chinner03135cf2007-02-10 18:35:15 +11002162 xfs_icsb_lock(mp);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002163 spin_lock(&mp->m_sb_lock);
David Chinnere8234a62006-03-14 13:23:52 +11002164 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
2165 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
2166 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
2167
2168 mp->m_sb.sb_icount += cntp->icsb_icount;
2169 mp->m_sb.sb_ifree += cntp->icsb_ifree;
2170 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
2171
David Chinner01e1b692006-03-14 13:29:16 +11002172 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinnere8234a62006-03-14 13:23:52 +11002173
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002174 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
2175 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
2176 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002177 spin_unlock(&mp->m_sb_lock);
David Chinner03135cf2007-02-10 18:35:15 +11002178 xfs_icsb_unlock(mp);
David Chinnere8234a62006-03-14 13:23:52 +11002179 break;
2180 }
2181
2182 return NOTIFY_OK;
2183}
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002184#endif /* CONFIG_HOTPLUG_CPU */
David Chinnere8234a62006-03-14 13:23:52 +11002185
David Chinner8d280b92006-03-14 13:13:09 +11002186int
2187xfs_icsb_init_counters(
2188 xfs_mount_t *mp)
2189{
2190 xfs_icsb_cnts_t *cntp;
2191 int i;
2192
2193 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2194 if (mp->m_sb_cnts == NULL)
2195 return -ENOMEM;
2196
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002197#ifdef CONFIG_HOTPLUG_CPU
David Chinnere8234a62006-03-14 13:23:52 +11002198 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2199 mp->m_icsb_notifier.priority = 0;
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002200 register_hotcpu_notifier(&mp->m_icsb_notifier);
2201#endif /* CONFIG_HOTPLUG_CPU */
David Chinnere8234a62006-03-14 13:23:52 +11002202
David Chinner8d280b92006-03-14 13:13:09 +11002203 for_each_online_cpu(i) {
2204 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11002205 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinner8d280b92006-03-14 13:13:09 +11002206 }
David Chinner20b64282007-02-10 18:35:09 +11002207
2208 mutex_init(&mp->m_icsb_mutex);
2209
David Chinner8d280b92006-03-14 13:13:09 +11002210 /*
2211 * start with all counters disabled so that the
2212 * initial balance kicks us off correctly
2213 */
2214 mp->m_icsb_counters = -1;
2215 return 0;
2216}
2217
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11002218void
2219xfs_icsb_reinit_counters(
2220 xfs_mount_t *mp)
2221{
2222 xfs_icsb_lock(mp);
2223 /*
2224 * start with all counters disabled so that the
2225 * initial balance kicks us off correctly
2226 */
2227 mp->m_icsb_counters = -1;
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002228 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2229 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2230 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11002231 xfs_icsb_unlock(mp);
2232}
2233
Christoph Hellwigc962fb72008-05-20 15:10:52 +10002234void
David Chinner8d280b92006-03-14 13:13:09 +11002235xfs_icsb_destroy_counters(
2236 xfs_mount_t *mp)
2237{
David Chinnere8234a62006-03-14 13:23:52 +11002238 if (mp->m_sb_cnts) {
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002239 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
David Chinner8d280b92006-03-14 13:13:09 +11002240 free_percpu(mp->m_sb_cnts);
David Chinnere8234a62006-03-14 13:23:52 +11002241 }
David Chinner03135cf2007-02-10 18:35:15 +11002242 mutex_destroy(&mp->m_icsb_mutex);
David Chinner8d280b92006-03-14 13:13:09 +11002243}
2244
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00002245STATIC void
David Chinner01e1b692006-03-14 13:29:16 +11002246xfs_icsb_lock_cntr(
2247 xfs_icsb_cnts_t *icsbp)
2248{
2249 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2250 ndelay(1000);
2251 }
2252}
2253
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00002254STATIC void
David Chinner01e1b692006-03-14 13:29:16 +11002255xfs_icsb_unlock_cntr(
2256 xfs_icsb_cnts_t *icsbp)
2257{
2258 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2259}
2260
David Chinner8d280b92006-03-14 13:13:09 +11002261
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00002262STATIC void
David Chinner8d280b92006-03-14 13:13:09 +11002263xfs_icsb_lock_all_counters(
2264 xfs_mount_t *mp)
2265{
2266 xfs_icsb_cnts_t *cntp;
2267 int i;
2268
2269 for_each_online_cpu(i) {
2270 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11002271 xfs_icsb_lock_cntr(cntp);
David Chinner8d280b92006-03-14 13:13:09 +11002272 }
2273}
2274
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00002275STATIC void
David Chinner8d280b92006-03-14 13:13:09 +11002276xfs_icsb_unlock_all_counters(
2277 xfs_mount_t *mp)
2278{
2279 xfs_icsb_cnts_t *cntp;
2280 int i;
2281
2282 for_each_online_cpu(i) {
2283 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11002284 xfs_icsb_unlock_cntr(cntp);
David Chinner8d280b92006-03-14 13:13:09 +11002285 }
2286}
2287
2288STATIC void
2289xfs_icsb_count(
2290 xfs_mount_t *mp,
2291 xfs_icsb_cnts_t *cnt,
2292 int flags)
2293{
2294 xfs_icsb_cnts_t *cntp;
2295 int i;
2296
2297 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2298
2299 if (!(flags & XFS_ICSB_LAZY_COUNT))
2300 xfs_icsb_lock_all_counters(mp);
2301
2302 for_each_online_cpu(i) {
2303 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2304 cnt->icsb_icount += cntp->icsb_icount;
2305 cnt->icsb_ifree += cntp->icsb_ifree;
2306 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2307 }
2308
2309 if (!(flags & XFS_ICSB_LAZY_COUNT))
2310 xfs_icsb_unlock_all_counters(mp);
2311}
2312
2313STATIC int
2314xfs_icsb_counter_disabled(
2315 xfs_mount_t *mp,
2316 xfs_sb_field_t field)
2317{
2318 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2319 return test_bit(field, &mp->m_icsb_counters);
2320}
2321
David Chinner36fbe6e2008-04-10 12:19:56 +10002322STATIC void
David Chinner8d280b92006-03-14 13:13:09 +11002323xfs_icsb_disable_counter(
2324 xfs_mount_t *mp,
2325 xfs_sb_field_t field)
2326{
2327 xfs_icsb_cnts_t cnt;
2328
2329 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2330
David Chinner20b64282007-02-10 18:35:09 +11002331 /*
2332 * If we are already disabled, then there is nothing to do
2333 * here. We check before locking all the counters to avoid
2334 * the expensive lock operation when being called in the
2335 * slow path and the counter is already disabled. This is
2336 * safe because the only time we set or clear this state is under
2337 * the m_icsb_mutex.
2338 */
2339 if (xfs_icsb_counter_disabled(mp, field))
David Chinner36fbe6e2008-04-10 12:19:56 +10002340 return;
David Chinner20b64282007-02-10 18:35:09 +11002341
David Chinner8d280b92006-03-14 13:13:09 +11002342 xfs_icsb_lock_all_counters(mp);
2343 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2344 /* drain back to superblock */
2345
Christoph Hellwigce461932008-04-22 17:34:50 +10002346 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
David Chinner8d280b92006-03-14 13:13:09 +11002347 switch(field) {
2348 case XFS_SBS_ICOUNT:
2349 mp->m_sb.sb_icount = cnt.icsb_icount;
2350 break;
2351 case XFS_SBS_IFREE:
2352 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2353 break;
2354 case XFS_SBS_FDBLOCKS:
2355 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2356 break;
2357 default:
2358 BUG();
2359 }
2360 }
2361
2362 xfs_icsb_unlock_all_counters(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002363}
2364
2365STATIC void
2366xfs_icsb_enable_counter(
2367 xfs_mount_t *mp,
2368 xfs_sb_field_t field,
2369 uint64_t count,
2370 uint64_t resid)
2371{
2372 xfs_icsb_cnts_t *cntp;
2373 int i;
2374
2375 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2376
2377 xfs_icsb_lock_all_counters(mp);
2378 for_each_online_cpu(i) {
2379 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2380 switch (field) {
2381 case XFS_SBS_ICOUNT:
2382 cntp->icsb_icount = count + resid;
2383 break;
2384 case XFS_SBS_IFREE:
2385 cntp->icsb_ifree = count + resid;
2386 break;
2387 case XFS_SBS_FDBLOCKS:
2388 cntp->icsb_fdblocks = count + resid;
2389 break;
2390 default:
2391 BUG();
2392 break;
2393 }
2394 resid = 0;
2395 }
2396 clear_bit(field, &mp->m_icsb_counters);
2397 xfs_icsb_unlock_all_counters(mp);
2398}
2399
David Chinnerdbcabad2007-02-10 18:36:17 +11002400void
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002401xfs_icsb_sync_counters_locked(
David Chinner8d280b92006-03-14 13:13:09 +11002402 xfs_mount_t *mp,
2403 int flags)
2404{
2405 xfs_icsb_cnts_t cnt;
David Chinner8d280b92006-03-14 13:13:09 +11002406
David Chinner8d280b92006-03-14 13:13:09 +11002407 xfs_icsb_count(mp, &cnt, flags);
2408
David Chinner8d280b92006-03-14 13:13:09 +11002409 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2410 mp->m_sb.sb_icount = cnt.icsb_icount;
2411 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2412 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2413 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2414 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
David Chinner8d280b92006-03-14 13:13:09 +11002415}
2416
2417/*
2418 * Accurate update of per-cpu counters to incore superblock
2419 */
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002420void
David Chinner8d280b92006-03-14 13:13:09 +11002421xfs_icsb_sync_counters(
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002422 xfs_mount_t *mp,
2423 int flags)
David Chinner8d280b92006-03-14 13:13:09 +11002424{
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002425 spin_lock(&mp->m_sb_lock);
2426 xfs_icsb_sync_counters_locked(mp, flags);
2427 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11002428}
2429
2430/*
2431 * Balance and enable/disable counters as necessary.
2432 *
David Chinner20b64282007-02-10 18:35:09 +11002433 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2434 * chosen to be the same number as single on disk allocation chunk per CPU, and
2435 * free blocks is something far enough zero that we aren't going thrash when we
2436 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2437 * prevent looping endlessly when xfs_alloc_space asks for more than will
2438 * be distributed to a single CPU but each CPU has enough blocks to be
2439 * reenabled.
2440 *
2441 * Note that we can be called when counters are already disabled.
2442 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2443 * prevent locking every per-cpu counter needlessly.
David Chinner8d280b92006-03-14 13:13:09 +11002444 */
David Chinner20b64282007-02-10 18:35:09 +11002445
2446#define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
David Chinner4be536d2006-09-07 14:26:50 +10002447#define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
David Chinner20b64282007-02-10 18:35:09 +11002448 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
David Chinner8d280b92006-03-14 13:13:09 +11002449STATIC void
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002450xfs_icsb_balance_counter_locked(
David Chinner8d280b92006-03-14 13:13:09 +11002451 xfs_mount_t *mp,
2452 xfs_sb_field_t field,
David Chinner20b64282007-02-10 18:35:09 +11002453 int min_per_cpu)
David Chinner8d280b92006-03-14 13:13:09 +11002454{
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002455 uint64_t count, resid;
David Chinner8d280b92006-03-14 13:13:09 +11002456 int weight = num_online_cpus();
David Chinner20b64282007-02-10 18:35:09 +11002457 uint64_t min = (uint64_t)min_per_cpu;
David Chinner8d280b92006-03-14 13:13:09 +11002458
David Chinner8d280b92006-03-14 13:13:09 +11002459 /* disable counter and sync counter */
2460 xfs_icsb_disable_counter(mp, field);
2461
2462 /* update counters - first CPU gets residual*/
2463 switch (field) {
2464 case XFS_SBS_ICOUNT:
2465 count = mp->m_sb.sb_icount;
2466 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002467 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002468 return;
David Chinner8d280b92006-03-14 13:13:09 +11002469 break;
2470 case XFS_SBS_IFREE:
2471 count = mp->m_sb.sb_ifree;
2472 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002473 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002474 return;
David Chinner8d280b92006-03-14 13:13:09 +11002475 break;
2476 case XFS_SBS_FDBLOCKS:
2477 count = mp->m_sb.sb_fdblocks;
2478 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002479 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002480 return;
David Chinner8d280b92006-03-14 13:13:09 +11002481 break;
2482 default:
2483 BUG();
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002484 count = resid = 0; /* quiet, gcc */
David Chinner8d280b92006-03-14 13:13:09 +11002485 break;
2486 }
2487
2488 xfs_icsb_enable_counter(mp, field, count, resid);
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002489}
2490
2491STATIC void
2492xfs_icsb_balance_counter(
2493 xfs_mount_t *mp,
2494 xfs_sb_field_t fields,
2495 int min_per_cpu)
2496{
2497 spin_lock(&mp->m_sb_lock);
2498 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2499 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11002500}
2501
David Chinnera8272ce2007-11-23 16:28:09 +11002502STATIC int
David Chinner20b64282007-02-10 18:35:09 +11002503xfs_icsb_modify_counters(
David Chinner8d280b92006-03-14 13:13:09 +11002504 xfs_mount_t *mp,
2505 xfs_sb_field_t field,
David Chinner20f4ebf2007-02-10 18:36:10 +11002506 int64_t delta,
David Chinner20b64282007-02-10 18:35:09 +11002507 int rsvd)
David Chinner8d280b92006-03-14 13:13:09 +11002508{
2509 xfs_icsb_cnts_t *icsbp;
2510 long long lcounter; /* long counter for 64 bit fields */
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002511 int ret = 0;
David Chinner8d280b92006-03-14 13:13:09 +11002512
David Chinner20b64282007-02-10 18:35:09 +11002513 might_sleep();
David Chinner8d280b92006-03-14 13:13:09 +11002514again:
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002515 preempt_disable();
2516 icsbp = this_cpu_ptr(mp->m_sb_cnts);
David Chinner20b64282007-02-10 18:35:09 +11002517
2518 /*
2519 * if the counter is disabled, go to slow path
2520 */
David Chinner8d280b92006-03-14 13:13:09 +11002521 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2522 goto slow_path;
David Chinner20b64282007-02-10 18:35:09 +11002523 xfs_icsb_lock_cntr(icsbp);
2524 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2525 xfs_icsb_unlock_cntr(icsbp);
2526 goto slow_path;
2527 }
David Chinner8d280b92006-03-14 13:13:09 +11002528
2529 switch (field) {
2530 case XFS_SBS_ICOUNT:
2531 lcounter = icsbp->icsb_icount;
2532 lcounter += delta;
2533 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002534 goto balance_counter;
David Chinner8d280b92006-03-14 13:13:09 +11002535 icsbp->icsb_icount = lcounter;
2536 break;
2537
2538 case XFS_SBS_IFREE:
2539 lcounter = icsbp->icsb_ifree;
2540 lcounter += delta;
2541 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002542 goto balance_counter;
David Chinner8d280b92006-03-14 13:13:09 +11002543 icsbp->icsb_ifree = lcounter;
2544 break;
2545
2546 case XFS_SBS_FDBLOCKS:
2547 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2548
David Chinner4be536d2006-09-07 14:26:50 +10002549 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002550 lcounter += delta;
2551 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002552 goto balance_counter;
David Chinner4be536d2006-09-07 14:26:50 +10002553 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002554 break;
2555 default:
2556 BUG();
2557 break;
2558 }
David Chinner01e1b692006-03-14 13:29:16 +11002559 xfs_icsb_unlock_cntr(icsbp);
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002560 preempt_enable();
David Chinner8d280b92006-03-14 13:13:09 +11002561 return 0;
2562
David Chinner8d280b92006-03-14 13:13:09 +11002563slow_path:
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002564 preempt_enable();
David Chinner20b64282007-02-10 18:35:09 +11002565
2566 /*
2567 * serialise with a mutex so we don't burn lots of cpu on
2568 * the superblock lock. We still need to hold the superblock
2569 * lock, however, when we modify the global structures.
2570 */
David Chinner03135cf2007-02-10 18:35:15 +11002571 xfs_icsb_lock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002572
2573 /*
2574 * Now running atomically.
2575 *
2576 * If the counter is enabled, someone has beaten us to rebalancing.
2577 * Drop the lock and try again in the fast path....
2578 */
2579 if (!(xfs_icsb_counter_disabled(mp, field))) {
David Chinner03135cf2007-02-10 18:35:15 +11002580 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002581 goto again;
2582 }
2583
2584 /*
2585 * The counter is currently disabled. Because we are
2586 * running atomically here, we know a rebalance cannot
2587 * be in progress. Hence we can go straight to operating
2588 * on the global superblock. We do not call xfs_mod_incore_sb()
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002589 * here even though we need to get the m_sb_lock. Doing so
David Chinner20b64282007-02-10 18:35:09 +11002590 * will cause us to re-enter this function and deadlock.
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002591 * Hence we get the m_sb_lock ourselves and then call
David Chinner20b64282007-02-10 18:35:09 +11002592 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2593 * directly on the global counters.
2594 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002595 spin_lock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11002596 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002597 spin_unlock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11002598
2599 /*
2600 * Now that we've modified the global superblock, we
2601 * may be able to re-enable the distributed counters
2602 * (e.g. lots of space just got freed). After that
2603 * we are done.
2604 */
2605 if (ret != ENOSPC)
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002606 xfs_icsb_balance_counter(mp, field, 0);
David Chinner03135cf2007-02-10 18:35:15 +11002607 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002608 return ret;
2609
2610balance_counter:
David Chinner01e1b692006-03-14 13:29:16 +11002611 xfs_icsb_unlock_cntr(icsbp);
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09002612 preempt_enable();
David Chinner8d280b92006-03-14 13:13:09 +11002613
David Chinner20b64282007-02-10 18:35:09 +11002614 /*
2615 * We may have multiple threads here if multiple per-cpu
2616 * counters run dry at the same time. This will mean we can
2617 * do more balances than strictly necessary but it is not
2618 * the common slowpath case.
2619 */
David Chinner03135cf2007-02-10 18:35:15 +11002620 xfs_icsb_lock(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002621
David Chinner20b64282007-02-10 18:35:09 +11002622 /*
2623 * running atomically.
2624 *
2625 * This will leave the counter in the correct state for future
2626 * accesses. After the rebalance, we simply try again and our retry
2627 * will either succeed through the fast path or slow path without
2628 * another balance operation being required.
2629 */
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002630 xfs_icsb_balance_counter(mp, field, delta);
David Chinner03135cf2007-02-10 18:35:15 +11002631 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002632 goto again;
David Chinner8d280b92006-03-14 13:13:09 +11002633}
2634
David Chinner8d280b92006-03-14 13:13:09 +11002635#endif