blob: 6e69e8a006710b8ff65846e09532ff71d810fbf8 [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"
Dave Chinner6ca1c902013-08-12 20:49:26 +100020#include "xfs_format.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
Christoph Hellwig211e4d42012-04-23 15:58:34 +100025#include "xfs_trans_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_sb.h"
27#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_mount.h"
Dave Chinner2b9ab5a2013-08-12 20:49:37 +100029#include "xfs_da_btree.h"
30#include "xfs_dir2_format.h"
31#include "xfs_dir2.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110033#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_ialloc_btree.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_quota.h"
44#include "xfs_fsops.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000045#include "xfs_trace.h"
Dave Chinner6d8b79c2012-10-08 21:56:09 +110046#include "xfs_icache.h"
Dave Chinner04a1e6c2013-04-03 16:11:31 +110047#include "xfs_cksum.h"
48#include "xfs_buf_item.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000049
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
David Chinner8d280b92006-03-14 13:13:09 +110051#ifdef HAVE_PERCPU_SB
David Chinner20f4ebf2007-02-10 18:36:10 +110052STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
Christoph Hellwig45af6c62008-04-22 17:34:44 +100053 int);
54STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t,
55 int);
David Chinner36fbe6e2008-04-10 12:19:56 +100056STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
David Chinner8d280b92006-03-14 13:13:09 +110057#else
58
Christoph Hellwig45af6c62008-04-22 17:34:44 +100059#define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
60#define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
David Chinner8d280b92006-03-14 13:13:09 +110061#endif
62
Christoph Hellwig27174202009-03-30 10:21:31 +020063static DEFINE_MUTEX(xfs_uuid_table_mutex);
64static int xfs_uuid_table_size;
65static uuid_t *xfs_uuid_table;
66
67/*
68 * See if the UUID is unique among mounted XFS filesystems.
69 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
70 */
71STATIC int
72xfs_uuid_mount(
73 struct xfs_mount *mp)
74{
75 uuid_t *uuid = &mp->m_sb.sb_uuid;
76 int hole, i;
77
78 if (mp->m_flags & XFS_MOUNT_NOUUID)
79 return 0;
80
81 if (uuid_is_nil(uuid)) {
Dave Chinner0b932cc2011-03-07 10:08:35 +110082 xfs_warn(mp, "Filesystem has nil UUID - can't mount");
Christoph Hellwig27174202009-03-30 10:21:31 +020083 return XFS_ERROR(EINVAL);
84 }
85
86 mutex_lock(&xfs_uuid_table_mutex);
87 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
88 if (uuid_is_nil(&xfs_uuid_table[i])) {
89 hole = i;
90 continue;
91 }
92 if (uuid_equal(uuid, &xfs_uuid_table[i]))
93 goto out_duplicate;
94 }
95
96 if (hole < 0) {
97 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
98 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
99 xfs_uuid_table_size * sizeof(*xfs_uuid_table),
100 KM_SLEEP);
101 hole = xfs_uuid_table_size++;
102 }
103 xfs_uuid_table[hole] = *uuid;
104 mutex_unlock(&xfs_uuid_table_mutex);
105
106 return 0;
107
108 out_duplicate:
109 mutex_unlock(&xfs_uuid_table_mutex);
Mitsuo Hayasaka021000e2012-01-13 05:58:39 +0000110 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
Christoph Hellwig27174202009-03-30 10:21:31 +0200111 return XFS_ERROR(EINVAL);
112}
113
114STATIC void
115xfs_uuid_unmount(
116 struct xfs_mount *mp)
117{
118 uuid_t *uuid = &mp->m_sb.sb_uuid;
119 int i;
120
121 if (mp->m_flags & XFS_MOUNT_NOUUID)
122 return;
123
124 mutex_lock(&xfs_uuid_table_mutex);
125 for (i = 0; i < xfs_uuid_table_size; i++) {
126 if (uuid_is_nil(&xfs_uuid_table[i]))
127 continue;
128 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
129 continue;
130 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
131 break;
132 }
133 ASSERT(i < xfs_uuid_table_size);
134 mutex_unlock(&xfs_uuid_table_mutex);
135}
136
137
Dave Chinnere1765792010-09-22 10:47:20 +1000138STATIC void
139__xfs_free_perag(
140 struct rcu_head *head)
141{
142 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
143
144 ASSERT(atomic_read(&pag->pag_ref) == 0);
145 kmem_free(pag);
146}
147
Dave Chinner0fa800f2010-01-11 11:47:46 +0000148/*
Dave Chinnere1765792010-09-22 10:47:20 +1000149 * Free up the per-ag resources associated with the mount structure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 */
Christoph Hellwigc962fb72008-05-20 15:10:52 +1000151STATIC void
Christoph Hellwigff4f0382008-08-13 16:50:47 +1000152xfs_free_perag(
Christoph Hellwig745f6912007-08-30 17:20:39 +1000153 xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154{
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000155 xfs_agnumber_t agno;
156 struct xfs_perag *pag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000158 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
159 spin_lock(&mp->m_perag_lock);
160 pag = radix_tree_delete(&mp->m_perag_tree, agno);
161 spin_unlock(&mp->m_perag_lock);
Dave Chinnere1765792010-09-22 10:47:20 +1000162 ASSERT(pag);
Dave Chinnerf83282a2010-11-08 08:55:04 +0000163 ASSERT(atomic_read(&pag->pag_ref) == 0);
Dave Chinnere1765792010-09-22 10:47:20 +1000164 call_rcu(&pag->rcu_head, __xfs_free_perag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166}
167
Nathan Scott4cc929e2007-05-14 18:24:02 +1000168/*
169 * Check size of device based on the (data/realtime) block count.
170 * Note: this check is used by the growfs code as well as mount.
171 */
172int
173xfs_sb_validate_fsb_count(
174 xfs_sb_t *sbp,
175 __uint64_t nblocks)
176{
177 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
178 ASSERT(sbp->sb_blocklog >= BBSHIFT);
179
180#if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
181 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
Eric Sandeen657a4cf2010-04-30 03:42:49 +0000182 return EFBIG;
Nathan Scott4cc929e2007-05-14 18:24:02 +1000183#else /* Limited by UINT_MAX of sectors */
184 if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
Eric Sandeen657a4cf2010-04-30 03:42:49 +0000185 return EFBIG;
Nathan Scott4cc929e2007-05-14 18:24:02 +1000186#endif
187 return 0;
188}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000190int
Nathan Scottc11e2c32005-11-02 15:11:45 +1100191xfs_initialize_perag(
Nathan Scottc11e2c32005-11-02 15:11:45 +1100192 xfs_mount_t *mp,
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000193 xfs_agnumber_t agcount,
194 xfs_agnumber_t *maxagi)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195{
Carlos Maiolino2d2194f2012-09-20 10:32:38 -0300196 xfs_agnumber_t index;
Dave Chinner8b26c582010-01-11 11:47:48 +0000197 xfs_agnumber_t first_initialised = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 xfs_perag_t *pag;
199 xfs_agino_t agino;
200 xfs_ino_t ino;
201 xfs_sb_t *sbp = &mp->m_sb;
Dave Chinner8b26c582010-01-11 11:47:48 +0000202 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000204 /*
205 * Walk the current per-ag tree so we don't try to initialise AGs
206 * that already exist (growfs case). Allocate and insert all the
207 * AGs we don't find ready for initialisation.
208 */
209 for (index = 0; index < agcount; index++) {
210 pag = xfs_perag_get(mp, index);
211 if (pag) {
212 xfs_perag_put(pag);
213 continue;
214 }
Dave Chinner8b26c582010-01-11 11:47:48 +0000215 if (!first_initialised)
216 first_initialised = index;
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000217
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000218 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
219 if (!pag)
Dave Chinner8b26c582010-01-11 11:47:48 +0000220 goto out_unwind;
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000221 pag->pag_agno = index;
222 pag->pag_mount = mp;
Dave Chinner1a427ab2010-12-16 17:08:41 +1100223 spin_lock_init(&pag->pag_ici_lock);
Dave Chinner69b491c2010-09-27 11:09:51 +1000224 mutex_init(&pag->pag_ici_reclaim_lock);
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000225 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
Dave Chinner74f75a02010-09-24 19:59:04 +1000226 spin_lock_init(&pag->pag_buf_lock);
227 pag->pag_buf_tree = RB_ROOT;
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000228
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000229 if (radix_tree_preload(GFP_NOFS))
Dave Chinner8b26c582010-01-11 11:47:48 +0000230 goto out_unwind;
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000231
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000232 spin_lock(&mp->m_perag_lock);
233 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
234 BUG();
235 spin_unlock(&mp->m_perag_lock);
Dave Chinner8b26c582010-01-11 11:47:48 +0000236 radix_tree_preload_end();
237 error = -EEXIST;
238 goto out_unwind;
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000239 }
240 spin_unlock(&mp->m_perag_lock);
241 radix_tree_preload_end();
242 }
243
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000244 /*
245 * If we mount with the inode64 option, or no inode overflows
246 * the legacy 32-bit address space clear the inode32 option.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 */
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000248 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
249 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000251 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
252 mp->m_flags |= XFS_MOUNT_32BITINODES;
253 else
254 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
255
Carlos Maiolino2d2194f2012-09-20 10:32:38 -0300256 if (mp->m_flags & XFS_MOUNT_32BITINODES)
257 index = xfs_set_inode32(mp);
258 else
259 index = xfs_set_inode64(mp);
Christoph Hellwigfb3b5042010-05-28 19:03:10 +0000260
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000261 if (maxagi)
262 *maxagi = index;
263 return 0;
Dave Chinner8b26c582010-01-11 11:47:48 +0000264
265out_unwind:
266 kmem_free(pag);
267 for (; index > first_initialised; index--) {
268 pag = radix_tree_delete(&mp->m_perag_tree, index);
269 kmem_free(pag);
270 }
271 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272}
273
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274/*
275 * xfs_readsb
276 *
277 * Does the initial read of the superblock.
278 */
279int
Dave Chinnerff550682013-08-12 20:49:41 +1000280xfs_readsb(
281 struct xfs_mount *mp,
282 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283{
284 unsigned int sector_size;
Dave Chinner04a1e6c2013-04-03 16:11:31 +1100285 struct xfs_buf *bp;
286 struct xfs_sb *sbp = &mp->m_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 int error;
Dave Chinneraf34e092011-03-07 10:04:35 +1100288 int loud = !(flags & XFS_MFSI_QUIET);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289
290 ASSERT(mp->m_sb_bp == NULL);
291 ASSERT(mp->m_ddev_targp != NULL);
292
293 /*
294 * Allocate a (locked) buffer to hold the superblock.
295 * This will be kept around at all times to optimize
296 * access to the superblock.
297 */
298 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
Dave Chinner26af6552010-09-22 10:47:20 +1000300reread:
Dave Chinnere70b73f2012-04-23 15:58:49 +1000301 bp = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
Dave Chinner98021822012-11-12 22:54:03 +1100302 BTOBB(sector_size), 0,
Dave Chinner1813dd62012-11-14 17:54:40 +1100303 loud ? &xfs_sb_buf_ops
304 : &xfs_sb_quiet_buf_ops);
Dave Chinner26af6552010-09-22 10:47:20 +1000305 if (!bp) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100306 if (loud)
307 xfs_warn(mp, "SB buffer read failed");
Dave Chinner26af6552010-09-22 10:47:20 +1000308 return EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 }
Dave Chinnereab4e632012-11-12 22:54:02 +1100310 if (bp->b_error) {
311 error = bp->b_error;
312 if (loud)
Dave Chinnere721f502013-04-03 16:11:32 +1100313 xfs_warn(mp, "SB validate failed with error %d.", error);
Dave Chinnereab4e632012-11-12 22:54:02 +1100314 goto release_buf;
315 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
317 /*
318 * Initialize the mount structure from the superblock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 */
Dave Chinner98021822012-11-12 22:54:03 +1100320 xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
Chandra Seetharaman83e782e2013-06-27 17:25:10 -0500321 xfs_sb_quota_from_disk(&mp->m_sb);
Dave Chinnerff550682013-08-12 20:49:41 +1000322
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 /*
324 * We must be able to do sector-sized and sector-aligned IO.
325 */
Dave Chinner04a1e6c2013-04-03 16:11:31 +1100326 if (sector_size > sbp->sb_sectsize) {
Dave Chinneraf34e092011-03-07 10:04:35 +1100327 if (loud)
328 xfs_warn(mp, "device supports %u byte sectors (not %u)",
Dave Chinner04a1e6c2013-04-03 16:11:31 +1100329 sector_size, sbp->sb_sectsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 error = ENOSYS;
Dave Chinner26af6552010-09-22 10:47:20 +1000331 goto release_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 }
333
334 /*
335 * If device sector size is smaller than the superblock size,
336 * re-read the superblock so the buffer is correctly sized.
337 */
Dave Chinner04a1e6c2013-04-03 16:11:31 +1100338 if (sector_size < sbp->sb_sectsize) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 xfs_buf_relse(bp);
Dave Chinner04a1e6c2013-04-03 16:11:31 +1100340 sector_size = sbp->sb_sectsize;
Dave Chinner26af6552010-09-22 10:47:20 +1000341 goto reread;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 }
343
Lachlan McIlroy5478eea2007-02-10 18:36:29 +1100344 /* Initialize per-cpu counters */
345 xfs_icsb_reinit_counters(mp);
David Chinner8d280b92006-03-14 13:13:09 +1100346
Dave Chinner04a1e6c2013-04-03 16:11:31 +1100347 /* no need to be quiet anymore, so reset the buf ops */
348 bp->b_ops = &xfs_sb_buf_ops;
349
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 mp->m_sb_bp = bp;
Dave Chinner26af6552010-09-22 10:47:20 +1000351 xfs_buf_unlock(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 return 0;
353
Dave Chinner26af6552010-09-22 10:47:20 +1000354release_buf:
355 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 return error;
357}
358
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359/*
Eric Sandeen0771fb42007-10-12 11:03:40 +1000360 * Update alignment values based on mount options and sb values
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 */
Eric Sandeen0771fb42007-10-12 11:03:40 +1000362STATIC int
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100363xfs_update_alignment(xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 xfs_sb_t *sbp = &(mp->m_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366
Christoph Hellwig42490232008-08-13 16:49:32 +1000367 if (mp->m_dalign) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 /*
369 * If stripe unit and stripe width are not multiples
370 * of the fs blocksize turn off alignment.
371 */
372 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
373 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
Jie Liu39a45d82013-05-02 19:27:47 +0800374 xfs_warn(mp,
375 "alignment check failed: sunit/swidth vs. blocksize(%d)",
376 sbp->sb_blocksize);
377 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 } else {
379 /*
380 * Convert the stripe unit and width to FSBs.
381 */
382 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
383 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
Dave Chinner53487782011-03-07 10:05:35 +1100384 xfs_warn(mp,
Jie Liu39a45d82013-05-02 19:27:47 +0800385 "alignment check failed: sunit/swidth vs. agsize(%d)",
386 sbp->sb_agblocks);
387 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 } else if (mp->m_dalign) {
389 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
390 } else {
Jie Liu39a45d82013-05-02 19:27:47 +0800391 xfs_warn(mp,
392 "alignment check failed: sunit(%d) less than bsize(%d)",
393 mp->m_dalign, sbp->sb_blocksize);
394 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 }
396 }
397
398 /*
399 * Update superblock with new values
400 * and log changes
401 */
Eric Sandeen62118702008-03-06 13:44:28 +1100402 if (xfs_sb_version_hasdalign(sbp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 if (sbp->sb_unit != mp->m_dalign) {
404 sbp->sb_unit = mp->m_dalign;
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100405 mp->m_update_flags |= XFS_SB_UNIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 }
407 if (sbp->sb_width != mp->m_swidth) {
408 sbp->sb_width = mp->m_swidth;
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100409 mp->m_update_flags |= XFS_SB_WIDTH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 }
Jie Liu34d7f602013-05-02 19:27:53 +0800411 } else {
412 xfs_warn(mp,
413 "cannot change alignment: superblock does not support data alignment");
414 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 }
416 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
Eric Sandeen62118702008-03-06 13:44:28 +1100417 xfs_sb_version_hasdalign(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 mp->m_dalign = sbp->sb_unit;
419 mp->m_swidth = sbp->sb_width;
420 }
421
Eric Sandeen0771fb42007-10-12 11:03:40 +1000422 return 0;
423}
424
425/*
426 * Set the maximum inode count for this filesystem
427 */
428STATIC void
429xfs_set_maxicount(xfs_mount_t *mp)
430{
431 xfs_sb_t *sbp = &(mp->m_sb);
432 __uint64_t icount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433
434 if (sbp->sb_imax_pct) {
Eric Sandeen0771fb42007-10-12 11:03:40 +1000435 /*
436 * Make sure the maximum inode count is a multiple
437 * of the units we allocate inodes in.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
440 do_div(icount, 100);
441 do_div(icount, mp->m_ialloc_blks);
442 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
443 sbp->sb_inopblog;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000444 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 mp->m_maxicount = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 }
Eric Sandeen0771fb42007-10-12 11:03:40 +1000447}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
Eric Sandeen0771fb42007-10-12 11:03:40 +1000449/*
450 * Set the default minimum read and write sizes unless
451 * already specified in a mount option.
452 * We use smaller I/O sizes when the file system
453 * is being used for NFS service (wsync mount option).
454 */
455STATIC void
456xfs_set_rw_sizes(xfs_mount_t *mp)
457{
458 xfs_sb_t *sbp = &(mp->m_sb);
459 int readio_log, writeio_log;
460
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
462 if (mp->m_flags & XFS_MOUNT_WSYNC) {
463 readio_log = XFS_WSYNC_READIO_LOG;
464 writeio_log = XFS_WSYNC_WRITEIO_LOG;
465 } else {
466 readio_log = XFS_READIO_LOG_LARGE;
467 writeio_log = XFS_WRITEIO_LOG_LARGE;
468 }
469 } else {
470 readio_log = mp->m_readio_log;
471 writeio_log = mp->m_writeio_log;
472 }
473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 if (sbp->sb_blocklog > readio_log) {
475 mp->m_readio_log = sbp->sb_blocklog;
476 } else {
477 mp->m_readio_log = readio_log;
478 }
479 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
480 if (sbp->sb_blocklog > writeio_log) {
481 mp->m_writeio_log = sbp->sb_blocklog;
482 } else {
483 mp->m_writeio_log = writeio_log;
484 }
485 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000486}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
Eric Sandeen0771fb42007-10-12 11:03:40 +1000488/*
Dave Chinner055388a2011-01-04 11:35:03 +1100489 * precalculate the low space thresholds for dynamic speculative preallocation.
490 */
491void
492xfs_set_low_space_thresholds(
493 struct xfs_mount *mp)
494{
495 int i;
496
497 for (i = 0; i < XFS_LOWSP_MAX; i++) {
498 __uint64_t space = mp->m_sb.sb_dblocks;
499
500 do_div(space, 100);
501 mp->m_low_space[i] = space * (i + 1);
502 }
503}
504
505
506/*
Eric Sandeen0771fb42007-10-12 11:03:40 +1000507 * Set whether we're using inode alignment.
508 */
509STATIC void
510xfs_set_inoalignment(xfs_mount_t *mp)
511{
Eric Sandeen62118702008-03-06 13:44:28 +1100512 if (xfs_sb_version_hasalign(&mp->m_sb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 mp->m_sb.sb_inoalignmt >=
514 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
515 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
516 else
517 mp->m_inoalign_mask = 0;
518 /*
519 * If we are using stripe alignment, check whether
520 * the stripe unit is a multiple of the inode alignment
521 */
522 if (mp->m_dalign && mp->m_inoalign_mask &&
523 !(mp->m_dalign & mp->m_inoalign_mask))
524 mp->m_sinoalign = mp->m_dalign;
525 else
526 mp->m_sinoalign = 0;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000527}
528
529/*
Zhi Yong Wu0471f622013-08-07 10:10:58 +0000530 * Check that the data (and log if separate) is an ok size.
Eric Sandeen0771fb42007-10-12 11:03:40 +1000531 */
532STATIC int
Christoph Hellwig42490232008-08-13 16:49:32 +1000533xfs_check_sizes(xfs_mount_t *mp)
Eric Sandeen0771fb42007-10-12 11:03:40 +1000534{
535 xfs_buf_t *bp;
536 xfs_daddr_t d;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000537
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
539 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100540 xfs_warn(mp, "filesystem size mismatch detected");
Eric Sandeen657a4cf2010-04-30 03:42:49 +0000541 return XFS_ERROR(EFBIG);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 }
Dave Chinnere70b73f2012-04-23 15:58:49 +1000543 bp = xfs_buf_read_uncached(mp->m_ddev_targp,
Dave Chinner1922c942010-09-22 10:47:20 +1000544 d - XFS_FSS_TO_BB(mp, 1),
Dave Chinnerc3f8fc72012-11-12 22:54:01 +1100545 XFS_FSS_TO_BB(mp, 1), 0, NULL);
Dave Chinner1922c942010-09-22 10:47:20 +1000546 if (!bp) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100547 xfs_warn(mp, "last sector read failed");
Dave Chinner1922c942010-09-22 10:47:20 +1000548 return EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 }
Dave Chinner1922c942010-09-22 10:47:20 +1000550 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
Christoph Hellwig42490232008-08-13 16:49:32 +1000552 if (mp->m_logdev_targp != mp->m_ddev_targp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
554 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100555 xfs_warn(mp, "log size mismatch detected");
Eric Sandeen657a4cf2010-04-30 03:42:49 +0000556 return XFS_ERROR(EFBIG);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 }
Dave Chinnere70b73f2012-04-23 15:58:49 +1000558 bp = xfs_buf_read_uncached(mp->m_logdev_targp,
Dave Chinner1922c942010-09-22 10:47:20 +1000559 d - XFS_FSB_TO_BB(mp, 1),
Dave Chinnerc3f8fc72012-11-12 22:54:01 +1100560 XFS_FSB_TO_BB(mp, 1), 0, NULL);
Dave Chinner1922c942010-09-22 10:47:20 +1000561 if (!bp) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100562 xfs_warn(mp, "log device read failed");
Dave Chinner1922c942010-09-22 10:47:20 +1000563 return EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 }
Dave Chinner1922c942010-09-22 10:47:20 +1000565 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 }
Eric Sandeen0771fb42007-10-12 11:03:40 +1000567 return 0;
568}
569
570/*
Christoph Hellwig7d095252009-06-08 15:33:32 +0200571 * Clear the quotaflags in memory and in the superblock.
572 */
573int
574xfs_mount_reset_sbqflags(
575 struct xfs_mount *mp)
576{
577 int error;
578 struct xfs_trans *tp;
579
580 mp->m_qflags = 0;
581
582 /*
583 * It is OK to look at sb_qflags here in mount path,
584 * without m_sb_lock.
585 */
586 if (mp->m_sb.sb_qflags == 0)
587 return 0;
588 spin_lock(&mp->m_sb_lock);
589 mp->m_sb.sb_qflags = 0;
590 spin_unlock(&mp->m_sb_lock);
591
592 /*
593 * If the fs is readonly, let the incore superblock run
594 * with quotas off but don't flush the update out to disk
595 */
596 if (mp->m_flags & XFS_MOUNT_RDONLY)
597 return 0;
598
Christoph Hellwig7d095252009-06-08 15:33:32 +0200599 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
Jie Liu3d3c8b52013-08-12 20:49:59 +1000600 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0);
Christoph Hellwig7d095252009-06-08 15:33:32 +0200601 if (error) {
602 xfs_trans_cancel(tp, 0);
Dave Chinner53487782011-03-07 10:05:35 +1100603 xfs_alert(mp, "%s: Superblock update failed!", __func__);
Christoph Hellwig7d095252009-06-08 15:33:32 +0200604 return error;
605 }
606
607 xfs_mod_sb(tp, XFS_SB_QFLAGS);
608 return xfs_trans_commit(tp, 0);
609}
610
Eric Sandeend5db0f92010-02-05 22:59:53 +0000611__uint64_t
612xfs_default_resblks(xfs_mount_t *mp)
613{
614 __uint64_t resblks;
615
616 /*
Dave Chinner8babd8a2010-03-04 01:46:25 +0000617 * We default to 5% or 8192 fsbs of space reserved, whichever is
618 * smaller. This is intended to cover concurrent allocation
619 * transactions when we initially hit enospc. These each require a 4
620 * block reservation. Hence by default we cover roughly 2000 concurrent
621 * allocation reservations.
Eric Sandeend5db0f92010-02-05 22:59:53 +0000622 */
623 resblks = mp->m_sb.sb_dblocks;
624 do_div(resblks, 20);
Dave Chinner8babd8a2010-03-04 01:46:25 +0000625 resblks = min_t(__uint64_t, resblks, 8192);
Eric Sandeend5db0f92010-02-05 22:59:53 +0000626 return resblks;
627}
628
Christoph Hellwig7d095252009-06-08 15:33:32 +0200629/*
Eric Sandeen0771fb42007-10-12 11:03:40 +1000630 * This function does the following on an initial mount of a file system:
631 * - reads the superblock from disk and init the mount struct
632 * - if we're a 32-bit kernel, do a size check on the superblock
633 * so we don't mount terabyte filesystems
634 * - init mount struct realtime fields
635 * - allocate inode hash table for fs
636 * - init directory manager
637 * - perform recovery and init the log manager
638 */
639int
640xfs_mountfs(
Christoph Hellwig42490232008-08-13 16:49:32 +1000641 xfs_mount_t *mp)
Eric Sandeen0771fb42007-10-12 11:03:40 +1000642{
643 xfs_sb_t *sbp = &(mp->m_sb);
644 xfs_inode_t *rip;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000645 __uint64_t resblks;
Christoph Hellwig7d095252009-06-08 15:33:32 +0200646 uint quotamount = 0;
647 uint quotaflags = 0;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000648 int error = 0;
649
Dave Chinnerff550682013-08-12 20:49:41 +1000650 xfs_sb_mount_common(mp, sbp);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000651
652 /*
Eric Sandeene6957ea2008-04-10 12:19:34 +1000653 * Check for a mismatched features2 values. Older kernels
654 * read & wrote into the wrong sb offset for sb_features2
655 * on some platforms due to xfs_sb_t not being 64bit size aligned
656 * when sb_features2 was added, which made older superblock
657 * reading/writing routines swap it as a 64-bit value.
David Chinneree1c0902008-03-06 13:45:50 +1100658 *
Eric Sandeene6957ea2008-04-10 12:19:34 +1000659 * For backwards compatibility, we make both slots equal.
660 *
661 * If we detect a mismatched field, we OR the set bits into the
662 * existing features2 field in case it has already been modified; we
663 * don't want to lose any features. We then update the bad location
664 * with the ORed value so that older kernels will see any features2
665 * flags, and mark the two fields as needing updates once the
666 * transaction subsystem is online.
David Chinneree1c0902008-03-06 13:45:50 +1100667 */
Eric Sandeene6957ea2008-04-10 12:19:34 +1000668 if (xfs_sb_has_mismatched_features2(sbp)) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100669 xfs_warn(mp, "correcting sb_features alignment problem");
David Chinneree1c0902008-03-06 13:45:50 +1100670 sbp->sb_features2 |= sbp->sb_bad_features2;
Eric Sandeene6957ea2008-04-10 12:19:34 +1000671 sbp->sb_bad_features2 = sbp->sb_features2;
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100672 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
Eric Sandeene6957ea2008-04-10 12:19:34 +1000673
674 /*
675 * Re-check for ATTR2 in case it was found in bad_features2
676 * slot.
677 */
Tim Shimmin7c12f292008-04-30 18:15:28 +1000678 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
679 !(mp->m_flags & XFS_MOUNT_NOATTR2))
Eric Sandeene6957ea2008-04-10 12:19:34 +1000680 mp->m_flags |= XFS_MOUNT_ATTR2;
Tim Shimmin7c12f292008-04-30 18:15:28 +1000681 }
Eric Sandeene6957ea2008-04-10 12:19:34 +1000682
Tim Shimmin7c12f292008-04-30 18:15:28 +1000683 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
684 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
685 xfs_sb_version_removeattr2(&mp->m_sb);
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100686 mp->m_update_flags |= XFS_SB_FEATURES2;
Tim Shimmin7c12f292008-04-30 18:15:28 +1000687
688 /* update sb_versionnum for the clearing of the morebits */
689 if (!sbp->sb_features2)
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100690 mp->m_update_flags |= XFS_SB_VERSIONNUM;
David Chinneree1c0902008-03-06 13:45:50 +1100691 }
692
693 /*
Eric Sandeen0771fb42007-10-12 11:03:40 +1000694 * Check if sb_agblocks is aligned at stripe boundary
695 * If sb_agblocks is NOT aligned turn off m_dalign since
696 * allocator alignment is within an ag, therefore ag has
697 * to be aligned at stripe boundary.
698 */
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100699 error = xfs_update_alignment(mp);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000700 if (error)
Christoph Hellwigf9057e32009-02-04 09:31:52 +0100701 goto out;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000702
703 xfs_alloc_compute_maxlevels(mp);
704 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
705 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
706 xfs_ialloc_compute_maxlevels(mp);
707
708 xfs_set_maxicount(mp);
709
Christoph Hellwig27174202009-03-30 10:21:31 +0200710 error = xfs_uuid_mount(mp);
711 if (error)
712 goto out;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000713
714 /*
715 * Set the minimum read and write sizes
716 */
717 xfs_set_rw_sizes(mp);
718
Dave Chinner055388a2011-01-04 11:35:03 +1100719 /* set the low space thresholds for dynamic preallocation */
720 xfs_set_low_space_thresholds(mp);
721
Eric Sandeen0771fb42007-10-12 11:03:40 +1000722 /*
723 * Set the inode cluster size.
724 * This may still be overridden by the file system
725 * block size if it is larger than the chosen cluster size.
726 */
727 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
728
729 /*
730 * Set inode alignment fields
731 */
732 xfs_set_inoalignment(mp);
733
734 /*
735 * Check that the data (and log if separate) are an ok size.
736 */
Christoph Hellwig42490232008-08-13 16:49:32 +1000737 error = xfs_check_sizes(mp);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000738 if (error)
Christoph Hellwigf9057e32009-02-04 09:31:52 +0100739 goto out_remove_uuid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740
741 /*
742 * Initialize realtime fields in the mount structure
743 */
Eric Sandeen0771fb42007-10-12 11:03:40 +1000744 error = xfs_rtmount_init(mp);
745 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100746 xfs_warn(mp, "RT mount failed");
Christoph Hellwigf9057e32009-02-04 09:31:52 +0100747 goto out_remove_uuid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 }
749
750 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 * Copies the low order bits of the timestamp and the randomly
752 * set "sequence" number out of a UUID.
753 */
754 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
755
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 mp->m_dmevmask = 0; /* not persistent; set after each mount */
757
Nathan Scottf6c2d1f2006-06-20 13:04:51 +1000758 xfs_dir_mount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759
760 /*
761 * Initialize the attribute manager's entries.
762 */
763 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
764
765 /*
766 * Initialize the precomputed transaction reservations values.
767 */
768 xfs_trans_init(mp);
769
770 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 * Allocate and initialize the per-ag data.
772 */
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000773 spin_lock_init(&mp->m_perag_lock);
Dave Chinner9b98b6f2010-05-27 01:58:13 +0000774 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000775 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
776 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100777 xfs_warn(mp, "Failed per-ag init: %d", error);
Christoph Hellwigf9057e32009-02-04 09:31:52 +0100778 goto out_remove_uuid;
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000779 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780
Christoph Hellwigf9057e32009-02-04 09:31:52 +0100781 if (!sbp->sb_logblocks) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100782 xfs_warn(mp, "no log defined");
Christoph Hellwigf9057e32009-02-04 09:31:52 +0100783 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
784 error = XFS_ERROR(EFSCORRUPTED);
785 goto out_free_perag;
786 }
787
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 /*
789 * log's mount-time initialization. Perform 1st part recovery if needed
790 */
Christoph Hellwigf9057e32009-02-04 09:31:52 +0100791 error = xfs_log_mount(mp, mp->m_logdev_targp,
792 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
793 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
794 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100795 xfs_warn(mp, "log mount failed");
Dave Chinnerd4f35122012-04-23 15:59:06 +1000796 goto out_fail_wait;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 }
798
799 /*
David Chinner92821e22007-05-24 15:26:31 +1000800 * Now the log is mounted, we know if it was an unclean shutdown or
801 * not. If it was, with the first phase of recovery has completed, we
802 * have consistent AG blocks on disk. We have not recovered EFIs yet,
803 * but they are recovered transactionally in the second recovery phase
804 * later.
805 *
806 * Hence we can safely re-initialise incore superblock counters from
807 * the per-ag data. These may not be correct if the filesystem was not
808 * cleanly unmounted, so we need to wait for recovery to finish before
809 * doing this.
810 *
811 * If the filesystem was cleanly unmounted, then we can trust the
812 * values in the superblock to be correct and we don't need to do
813 * anything here.
814 *
815 * If we are currently making the filesystem, the initialisation will
816 * fail as the perag data is in an undefined state.
817 */
David Chinner92821e22007-05-24 15:26:31 +1000818 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
819 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
820 !mp->m_sb.sb_inprogress) {
821 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
Christoph Hellwigf9057e32009-02-04 09:31:52 +0100822 if (error)
Dave Chinnerd4f35122012-04-23 15:59:06 +1000823 goto out_fail_wait;
David Chinner92821e22007-05-24 15:26:31 +1000824 }
Christoph Hellwigf9057e32009-02-04 09:31:52 +0100825
David Chinner92821e22007-05-24 15:26:31 +1000826 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 * Get and sanity-check the root inode.
828 * Save the pointer to it in the mount structure.
829 */
Dave Chinner7b6259e2010-06-24 11:35:17 +1000830 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100832 xfs_warn(mp, "failed to read root inode");
Christoph Hellwigf9057e32009-02-04 09:31:52 +0100833 goto out_log_dealloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 }
835
836 ASSERT(rip != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837
Al Viroabbede12011-07-26 02:31:30 -0400838 if (unlikely(!S_ISDIR(rip->i_d.di_mode))) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100839 xfs_warn(mp, "corrupted root inode %llu: not a directory",
Nathan Scottb6574522006-06-09 15:29:40 +1000840 (unsigned long long)rip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 xfs_iunlock(rip, XFS_ILOCK_EXCL);
842 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
843 mp);
844 error = XFS_ERROR(EFSCORRUPTED);
Christoph Hellwigf9057e32009-02-04 09:31:52 +0100845 goto out_rele_rip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 }
847 mp->m_rootip = rip; /* save it */
848
849 xfs_iunlock(rip, XFS_ILOCK_EXCL);
850
851 /*
852 * Initialize realtime inode pointers in the mount structure
853 */
Eric Sandeen0771fb42007-10-12 11:03:40 +1000854 error = xfs_rtmount_inodes(mp);
855 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 /*
857 * Free up the root inode.
858 */
Dave Chinner0b932cc2011-03-07 10:08:35 +1100859 xfs_warn(mp, "failed to read RT inodes");
Christoph Hellwigf9057e32009-02-04 09:31:52 +0100860 goto out_rele_rip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 }
862
863 /*
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100864 * If this is a read-only mount defer the superblock updates until
865 * the next remount into writeable mode. Otherwise we would never
866 * perform the update e.g. for the root filesystem.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 */
Christoph Hellwig7884bc82009-01-19 02:04:07 +0100868 if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
869 error = xfs_mount_log_sb(mp, mp->m_update_flags);
David Chinnere5720ee2008-04-10 12:21:18 +1000870 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100871 xfs_warn(mp, "failed to write sb changes");
Christoph Hellwigb93b6e42009-02-04 09:33:58 +0100872 goto out_rtunmount;
David Chinnere5720ee2008-04-10 12:21:18 +1000873 }
874 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875
876 /*
877 * Initialise the XFS quota management subsystem for this mount
878 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200879 if (XFS_IS_QUOTA_RUNNING(mp)) {
880 error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
881 if (error)
882 goto out_rtunmount;
883 } else {
884 ASSERT(!XFS_IS_QUOTA_ON(mp));
885
886 /*
887 * If a file system had quotas running earlier, but decided to
888 * mount without -o uquota/pquota/gquota options, revoke the
889 * quotachecked license.
890 */
891 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100892 xfs_notice(mp, "resetting quota flags");
Christoph Hellwig7d095252009-06-08 15:33:32 +0200893 error = xfs_mount_reset_sbqflags(mp);
894 if (error)
895 return error;
896 }
897 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
899 /*
900 * Finish recovering the file system. This part needed to be
901 * delayed until after the root and real-time bitmap inodes
902 * were consistently read in.
903 */
Christoph Hellwig42490232008-08-13 16:49:32 +1000904 error = xfs_log_mount_finish(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100906 xfs_warn(mp, "log mount finish failed");
Christoph Hellwigb93b6e42009-02-04 09:33:58 +0100907 goto out_rtunmount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 }
909
910 /*
911 * Complete the quota initialisation, post-log-replay component.
912 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200913 if (quotamount) {
914 ASSERT(mp->m_qflags == 0);
915 mp->m_qflags = quotaflags;
916
917 xfs_qm_mount_quotas(mp);
918 }
919
David Chinner84e1e992007-06-18 16:50:27 +1000920 /*
921 * Now we are mounted, reserve a small amount of unused space for
922 * privileged transactions. This is needed so that transaction
923 * space required for critical operations can dip into this pool
924 * when at ENOSPC. This is needed for operations like create with
925 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
926 * are not allowed to use this reserved space.
Dave Chinner8babd8a2010-03-04 01:46:25 +0000927 *
928 * This may drive us straight to ENOSPC on mount, but that implies
929 * we were already there on the last unmount. Warn if this occurs.
David Chinner84e1e992007-06-18 16:50:27 +1000930 */
Eric Sandeend5db0f92010-02-05 22:59:53 +0000931 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
932 resblks = xfs_default_resblks(mp);
933 error = xfs_reserve_blocks(mp, &resblks, NULL);
934 if (error)
Dave Chinner0b932cc2011-03-07 10:08:35 +1100935 xfs_warn(mp,
936 "Unable to allocate reserve blocks. Continuing without reserve pool.");
Eric Sandeend5db0f92010-02-05 22:59:53 +0000937 }
David Chinner84e1e992007-06-18 16:50:27 +1000938
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 return 0;
940
Christoph Hellwigb93b6e42009-02-04 09:33:58 +0100941 out_rtunmount:
942 xfs_rtunmount_inodes(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +0100943 out_rele_rip:
Christoph Hellwig43355092008-03-27 18:01:08 +1100944 IRELE(rip);
Christoph Hellwigf9057e32009-02-04 09:31:52 +0100945 out_log_dealloc:
Christoph Hellwig21b699c2009-03-16 08:19:29 +0100946 xfs_log_unmount(mp);
Dave Chinnerd4f35122012-04-23 15:59:06 +1000947 out_fail_wait:
948 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
949 xfs_wait_buftarg(mp->m_logdev_targp);
950 xfs_wait_buftarg(mp->m_ddev_targp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +0100951 out_free_perag:
Christoph Hellwigff4f0382008-08-13 16:50:47 +1000952 xfs_free_perag(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +0100953 out_remove_uuid:
Christoph Hellwig27174202009-03-30 10:21:31 +0200954 xfs_uuid_unmount(mp);
Christoph Hellwigf9057e32009-02-04 09:31:52 +0100955 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 return error;
957}
958
959/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 * This flushes out the inodes,dquots and the superblock, unmounts the
961 * log and makes sure that incore structures are freed.
962 */
Christoph Hellwig41b5c2e2008-08-13 16:49:57 +1000963void
964xfs_unmountfs(
965 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966{
Christoph Hellwig41b5c2e2008-08-13 16:49:57 +1000967 __uint64_t resblks;
968 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969
Brian Foster579b62f2012-11-06 09:50:47 -0500970 cancel_delayed_work_sync(&mp->m_eofblocks_work);
971
Christoph Hellwig7d095252009-06-08 15:33:32 +0200972 xfs_qm_unmount_quotas(mp);
Christoph Hellwigb93b6e42009-02-04 09:33:58 +0100973 xfs_rtunmount_inodes(mp);
Christoph Hellwig77508ec2008-08-13 16:49:04 +1000974 IRELE(mp->m_rootip);
975
David Chinner641c56f2007-06-18 16:50:17 +1000976 /*
977 * We can potentially deadlock here if we have an inode cluster
Malcolm Parsons9da096f2009-03-29 09:55:42 +0200978 * that has been freed has its buffer still pinned in memory because
David Chinner641c56f2007-06-18 16:50:17 +1000979 * the transaction is still sitting in a iclog. The stale inodes
980 * on that buffer will have their flush locks held until the
981 * transaction hits the disk and the callbacks run. the inode
982 * flush takes the flush lock unconditionally and with nothing to
983 * push out the iclog we will never get that unlocked. hence we
984 * need to force the log first.
985 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +0000986 xfs_log_force(mp, XFS_LOG_SYNC);
Dave Chinnerc8543632010-02-06 12:39:36 +1100987
988 /*
Christoph Hellwig211e4d42012-04-23 15:58:34 +1000989 * Flush all pending changes from the AIL.
Dave Chinnerc8543632010-02-06 12:39:36 +1100990 */
Christoph Hellwig211e4d42012-04-23 15:58:34 +1000991 xfs_ail_push_all_sync(mp->m_ail);
992
993 /*
994 * And reclaim all inodes. At this point there should be no dirty
Dave Chinner7e185302012-10-08 21:56:00 +1100995 * inodes and none should be pinned or locked, but use synchronous
996 * reclaim just to be sure. We can stop background inode reclaim
997 * here as well if it is still running.
Christoph Hellwig211e4d42012-04-23 15:58:34 +1000998 */
Dave Chinner7e185302012-10-08 21:56:00 +1100999 cancel_delayed_work_sync(&mp->m_reclaim_work);
Dave Chinnerc8543632010-02-06 12:39:36 +11001000 xfs_reclaim_inodes(mp, SYNC_WAIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001
Christoph Hellwig7d095252009-06-08 15:33:32 +02001002 xfs_qm_unmount(mp);
Lachlan McIlroya357a122008-10-30 16:53:25 +11001003
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 /*
David Chinner84e1e992007-06-18 16:50:27 +10001005 * Unreserve any blocks we have so that when we unmount we don't account
1006 * the reserved free space as used. This is really only necessary for
1007 * lazy superblock counting because it trusts the incore superblock
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001008 * counters to be absolutely correct on clean unmount.
David Chinner84e1e992007-06-18 16:50:27 +10001009 *
1010 * We don't bother correcting this elsewhere for lazy superblock
1011 * counting because on mount of an unclean filesystem we reconstruct the
1012 * correct counter value and this is irrelevant.
1013 *
1014 * For non-lazy counter filesystems, this doesn't matter at all because
1015 * we only every apply deltas to the superblock and hence the incore
1016 * value does not matter....
1017 */
1018 resblks = 0;
David Chinner714082b2008-04-10 12:20:03 +10001019 error = xfs_reserve_blocks(mp, &resblks, NULL);
1020 if (error)
Dave Chinner0b932cc2011-03-07 10:08:35 +11001021 xfs_warn(mp, "Unable to free reserved block pool. "
David Chinner714082b2008-04-10 12:20:03 +10001022 "Freespace may not be correct on next mount.");
1023
Chandra Seetharamanadab0f62011-06-29 22:10:14 +00001024 error = xfs_log_sbcount(mp);
David Chinnere5720ee2008-04-10 12:21:18 +10001025 if (error)
Dave Chinner0b932cc2011-03-07 10:08:35 +11001026 xfs_warn(mp, "Unable to update superblock counters. "
David Chinnere5720ee2008-04-10 12:21:18 +10001027 "Freespace may not be correct on next mount.");
Christoph Hellwig87c7bec2011-09-14 14:08:26 +00001028
Christoph Hellwig21b699c2009-03-16 08:19:29 +01001029 xfs_log_unmount(mp);
Christoph Hellwig27174202009-03-30 10:21:31 +02001030 xfs_uuid_unmount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031
Christoph Hellwig1550d0b2008-08-13 16:17:37 +10001032#if defined(DEBUG)
Christoph Hellwig0ce4cfd2007-08-30 17:20:53 +10001033 xfs_errortag_clearall(mp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034#endif
Christoph Hellwigff4f0382008-08-13 16:50:47 +10001035 xfs_free_perag(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036}
1037
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038int
David Chinner92821e22007-05-24 15:26:31 +10001039xfs_fs_writable(xfs_mount_t *mp)
1040{
Jan Karad9457dc2012-06-12 16:20:39 +02001041 return !(mp->m_super->s_writers.frozen || XFS_FORCED_SHUTDOWN(mp) ||
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001042 (mp->m_flags & XFS_MOUNT_RDONLY));
David Chinner92821e22007-05-24 15:26:31 +10001043}
1044
1045/*
Alex Elderb2ce3972011-07-11 09:51:44 -05001046 * xfs_log_sbcount
1047 *
Chandra Seetharamanadab0f62011-06-29 22:10:14 +00001048 * Sync the superblock counters to disk.
Alex Elderb2ce3972011-07-11 09:51:44 -05001049 *
1050 * Note this code can be called during the process of freezing, so
Chandra Seetharamanadab0f62011-06-29 22:10:14 +00001051 * we may need to use the transaction allocator which does not
Alex Elderb2ce3972011-07-11 09:51:44 -05001052 * block when the transaction subsystem is in its frozen state.
David Chinner92821e22007-05-24 15:26:31 +10001053 */
1054int
Chandra Seetharamanadab0f62011-06-29 22:10:14 +00001055xfs_log_sbcount(xfs_mount_t *mp)
David Chinner92821e22007-05-24 15:26:31 +10001056{
1057 xfs_trans_t *tp;
1058 int error;
1059
1060 if (!xfs_fs_writable(mp))
1061 return 0;
1062
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10001063 xfs_icsb_sync_counters(mp, 0);
David Chinner92821e22007-05-24 15:26:31 +10001064
1065 /*
1066 * we don't need to do this if we are updating the superblock
1067 * counters on every modification.
1068 */
1069 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1070 return 0;
1071
Alex Elderb2ce3972011-07-11 09:51:44 -05001072 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
Jie Liu3d3c8b52013-08-12 20:49:59 +10001073 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
David Chinner92821e22007-05-24 15:26:31 +10001074 if (error) {
1075 xfs_trans_cancel(tp, 0);
1076 return error;
1077 }
1078
1079 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
Chandra Seetharamanadab0f62011-06-29 22:10:14 +00001080 xfs_trans_set_sync(tp);
David Chinnere5720ee2008-04-10 12:21:18 +10001081 error = xfs_trans_commit(tp, 0);
1082 return error;
David Chinner92821e22007-05-24 15:26:31 +10001083}
1084
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1087 * a delta to a specified field in the in-core superblock. Simply
1088 * switch on the field indicated and apply the delta to that field.
1089 * Fields are not allowed to dip below zero, so if the delta would
1090 * do this do not apply it and return EINVAL.
1091 *
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001092 * The m_sb_lock must be held when this routine is called.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05001094STATIC int
David Chinner20f4ebf2007-02-10 18:36:10 +11001095xfs_mod_incore_sb_unlocked(
1096 xfs_mount_t *mp,
1097 xfs_sb_field_t field,
1098 int64_t delta,
1099 int rsvd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100{
1101 int scounter; /* short counter for 32 bit fields */
1102 long long lcounter; /* long counter for 64 bit fields */
1103 long long res_used, rem;
1104
1105 /*
1106 * With the in-core superblock spin lock held, switch
1107 * on the indicated field. Apply the delta to the
1108 * proper field. If the fields value would dip below
1109 * 0, then do not apply the delta and return EINVAL.
1110 */
1111 switch (field) {
1112 case XFS_SBS_ICOUNT:
1113 lcounter = (long long)mp->m_sb.sb_icount;
1114 lcounter += delta;
1115 if (lcounter < 0) {
1116 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001117 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 }
1119 mp->m_sb.sb_icount = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001120 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 case XFS_SBS_IFREE:
1122 lcounter = (long long)mp->m_sb.sb_ifree;
1123 lcounter += delta;
1124 if (lcounter < 0) {
1125 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001126 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 }
1128 mp->m_sb.sb_ifree = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001129 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 case XFS_SBS_FDBLOCKS:
David Chinner4be536d2006-09-07 14:26:50 +10001131 lcounter = (long long)
1132 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1134
1135 if (delta > 0) { /* Putting blocks back */
1136 if (res_used > delta) {
1137 mp->m_resblks_avail += delta;
1138 } else {
1139 rem = delta - res_used;
1140 mp->m_resblks_avail = mp->m_resblks;
1141 lcounter += rem;
1142 }
1143 } else { /* Taking blocks away */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 lcounter += delta;
Dave Chinner8babd8a2010-03-04 01:46:25 +00001145 if (lcounter >= 0) {
1146 mp->m_sb.sb_fdblocks = lcounter +
1147 XFS_ALLOC_SET_ASIDE(mp);
1148 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 }
Dave Chinner8babd8a2010-03-04 01:46:25 +00001150
1151 /*
1152 * We are out of blocks, use any available reserved
1153 * blocks if were allowed to.
1154 */
1155 if (!rsvd)
1156 return XFS_ERROR(ENOSPC);
1157
1158 lcounter = (long long)mp->m_resblks_avail + delta;
1159 if (lcounter >= 0) {
1160 mp->m_resblks_avail = lcounter;
1161 return 0;
1162 }
1163 printk_once(KERN_WARNING
1164 "Filesystem \"%s\": reserve blocks depleted! "
1165 "Consider increasing reserve pool size.",
1166 mp->m_fsname);
1167 return XFS_ERROR(ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 }
1169
David Chinner4be536d2006-09-07 14:26:50 +10001170 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
Jesper Juhl014c2542006-01-15 02:37:08 +01001171 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 case XFS_SBS_FREXTENTS:
1173 lcounter = (long long)mp->m_sb.sb_frextents;
1174 lcounter += delta;
1175 if (lcounter < 0) {
Jesper Juhl014c2542006-01-15 02:37:08 +01001176 return XFS_ERROR(ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 }
1178 mp->m_sb.sb_frextents = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001179 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 case XFS_SBS_DBLOCKS:
1181 lcounter = (long long)mp->m_sb.sb_dblocks;
1182 lcounter += delta;
1183 if (lcounter < 0) {
1184 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001185 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 }
1187 mp->m_sb.sb_dblocks = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001188 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 case XFS_SBS_AGCOUNT:
1190 scounter = mp->m_sb.sb_agcount;
1191 scounter += delta;
1192 if (scounter < 0) {
1193 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001194 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 }
1196 mp->m_sb.sb_agcount = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001197 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 case XFS_SBS_IMAX_PCT:
1199 scounter = mp->m_sb.sb_imax_pct;
1200 scounter += delta;
1201 if (scounter < 0) {
1202 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001203 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 }
1205 mp->m_sb.sb_imax_pct = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001206 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 case XFS_SBS_REXTSIZE:
1208 scounter = mp->m_sb.sb_rextsize;
1209 scounter += delta;
1210 if (scounter < 0) {
1211 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001212 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 }
1214 mp->m_sb.sb_rextsize = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001215 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 case XFS_SBS_RBMBLOCKS:
1217 scounter = mp->m_sb.sb_rbmblocks;
1218 scounter += delta;
1219 if (scounter < 0) {
1220 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001221 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 }
1223 mp->m_sb.sb_rbmblocks = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001224 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 case XFS_SBS_RBLOCKS:
1226 lcounter = (long long)mp->m_sb.sb_rblocks;
1227 lcounter += delta;
1228 if (lcounter < 0) {
1229 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001230 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 }
1232 mp->m_sb.sb_rblocks = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001233 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 case XFS_SBS_REXTENTS:
1235 lcounter = (long long)mp->m_sb.sb_rextents;
1236 lcounter += delta;
1237 if (lcounter < 0) {
1238 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001239 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 }
1241 mp->m_sb.sb_rextents = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001242 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 case XFS_SBS_REXTSLOG:
1244 scounter = mp->m_sb.sb_rextslog;
1245 scounter += delta;
1246 if (scounter < 0) {
1247 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001248 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 }
1250 mp->m_sb.sb_rextslog = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001251 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 default:
1253 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001254 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 }
1256}
1257
1258/*
1259 * xfs_mod_incore_sb() is used to change a field in the in-core
1260 * superblock structure by the specified delta. This modification
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001261 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 * routine to do the work.
1263 */
1264int
David Chinner20f4ebf2007-02-10 18:36:10 +11001265xfs_mod_incore_sb(
Christoph Hellwig96540c72010-09-30 02:25:55 +00001266 struct xfs_mount *mp,
1267 xfs_sb_field_t field,
1268 int64_t delta,
1269 int rsvd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270{
Christoph Hellwig96540c72010-09-30 02:25:55 +00001271 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272
David Chinner8d280b92006-03-14 13:13:09 +11001273#ifdef HAVE_PERCPU_SB
Christoph Hellwig96540c72010-09-30 02:25:55 +00001274 ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS);
David Chinner8d280b92006-03-14 13:13:09 +11001275#endif
Christoph Hellwig96540c72010-09-30 02:25:55 +00001276 spin_lock(&mp->m_sb_lock);
1277 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1278 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001279
Jesper Juhl014c2542006-01-15 02:37:08 +01001280 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281}
1282
1283/*
Christoph Hellwig1b040712010-09-30 02:25:56 +00001284 * Change more than one field in the in-core superblock structure at a time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 *
Christoph Hellwig1b040712010-09-30 02:25:56 +00001286 * The fields and changes to those fields are specified in the array of
1287 * xfs_mod_sb structures passed in. Either all of the specified deltas
1288 * will be applied or none of them will. If any modified field dips below 0,
1289 * then all modifications will be backed out and EINVAL will be returned.
1290 *
1291 * Note that this function may not be used for the superblock values that
1292 * are tracked with the in-memory per-cpu counters - a direct call to
1293 * xfs_icsb_modify_counters is required for these.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 */
1295int
Christoph Hellwig1b040712010-09-30 02:25:56 +00001296xfs_mod_incore_sb_batch(
1297 struct xfs_mount *mp,
1298 xfs_mod_sb_t *msb,
1299 uint nmsb,
1300 int rsvd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301{
David Sterba45c51b92011-04-13 22:03:28 +00001302 xfs_mod_sb_t *msbp;
Christoph Hellwig1b040712010-09-30 02:25:56 +00001303 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
1305 /*
Christoph Hellwig1b040712010-09-30 02:25:56 +00001306 * Loop through the array of mod structures and apply each individually.
1307 * If any fail, then back out all those which have already been applied.
1308 * Do all of this within the scope of the m_sb_lock so that all of the
1309 * changes will be atomic.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001311 spin_lock(&mp->m_sb_lock);
David Sterba45c51b92011-04-13 22:03:28 +00001312 for (msbp = msb; msbp < (msb + nmsb); msbp++) {
Christoph Hellwig1b040712010-09-30 02:25:56 +00001313 ASSERT(msbp->msb_field < XFS_SBS_ICOUNT ||
1314 msbp->msb_field > XFS_SBS_FDBLOCKS);
David Chinner8d280b92006-03-14 13:13:09 +11001315
Christoph Hellwig1b040712010-09-30 02:25:56 +00001316 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1317 msbp->msb_delta, rsvd);
1318 if (error)
1319 goto unwind;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 }
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001321 spin_unlock(&mp->m_sb_lock);
Christoph Hellwig1b040712010-09-30 02:25:56 +00001322 return 0;
1323
1324unwind:
1325 while (--msbp >= msb) {
1326 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1327 -msbp->msb_delta, rsvd);
1328 ASSERT(error == 0);
1329 }
1330 spin_unlock(&mp->m_sb_lock);
1331 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332}
1333
1334/*
1335 * xfs_getsb() is called to obtain the buffer for the superblock.
1336 * The buffer is returned locked and read in from disk.
1337 * The buffer should be released with a call to xfs_brelse().
1338 *
1339 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1340 * the superblock buffer if it can be locked without sleeping.
1341 * If it can't then we'll return NULL.
1342 */
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001343struct xfs_buf *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344xfs_getsb(
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001345 struct xfs_mount *mp,
1346 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347{
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001348 struct xfs_buf *bp = mp->m_sb_bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001350 if (!xfs_buf_trylock(bp)) {
1351 if (flags & XBF_TRYLOCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 return NULL;
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001353 xfs_buf_lock(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 }
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001355
Chandra Seetharaman72790aa2011-07-22 23:40:04 +00001356 xfs_buf_hold(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 ASSERT(XFS_BUF_ISDONE(bp));
Jesper Juhl014c2542006-01-15 02:37:08 +01001358 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359}
1360
1361/*
1362 * Used to free the superblock along various error paths.
1363 */
1364void
1365xfs_freesb(
Dave Chinner26af6552010-09-22 10:47:20 +10001366 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367{
Dave Chinner26af6552010-09-22 10:47:20 +10001368 struct xfs_buf *bp = mp->m_sb_bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
Dave Chinner26af6552010-09-22 10:47:20 +10001370 xfs_buf_lock(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 mp->m_sb_bp = NULL;
Dave Chinner26af6552010-09-22 10:47:20 +10001372 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373}
1374
1375/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 * Used to log changes to the superblock unit and width fields which could
Eric Sandeene6957ea2008-04-10 12:19:34 +10001377 * be altered by the mount options, as well as any potential sb_features2
1378 * fixup. Only the first superblock is updated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 */
Christoph Hellwig7884bc82009-01-19 02:04:07 +01001380int
David Chinneree1c0902008-03-06 13:45:50 +11001381xfs_mount_log_sb(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 xfs_mount_t *mp,
1383 __int64_t fields)
1384{
1385 xfs_trans_t *tp;
David Chinnere5720ee2008-04-10 12:21:18 +10001386 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
David Chinneree1c0902008-03-06 13:45:50 +11001388 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
David Chinner4b166de2008-05-20 11:30:27 +10001389 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1390 XFS_SB_VERSIONNUM));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391
1392 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
Jie Liu3d3c8b52013-08-12 20:49:59 +10001393 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
David Chinnere5720ee2008-04-10 12:21:18 +10001394 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 xfs_trans_cancel(tp, 0);
David Chinnere5720ee2008-04-10 12:21:18 +10001396 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 }
1398 xfs_mod_sb(tp, fields);
David Chinnere5720ee2008-04-10 12:21:18 +10001399 error = xfs_trans_commit(tp, 0);
1400 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401}
David Chinner8d280b92006-03-14 13:13:09 +11001402
Christoph Hellwigdda35b82010-02-15 09:44:46 +00001403/*
1404 * If the underlying (data/log/rt) device is readonly, there are some
1405 * operations that cannot proceed.
1406 */
1407int
1408xfs_dev_is_read_only(
1409 struct xfs_mount *mp,
1410 char *message)
1411{
1412 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1413 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1414 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001415 xfs_notice(mp, "%s required on read-only device.", message);
1416 xfs_notice(mp, "write access unavailable, cannot proceed.");
Christoph Hellwigdda35b82010-02-15 09:44:46 +00001417 return EROFS;
1418 }
1419 return 0;
1420}
David Chinner8d280b92006-03-14 13:13:09 +11001421
1422#ifdef HAVE_PERCPU_SB
1423/*
1424 * Per-cpu incore superblock counters
1425 *
1426 * Simple concept, difficult implementation
1427 *
1428 * Basically, replace the incore superblock counters with a distributed per cpu
1429 * counter for contended fields (e.g. free block count).
1430 *
1431 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1432 * hence needs to be accurately read when we are running low on space. Hence
1433 * there is a method to enable and disable the per-cpu counters based on how
1434 * much "stuff" is available in them.
1435 *
1436 * Basically, a counter is enabled if there is enough free resource to justify
1437 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1438 * ENOSPC), then we disable the counters to synchronise all callers and
1439 * re-distribute the available resources.
1440 *
1441 * If, once we redistributed the available resources, we still get a failure,
1442 * we disable the per-cpu counter and go through the slow path.
1443 *
1444 * The slow path is the current xfs_mod_incore_sb() function. This means that
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001445 * when we disable a per-cpu counter, we need to drain its resources back to
David Chinner8d280b92006-03-14 13:13:09 +11001446 * the global superblock. We do this after disabling the counter to prevent
1447 * more threads from queueing up on the counter.
1448 *
1449 * Essentially, this means that we still need a lock in the fast path to enable
1450 * synchronisation between the global counters and the per-cpu counters. This
1451 * is not a problem because the lock will be local to a CPU almost all the time
1452 * and have little contention except when we get to ENOSPC conditions.
1453 *
1454 * Basically, this lock becomes a barrier that enables us to lock out the fast
1455 * path while we do things like enabling and disabling counters and
1456 * synchronising the counters.
1457 *
1458 * Locking rules:
1459 *
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001460 * 1. m_sb_lock before picking up per-cpu locks
David Chinner8d280b92006-03-14 13:13:09 +11001461 * 2. per-cpu locks always picked up via for_each_online_cpu() order
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001462 * 3. accurate counter sync requires m_sb_lock + per cpu locks
David Chinner8d280b92006-03-14 13:13:09 +11001463 * 4. modifying per-cpu counters requires holding per-cpu lock
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001464 * 5. modifying global counters requires holding m_sb_lock
1465 * 6. enabling or disabling a counter requires holding the m_sb_lock
David Chinner8d280b92006-03-14 13:13:09 +11001466 * and _none_ of the per-cpu locks.
1467 *
1468 * Disabled counters are only ever re-enabled by a balance operation
1469 * that results in more free resources per CPU than a given threshold.
1470 * To ensure counters don't remain disabled, they are rebalanced when
1471 * the global resource goes above a higher threshold (i.e. some hysteresis
1472 * is present to prevent thrashing).
David Chinner8d280b92006-03-14 13:13:09 +11001473 */
David Chinnere8234a62006-03-14 13:23:52 +11001474
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001475#ifdef CONFIG_HOTPLUG_CPU
David Chinnere8234a62006-03-14 13:23:52 +11001476/*
1477 * hot-plug CPU notifier support.
1478 *
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001479 * We need a notifier per filesystem as we need to be able to identify
1480 * the filesystem to balance the counters out. This is achieved by
1481 * having a notifier block embedded in the xfs_mount_t and doing pointer
1482 * magic to get the mount pointer from the notifier block address.
David Chinnere8234a62006-03-14 13:23:52 +11001483 */
1484STATIC int
1485xfs_icsb_cpu_notify(
1486 struct notifier_block *nfb,
1487 unsigned long action,
1488 void *hcpu)
1489{
1490 xfs_icsb_cnts_t *cntp;
1491 xfs_mount_t *mp;
David Chinnere8234a62006-03-14 13:23:52 +11001492
1493 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
1494 cntp = (xfs_icsb_cnts_t *)
1495 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
1496 switch (action) {
1497 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001498 case CPU_UP_PREPARE_FROZEN:
David Chinnere8234a62006-03-14 13:23:52 +11001499 /* Easy Case - initialize the area and locks, and
1500 * then rebalance when online does everything else for us. */
David Chinner01e1b692006-03-14 13:29:16 +11001501 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinnere8234a62006-03-14 13:23:52 +11001502 break;
1503 case CPU_ONLINE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001504 case CPU_ONLINE_FROZEN:
David Chinner03135cf2007-02-10 18:35:15 +11001505 xfs_icsb_lock(mp);
Christoph Hellwig45af6c62008-04-22 17:34:44 +10001506 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
1507 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
1508 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
David Chinner03135cf2007-02-10 18:35:15 +11001509 xfs_icsb_unlock(mp);
David Chinnere8234a62006-03-14 13:23:52 +11001510 break;
1511 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001512 case CPU_DEAD_FROZEN:
David Chinnere8234a62006-03-14 13:23:52 +11001513 /* Disable all the counters, then fold the dead cpu's
1514 * count into the total on the global superblock and
1515 * re-enable the counters. */
David Chinner03135cf2007-02-10 18:35:15 +11001516 xfs_icsb_lock(mp);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001517 spin_lock(&mp->m_sb_lock);
David Chinnere8234a62006-03-14 13:23:52 +11001518 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
1519 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
1520 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
1521
1522 mp->m_sb.sb_icount += cntp->icsb_icount;
1523 mp->m_sb.sb_ifree += cntp->icsb_ifree;
1524 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
1525
David Chinner01e1b692006-03-14 13:29:16 +11001526 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinnere8234a62006-03-14 13:23:52 +11001527
Christoph Hellwig45af6c62008-04-22 17:34:44 +10001528 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
1529 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
1530 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001531 spin_unlock(&mp->m_sb_lock);
David Chinner03135cf2007-02-10 18:35:15 +11001532 xfs_icsb_unlock(mp);
David Chinnere8234a62006-03-14 13:23:52 +11001533 break;
1534 }
1535
1536 return NOTIFY_OK;
1537}
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001538#endif /* CONFIG_HOTPLUG_CPU */
David Chinnere8234a62006-03-14 13:23:52 +11001539
David Chinner8d280b92006-03-14 13:13:09 +11001540int
1541xfs_icsb_init_counters(
1542 xfs_mount_t *mp)
1543{
1544 xfs_icsb_cnts_t *cntp;
1545 int i;
1546
1547 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
1548 if (mp->m_sb_cnts == NULL)
1549 return -ENOMEM;
1550
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001551#ifdef CONFIG_HOTPLUG_CPU
David Chinnere8234a62006-03-14 13:23:52 +11001552 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
1553 mp->m_icsb_notifier.priority = 0;
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001554 register_hotcpu_notifier(&mp->m_icsb_notifier);
1555#endif /* CONFIG_HOTPLUG_CPU */
David Chinnere8234a62006-03-14 13:23:52 +11001556
David Chinner8d280b92006-03-14 13:13:09 +11001557 for_each_online_cpu(i) {
1558 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11001559 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinner8d280b92006-03-14 13:13:09 +11001560 }
David Chinner20b64282007-02-10 18:35:09 +11001561
1562 mutex_init(&mp->m_icsb_mutex);
1563
David Chinner8d280b92006-03-14 13:13:09 +11001564 /*
1565 * start with all counters disabled so that the
1566 * initial balance kicks us off correctly
1567 */
1568 mp->m_icsb_counters = -1;
1569 return 0;
1570}
1571
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11001572void
1573xfs_icsb_reinit_counters(
1574 xfs_mount_t *mp)
1575{
1576 xfs_icsb_lock(mp);
1577 /*
1578 * start with all counters disabled so that the
1579 * initial balance kicks us off correctly
1580 */
1581 mp->m_icsb_counters = -1;
Christoph Hellwig45af6c62008-04-22 17:34:44 +10001582 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
1583 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
1584 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11001585 xfs_icsb_unlock(mp);
1586}
1587
Christoph Hellwigc962fb72008-05-20 15:10:52 +10001588void
David Chinner8d280b92006-03-14 13:13:09 +11001589xfs_icsb_destroy_counters(
1590 xfs_mount_t *mp)
1591{
David Chinnere8234a62006-03-14 13:23:52 +11001592 if (mp->m_sb_cnts) {
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001593 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
David Chinner8d280b92006-03-14 13:13:09 +11001594 free_percpu(mp->m_sb_cnts);
David Chinnere8234a62006-03-14 13:23:52 +11001595 }
David Chinner03135cf2007-02-10 18:35:15 +11001596 mutex_destroy(&mp->m_icsb_mutex);
David Chinner8d280b92006-03-14 13:13:09 +11001597}
1598
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00001599STATIC void
David Chinner01e1b692006-03-14 13:29:16 +11001600xfs_icsb_lock_cntr(
1601 xfs_icsb_cnts_t *icsbp)
1602{
1603 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
1604 ndelay(1000);
1605 }
1606}
1607
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00001608STATIC void
David Chinner01e1b692006-03-14 13:29:16 +11001609xfs_icsb_unlock_cntr(
1610 xfs_icsb_cnts_t *icsbp)
1611{
1612 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
1613}
1614
David Chinner8d280b92006-03-14 13:13:09 +11001615
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00001616STATIC void
David Chinner8d280b92006-03-14 13:13:09 +11001617xfs_icsb_lock_all_counters(
1618 xfs_mount_t *mp)
1619{
1620 xfs_icsb_cnts_t *cntp;
1621 int i;
1622
1623 for_each_online_cpu(i) {
1624 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11001625 xfs_icsb_lock_cntr(cntp);
David Chinner8d280b92006-03-14 13:13:09 +11001626 }
1627}
1628
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00001629STATIC void
David Chinner8d280b92006-03-14 13:13:09 +11001630xfs_icsb_unlock_all_counters(
1631 xfs_mount_t *mp)
1632{
1633 xfs_icsb_cnts_t *cntp;
1634 int i;
1635
1636 for_each_online_cpu(i) {
1637 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11001638 xfs_icsb_unlock_cntr(cntp);
David Chinner8d280b92006-03-14 13:13:09 +11001639 }
1640}
1641
1642STATIC void
1643xfs_icsb_count(
1644 xfs_mount_t *mp,
1645 xfs_icsb_cnts_t *cnt,
1646 int flags)
1647{
1648 xfs_icsb_cnts_t *cntp;
1649 int i;
1650
1651 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
1652
1653 if (!(flags & XFS_ICSB_LAZY_COUNT))
1654 xfs_icsb_lock_all_counters(mp);
1655
1656 for_each_online_cpu(i) {
1657 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1658 cnt->icsb_icount += cntp->icsb_icount;
1659 cnt->icsb_ifree += cntp->icsb_ifree;
1660 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
1661 }
1662
1663 if (!(flags & XFS_ICSB_LAZY_COUNT))
1664 xfs_icsb_unlock_all_counters(mp);
1665}
1666
1667STATIC int
1668xfs_icsb_counter_disabled(
1669 xfs_mount_t *mp,
1670 xfs_sb_field_t field)
1671{
1672 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1673 return test_bit(field, &mp->m_icsb_counters);
1674}
1675
David Chinner36fbe6e2008-04-10 12:19:56 +10001676STATIC void
David Chinner8d280b92006-03-14 13:13:09 +11001677xfs_icsb_disable_counter(
1678 xfs_mount_t *mp,
1679 xfs_sb_field_t field)
1680{
1681 xfs_icsb_cnts_t cnt;
1682
1683 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1684
David Chinner20b64282007-02-10 18:35:09 +11001685 /*
1686 * If we are already disabled, then there is nothing to do
1687 * here. We check before locking all the counters to avoid
1688 * the expensive lock operation when being called in the
1689 * slow path and the counter is already disabled. This is
1690 * safe because the only time we set or clear this state is under
1691 * the m_icsb_mutex.
1692 */
1693 if (xfs_icsb_counter_disabled(mp, field))
David Chinner36fbe6e2008-04-10 12:19:56 +10001694 return;
David Chinner20b64282007-02-10 18:35:09 +11001695
David Chinner8d280b92006-03-14 13:13:09 +11001696 xfs_icsb_lock_all_counters(mp);
1697 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
1698 /* drain back to superblock */
1699
Christoph Hellwigce461932008-04-22 17:34:50 +10001700 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
David Chinner8d280b92006-03-14 13:13:09 +11001701 switch(field) {
1702 case XFS_SBS_ICOUNT:
1703 mp->m_sb.sb_icount = cnt.icsb_icount;
1704 break;
1705 case XFS_SBS_IFREE:
1706 mp->m_sb.sb_ifree = cnt.icsb_ifree;
1707 break;
1708 case XFS_SBS_FDBLOCKS:
1709 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
1710 break;
1711 default:
1712 BUG();
1713 }
1714 }
1715
1716 xfs_icsb_unlock_all_counters(mp);
David Chinner8d280b92006-03-14 13:13:09 +11001717}
1718
1719STATIC void
1720xfs_icsb_enable_counter(
1721 xfs_mount_t *mp,
1722 xfs_sb_field_t field,
1723 uint64_t count,
1724 uint64_t resid)
1725{
1726 xfs_icsb_cnts_t *cntp;
1727 int i;
1728
1729 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1730
1731 xfs_icsb_lock_all_counters(mp);
1732 for_each_online_cpu(i) {
1733 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
1734 switch (field) {
1735 case XFS_SBS_ICOUNT:
1736 cntp->icsb_icount = count + resid;
1737 break;
1738 case XFS_SBS_IFREE:
1739 cntp->icsb_ifree = count + resid;
1740 break;
1741 case XFS_SBS_FDBLOCKS:
1742 cntp->icsb_fdblocks = count + resid;
1743 break;
1744 default:
1745 BUG();
1746 break;
1747 }
1748 resid = 0;
1749 }
1750 clear_bit(field, &mp->m_icsb_counters);
1751 xfs_icsb_unlock_all_counters(mp);
1752}
1753
David Chinnerdbcabad2007-02-10 18:36:17 +11001754void
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10001755xfs_icsb_sync_counters_locked(
David Chinner8d280b92006-03-14 13:13:09 +11001756 xfs_mount_t *mp,
1757 int flags)
1758{
1759 xfs_icsb_cnts_t cnt;
David Chinner8d280b92006-03-14 13:13:09 +11001760
David Chinner8d280b92006-03-14 13:13:09 +11001761 xfs_icsb_count(mp, &cnt, flags);
1762
David Chinner8d280b92006-03-14 13:13:09 +11001763 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
1764 mp->m_sb.sb_icount = cnt.icsb_icount;
1765 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
1766 mp->m_sb.sb_ifree = cnt.icsb_ifree;
1767 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
1768 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
David Chinner8d280b92006-03-14 13:13:09 +11001769}
1770
1771/*
1772 * Accurate update of per-cpu counters to incore superblock
1773 */
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10001774void
David Chinner8d280b92006-03-14 13:13:09 +11001775xfs_icsb_sync_counters(
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10001776 xfs_mount_t *mp,
1777 int flags)
David Chinner8d280b92006-03-14 13:13:09 +11001778{
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10001779 spin_lock(&mp->m_sb_lock);
1780 xfs_icsb_sync_counters_locked(mp, flags);
1781 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001782}
1783
1784/*
1785 * Balance and enable/disable counters as necessary.
1786 *
David Chinner20b64282007-02-10 18:35:09 +11001787 * Thresholds for re-enabling counters are somewhat magic. inode counts are
1788 * chosen to be the same number as single on disk allocation chunk per CPU, and
1789 * free blocks is something far enough zero that we aren't going thrash when we
1790 * get near ENOSPC. We also need to supply a minimum we require per cpu to
1791 * prevent looping endlessly when xfs_alloc_space asks for more than will
1792 * be distributed to a single CPU but each CPU has enough blocks to be
1793 * reenabled.
1794 *
1795 * Note that we can be called when counters are already disabled.
1796 * xfs_icsb_disable_counter() optimises the counter locking in this case to
1797 * prevent locking every per-cpu counter needlessly.
David Chinner8d280b92006-03-14 13:13:09 +11001798 */
David Chinner20b64282007-02-10 18:35:09 +11001799
1800#define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
David Chinner4be536d2006-09-07 14:26:50 +10001801#define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
David Chinner20b64282007-02-10 18:35:09 +11001802 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
David Chinner8d280b92006-03-14 13:13:09 +11001803STATIC void
Christoph Hellwig45af6c62008-04-22 17:34:44 +10001804xfs_icsb_balance_counter_locked(
David Chinner8d280b92006-03-14 13:13:09 +11001805 xfs_mount_t *mp,
1806 xfs_sb_field_t field,
David Chinner20b64282007-02-10 18:35:09 +11001807 int min_per_cpu)
David Chinner8d280b92006-03-14 13:13:09 +11001808{
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001809 uint64_t count, resid;
David Chinner8d280b92006-03-14 13:13:09 +11001810 int weight = num_online_cpus();
David Chinner20b64282007-02-10 18:35:09 +11001811 uint64_t min = (uint64_t)min_per_cpu;
David Chinner8d280b92006-03-14 13:13:09 +11001812
David Chinner8d280b92006-03-14 13:13:09 +11001813 /* disable counter and sync counter */
1814 xfs_icsb_disable_counter(mp, field);
1815
1816 /* update counters - first CPU gets residual*/
1817 switch (field) {
1818 case XFS_SBS_ICOUNT:
1819 count = mp->m_sb.sb_icount;
1820 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11001821 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10001822 return;
David Chinner8d280b92006-03-14 13:13:09 +11001823 break;
1824 case XFS_SBS_IFREE:
1825 count = mp->m_sb.sb_ifree;
1826 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11001827 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10001828 return;
David Chinner8d280b92006-03-14 13:13:09 +11001829 break;
1830 case XFS_SBS_FDBLOCKS:
1831 count = mp->m_sb.sb_fdblocks;
1832 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11001833 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10001834 return;
David Chinner8d280b92006-03-14 13:13:09 +11001835 break;
1836 default:
1837 BUG();
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001838 count = resid = 0; /* quiet, gcc */
David Chinner8d280b92006-03-14 13:13:09 +11001839 break;
1840 }
1841
1842 xfs_icsb_enable_counter(mp, field, count, resid);
Christoph Hellwig45af6c62008-04-22 17:34:44 +10001843}
1844
1845STATIC void
1846xfs_icsb_balance_counter(
1847 xfs_mount_t *mp,
1848 xfs_sb_field_t fields,
1849 int min_per_cpu)
1850{
1851 spin_lock(&mp->m_sb_lock);
1852 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
1853 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001854}
1855
Christoph Hellwig1b040712010-09-30 02:25:56 +00001856int
David Chinner20b64282007-02-10 18:35:09 +11001857xfs_icsb_modify_counters(
David Chinner8d280b92006-03-14 13:13:09 +11001858 xfs_mount_t *mp,
1859 xfs_sb_field_t field,
David Chinner20f4ebf2007-02-10 18:36:10 +11001860 int64_t delta,
David Chinner20b64282007-02-10 18:35:09 +11001861 int rsvd)
David Chinner8d280b92006-03-14 13:13:09 +11001862{
1863 xfs_icsb_cnts_t *icsbp;
1864 long long lcounter; /* long counter for 64 bit fields */
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09001865 int ret = 0;
David Chinner8d280b92006-03-14 13:13:09 +11001866
David Chinner20b64282007-02-10 18:35:09 +11001867 might_sleep();
David Chinner8d280b92006-03-14 13:13:09 +11001868again:
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09001869 preempt_disable();
1870 icsbp = this_cpu_ptr(mp->m_sb_cnts);
David Chinner20b64282007-02-10 18:35:09 +11001871
1872 /*
1873 * if the counter is disabled, go to slow path
1874 */
David Chinner8d280b92006-03-14 13:13:09 +11001875 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
1876 goto slow_path;
David Chinner20b64282007-02-10 18:35:09 +11001877 xfs_icsb_lock_cntr(icsbp);
1878 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
1879 xfs_icsb_unlock_cntr(icsbp);
1880 goto slow_path;
1881 }
David Chinner8d280b92006-03-14 13:13:09 +11001882
1883 switch (field) {
1884 case XFS_SBS_ICOUNT:
1885 lcounter = icsbp->icsb_icount;
1886 lcounter += delta;
1887 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11001888 goto balance_counter;
David Chinner8d280b92006-03-14 13:13:09 +11001889 icsbp->icsb_icount = lcounter;
1890 break;
1891
1892 case XFS_SBS_IFREE:
1893 lcounter = icsbp->icsb_ifree;
1894 lcounter += delta;
1895 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11001896 goto balance_counter;
David Chinner8d280b92006-03-14 13:13:09 +11001897 icsbp->icsb_ifree = lcounter;
1898 break;
1899
1900 case XFS_SBS_FDBLOCKS:
1901 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
1902
David Chinner4be536d2006-09-07 14:26:50 +10001903 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
David Chinner8d280b92006-03-14 13:13:09 +11001904 lcounter += delta;
1905 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11001906 goto balance_counter;
David Chinner4be536d2006-09-07 14:26:50 +10001907 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
David Chinner8d280b92006-03-14 13:13:09 +11001908 break;
1909 default:
1910 BUG();
1911 break;
1912 }
David Chinner01e1b692006-03-14 13:29:16 +11001913 xfs_icsb_unlock_cntr(icsbp);
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09001914 preempt_enable();
David Chinner8d280b92006-03-14 13:13:09 +11001915 return 0;
1916
David Chinner8d280b92006-03-14 13:13:09 +11001917slow_path:
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09001918 preempt_enable();
David Chinner20b64282007-02-10 18:35:09 +11001919
1920 /*
1921 * serialise with a mutex so we don't burn lots of cpu on
1922 * the superblock lock. We still need to hold the superblock
1923 * lock, however, when we modify the global structures.
1924 */
David Chinner03135cf2007-02-10 18:35:15 +11001925 xfs_icsb_lock(mp);
David Chinner20b64282007-02-10 18:35:09 +11001926
1927 /*
1928 * Now running atomically.
1929 *
1930 * If the counter is enabled, someone has beaten us to rebalancing.
1931 * Drop the lock and try again in the fast path....
1932 */
1933 if (!(xfs_icsb_counter_disabled(mp, field))) {
David Chinner03135cf2007-02-10 18:35:15 +11001934 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11001935 goto again;
1936 }
1937
1938 /*
1939 * The counter is currently disabled. Because we are
1940 * running atomically here, we know a rebalance cannot
1941 * be in progress. Hence we can go straight to operating
1942 * on the global superblock. We do not call xfs_mod_incore_sb()
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001943 * here even though we need to get the m_sb_lock. Doing so
David Chinner20b64282007-02-10 18:35:09 +11001944 * will cause us to re-enter this function and deadlock.
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001945 * Hence we get the m_sb_lock ourselves and then call
David Chinner20b64282007-02-10 18:35:09 +11001946 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
1947 * directly on the global counters.
1948 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001949 spin_lock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11001950 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001951 spin_unlock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11001952
1953 /*
1954 * Now that we've modified the global superblock, we
1955 * may be able to re-enable the distributed counters
1956 * (e.g. lots of space just got freed). After that
1957 * we are done.
1958 */
1959 if (ret != ENOSPC)
Christoph Hellwig45af6c62008-04-22 17:34:44 +10001960 xfs_icsb_balance_counter(mp, field, 0);
David Chinner03135cf2007-02-10 18:35:15 +11001961 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11001962 return ret;
1963
1964balance_counter:
David Chinner01e1b692006-03-14 13:29:16 +11001965 xfs_icsb_unlock_cntr(icsbp);
Christoph Lameter7a9e02d2009-10-03 19:48:23 +09001966 preempt_enable();
David Chinner8d280b92006-03-14 13:13:09 +11001967
David Chinner20b64282007-02-10 18:35:09 +11001968 /*
1969 * We may have multiple threads here if multiple per-cpu
1970 * counters run dry at the same time. This will mean we can
1971 * do more balances than strictly necessary but it is not
1972 * the common slowpath case.
1973 */
David Chinner03135cf2007-02-10 18:35:15 +11001974 xfs_icsb_lock(mp);
David Chinner8d280b92006-03-14 13:13:09 +11001975
David Chinner20b64282007-02-10 18:35:09 +11001976 /*
1977 * running atomically.
1978 *
1979 * This will leave the counter in the correct state for future
1980 * accesses. After the rebalance, we simply try again and our retry
1981 * will either succeed through the fast path or slow path without
1982 * another balance operation being required.
1983 */
Christoph Hellwig45af6c62008-04-22 17:34:44 +10001984 xfs_icsb_balance_counter(mp, field, delta);
David Chinner03135cf2007-02-10 18:35:15 +11001985 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11001986 goto again;
David Chinner8d280b92006-03-14 13:13:09 +11001987}
1988
David Chinner8d280b92006-03-14 13:13:09 +11001989#endif