blob: 3c97c6463a4e418af83145d350d7aac3e49cc617 [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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
David Chinnere5720ee2008-04-10 12:21:18 +100048STATIC int xfs_mount_log_sb(xfs_mount_t *, __int64_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -070049STATIC int xfs_uuid_mount(xfs_mount_t *);
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
125/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 * Free up the resources associated with a mount structure. Assume that
127 * the structure was initially zeroed, so we can tell which fields got
128 * initialized.
129 */
Christoph Hellwigc962fb72008-05-20 15:10:52 +1000130STATIC void
Christoph Hellwigff4f0382008-08-13 16:50:47 +1000131xfs_free_perag(
Christoph Hellwig745f6912007-08-30 17:20:39 +1000132 xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 if (mp->m_perag) {
135 int agno;
136
137 for (agno = 0; agno < mp->m_maxagi; agno++)
138 if (mp->m_perag[agno].pagb_list)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +1000139 kmem_free(mp->m_perag[agno].pagb_list);
140 kmem_free(mp->m_perag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142}
143
Nathan Scott4cc929e2007-05-14 18:24:02 +1000144/*
145 * Check size of device based on the (data/realtime) block count.
146 * Note: this check is used by the growfs code as well as mount.
147 */
148int
149xfs_sb_validate_fsb_count(
150 xfs_sb_t *sbp,
151 __uint64_t nblocks)
152{
153 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
154 ASSERT(sbp->sb_blocklog >= BBSHIFT);
155
156#if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
157 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
158 return E2BIG;
159#else /* Limited by UINT_MAX of sectors */
160 if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
161 return E2BIG;
162#endif
163 return 0;
164}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165
166/*
167 * Check the validity of the SB found.
168 */
169STATIC int
170xfs_mount_validate_sb(
171 xfs_mount_t *mp,
Nathan Scott764d1f82006-03-31 13:04:17 +1000172 xfs_sb_t *sbp,
173 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174{
175 /*
176 * If the log device and data device have the
177 * same device number, the log is internal.
178 * Consequently, the sb_logstart should be non-zero. If
179 * we have a zero sb_logstart in this case, we may be trying to mount
180 * a volume filesystem in a non-volume manner.
181 */
182 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000183 xfs_fs_mount_cmn_err(flags, "bad magic number");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 return XFS_ERROR(EWRONGFS);
185 }
186
Eric Sandeen62118702008-03-06 13:44:28 +1100187 if (!xfs_sb_good_version(sbp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000188 xfs_fs_mount_cmn_err(flags, "bad version");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 return XFS_ERROR(EWRONGFS);
190 }
191
192 if (unlikely(
193 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000194 xfs_fs_mount_cmn_err(flags,
195 "filesystem is marked as having an external log; "
196 "specify logdev on the\nmount command line.");
197 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 }
199
200 if (unlikely(
201 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000202 xfs_fs_mount_cmn_err(flags,
203 "filesystem is marked as having an internal log; "
204 "do not specify logdev on\nthe mount command line.");
205 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 }
207
208 /*
209 * More sanity checking. These were stolen directly from
210 * xfs_repair.
211 */
212 if (unlikely(
213 sbp->sb_agcount <= 0 ||
214 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
215 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
216 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
217 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
218 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
219 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
220 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
221 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
222 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
223 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
Nathan Scott9f989c92006-03-14 13:29:32 +1100224 sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
225 sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
226 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
228 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
Nathan Scotte50bd162006-04-11 15:10:45 +1000229 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000230 xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 return XFS_ERROR(EFSCORRUPTED);
232 }
233
234 /*
235 * Sanity check AG count, size fields against data size field
236 */
237 if (unlikely(
238 sbp->sb_dblocks == 0 ||
239 sbp->sb_dblocks >
240 (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
241 sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
242 sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000243 xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 return XFS_ERROR(EFSCORRUPTED);
245 }
246
Lachlan McIlroy2edbddd2008-06-27 13:34:34 +1000247 /*
248 * Until this is fixed only page-sized or smaller data blocks work.
249 */
250 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
251 xfs_fs_mount_cmn_err(flags,
252 "file system with blocksize %d bytes",
253 sbp->sb_blocksize);
254 xfs_fs_mount_cmn_err(flags,
255 "only pagesize (%ld) or less will currently work.",
256 PAGE_SIZE);
257 return XFS_ERROR(ENOSYS);
258 }
259
Nathan Scott4cc929e2007-05-14 18:24:02 +1000260 if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
261 xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000262 xfs_fs_mount_cmn_err(flags,
263 "file system too large to be mounted on this system.");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 return XFS_ERROR(E2BIG);
265 }
266
267 if (unlikely(sbp->sb_inprogress)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000268 xfs_fs_mount_cmn_err(flags, "file system busy");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 return XFS_ERROR(EFSCORRUPTED);
270 }
271
272 /*
Nathan Scottde206142005-05-05 13:24:13 -0700273 * Version 1 directory format has never worked on Linux.
274 */
Eric Sandeen62118702008-03-06 13:44:28 +1100275 if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000276 xfs_fs_mount_cmn_err(flags,
277 "file system using version 1 directory format");
Nathan Scottde206142005-05-05 13:24:13 -0700278 return XFS_ERROR(ENOSYS);
279 }
280
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 return 0;
282}
283
David Chinnerda353b02007-08-28 14:00:13 +1000284STATIC void
285xfs_initialize_perag_icache(
286 xfs_perag_t *pag)
287{
288 if (!pag->pag_ici_init) {
289 rwlock_init(&pag->pag_ici_lock);
290 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
291 pag->pag_ici_init = 1;
292 }
293}
294
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295xfs_agnumber_t
Nathan Scottc11e2c32005-11-02 15:11:45 +1100296xfs_initialize_perag(
Nathan Scottc11e2c32005-11-02 15:11:45 +1100297 xfs_mount_t *mp,
298 xfs_agnumber_t agcount)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
300 xfs_agnumber_t index, max_metadata;
301 xfs_perag_t *pag;
302 xfs_agino_t agino;
303 xfs_ino_t ino;
304 xfs_sb_t *sbp = &mp->m_sb;
305 xfs_ino_t max_inum = XFS_MAXINUMBER_32;
306
307 /* Check to see if the filesystem can overflow 32 bit inodes */
308 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
309 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
310
311 /* Clear the mount flag if no inode can overflow 32 bits
312 * on this filesystem, or if specifically requested..
313 */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000314 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > max_inum) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 mp->m_flags |= XFS_MOUNT_32BITINODES;
316 } else {
317 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
318 }
319
320 /* If we can overflow then setup the ag headers accordingly */
321 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
322 /* Calculate how much should be reserved for inodes to
323 * meet the max inode percentage.
324 */
325 if (mp->m_maxicount) {
326 __uint64_t icount;
327
328 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
329 do_div(icount, 100);
330 icount += sbp->sb_agblocks - 1;
Eric Sandeena749ee82005-11-02 15:13:42 +1100331 do_div(icount, sbp->sb_agblocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 max_metadata = icount;
333 } else {
334 max_metadata = agcount;
335 }
336 for (index = 0; index < agcount; index++) {
337 ino = XFS_AGINO_TO_INO(mp, index, agino);
338 if (ino > max_inum) {
339 index++;
340 break;
341 }
342
Nathan Scottc41564b2006-03-29 08:55:14 +1000343 /* This ag is preferred for inodes */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 pag = &mp->m_perag[index];
345 pag->pagi_inodeok = 1;
346 if (index < max_metadata)
347 pag->pagf_metadata = 1;
David Chinnerda353b02007-08-28 14:00:13 +1000348 xfs_initialize_perag_icache(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 }
350 } else {
351 /* Setup default behavior for smaller filesystems */
352 for (index = 0; index < agcount; index++) {
353 pag = &mp->m_perag[index];
354 pag->pagi_inodeok = 1;
David Chinnerda353b02007-08-28 14:00:13 +1000355 xfs_initialize_perag_icache(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 }
357 }
358 return index;
359}
360
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000361void
362xfs_sb_from_disk(
363 xfs_sb_t *to,
364 xfs_dsb_t *from)
365{
366 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
367 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
368 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
369 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
370 to->sb_rextents = be64_to_cpu(from->sb_rextents);
371 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
372 to->sb_logstart = be64_to_cpu(from->sb_logstart);
373 to->sb_rootino = be64_to_cpu(from->sb_rootino);
374 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
375 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
376 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
377 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
378 to->sb_agcount = be32_to_cpu(from->sb_agcount);
379 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
380 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
381 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
382 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
383 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
384 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
385 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
386 to->sb_blocklog = from->sb_blocklog;
387 to->sb_sectlog = from->sb_sectlog;
388 to->sb_inodelog = from->sb_inodelog;
389 to->sb_inopblog = from->sb_inopblog;
390 to->sb_agblklog = from->sb_agblklog;
391 to->sb_rextslog = from->sb_rextslog;
392 to->sb_inprogress = from->sb_inprogress;
393 to->sb_imax_pct = from->sb_imax_pct;
394 to->sb_icount = be64_to_cpu(from->sb_icount);
395 to->sb_ifree = be64_to_cpu(from->sb_ifree);
396 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
397 to->sb_frextents = be64_to_cpu(from->sb_frextents);
398 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
399 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
400 to->sb_qflags = be16_to_cpu(from->sb_qflags);
401 to->sb_flags = from->sb_flags;
402 to->sb_shared_vn = from->sb_shared_vn;
403 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
404 to->sb_unit = be32_to_cpu(from->sb_unit);
405 to->sb_width = be32_to_cpu(from->sb_width);
406 to->sb_dirblklog = from->sb_dirblklog;
407 to->sb_logsectlog = from->sb_logsectlog;
408 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
409 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
410 to->sb_features2 = be32_to_cpu(from->sb_features2);
David Chinneree1c0902008-03-06 13:45:50 +1100411 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000412}
413
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414/*
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000415 * Copy in core superblock to ondisk one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 *
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000417 * The fields argument is mask of superblock fields to copy.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 */
419void
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000420xfs_sb_to_disk(
421 xfs_dsb_t *to,
422 xfs_sb_t *from,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 __int64_t fields)
424{
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000425 xfs_caddr_t to_ptr = (xfs_caddr_t)to;
426 xfs_caddr_t from_ptr = (xfs_caddr_t)from;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 xfs_sb_field_t f;
428 int first;
429 int size;
430
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 ASSERT(fields);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 if (!fields)
433 return;
434
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 while (fields) {
436 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
437 first = xfs_sb_info[f].offset;
438 size = xfs_sb_info[f + 1].offset - first;
439
440 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
441
442 if (size == 1 || xfs_sb_info[f].type == 1) {
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000443 memcpy(to_ptr + first, from_ptr + first, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 } else {
445 switch (size) {
446 case 2:
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000447 *(__be16 *)(to_ptr + first) =
448 cpu_to_be16(*(__u16 *)(from_ptr + first));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 break;
450 case 4:
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000451 *(__be32 *)(to_ptr + first) =
452 cpu_to_be32(*(__u32 *)(from_ptr + first));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 break;
454 case 8:
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000455 *(__be64 *)(to_ptr + first) =
456 cpu_to_be64(*(__u64 *)(from_ptr + first));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 break;
458 default:
459 ASSERT(0);
460 }
461 }
462
463 fields &= ~(1LL << f);
464 }
465}
466
467/*
468 * xfs_readsb
469 *
470 * Does the initial read of the superblock.
471 */
472int
Nathan Scott764d1f82006-03-31 13:04:17 +1000473xfs_readsb(xfs_mount_t *mp, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474{
475 unsigned int sector_size;
476 unsigned int extra_flags;
477 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 int error;
479
480 ASSERT(mp->m_sb_bp == NULL);
481 ASSERT(mp->m_ddev_targp != NULL);
482
483 /*
484 * Allocate a (locked) buffer to hold the superblock.
485 * This will be kept around at all times to optimize
486 * access to the superblock.
487 */
488 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
489 extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED;
490
491 bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
492 BTOBB(sector_size), extra_flags);
493 if (!bp || XFS_BUF_ISERROR(bp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000494 xfs_fs_mount_cmn_err(flags, "SB read failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
496 goto fail;
497 }
498 ASSERT(XFS_BUF_ISBUSY(bp));
499 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
500
501 /*
502 * Initialize the mount structure from the superblock.
503 * But first do some basic consistency checking.
504 */
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000505 xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
Nathan Scott764d1f82006-03-31 13:04:17 +1000507 error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 if (error) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000509 xfs_fs_mount_cmn_err(flags, "SB validate failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 goto fail;
511 }
512
513 /*
514 * We must be able to do sector-sized and sector-aligned IO.
515 */
516 if (sector_size > mp->m_sb.sb_sectsize) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000517 xfs_fs_mount_cmn_err(flags,
518 "device supports only %u byte sectors (not %u)",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 sector_size, mp->m_sb.sb_sectsize);
520 error = ENOSYS;
521 goto fail;
522 }
523
524 /*
525 * If device sector size is smaller than the superblock size,
526 * re-read the superblock so the buffer is correctly sized.
527 */
528 if (sector_size < mp->m_sb.sb_sectsize) {
529 XFS_BUF_UNMANAGE(bp);
530 xfs_buf_relse(bp);
531 sector_size = mp->m_sb.sb_sectsize;
532 bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
533 BTOBB(sector_size), extra_flags);
534 if (!bp || XFS_BUF_ISERROR(bp)) {
Nathan Scott764d1f82006-03-31 13:04:17 +1000535 xfs_fs_mount_cmn_err(flags, "SB re-read failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
537 goto fail;
538 }
539 ASSERT(XFS_BUF_ISBUSY(bp));
540 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
541 }
542
Lachlan McIlroy5478eea2007-02-10 18:36:29 +1100543 /* Initialize per-cpu counters */
544 xfs_icsb_reinit_counters(mp);
David Chinner8d280b92006-03-14 13:13:09 +1100545
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 mp->m_sb_bp = bp;
547 xfs_buf_relse(bp);
548 ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
549 return 0;
550
551 fail:
552 if (bp) {
553 XFS_BUF_UNMANAGE(bp);
554 xfs_buf_relse(bp);
555 }
556 return error;
557}
558
559
560/*
561 * xfs_mount_common
562 *
563 * Mount initialization code establishing various mount
564 * fields from the superblock associated with the given
565 * mount structure
566 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000567STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
569{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 mp->m_agfrotor = mp->m_agirotor = 0;
Eric Sandeen007c61c2007-10-11 17:43:56 +1000571 spin_lock_init(&mp->m_agirotor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 mp->m_maxagi = mp->m_sb.sb_agcount;
573 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
574 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
575 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
576 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
577 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100578 mp->m_litino = sbp->sb_inodesize - sizeof(struct xfs_dinode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 mp->m_blockmask = sbp->sb_blocksize - 1;
580 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
581 mp->m_blockwmask = mp->m_blockwsize - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582
583 /*
584 * Setup for attributes, in case they get created.
585 * This value is for inodes getting attributes for the first time,
586 * the per-inode value is for old attribute values.
587 */
588 ASSERT(sbp->sb_inodesize >= 256 && sbp->sb_inodesize <= 2048);
589 switch (sbp->sb_inodesize) {
590 case 256:
Nathan Scottd8cc8902005-11-02 10:34:53 +1100591 mp->m_attroffset = XFS_LITINO(mp) -
592 XFS_BMDR_SPACE_CALC(MINABTPTRS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 break;
594 case 512:
595 case 1024:
596 case 2048:
Nathan Scottd8cc8902005-11-02 10:34:53 +1100597 mp->m_attroffset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 break;
599 default:
600 ASSERT(0);
601 }
602 ASSERT(mp->m_attroffset < XFS_LITINO(mp));
603
Christoph Hellwig60197e82008-10-30 17:11:19 +1100604 mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
605 mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
606 mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
607 mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
608
609 mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
610 mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
611 mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
612 mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
613
614 mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
615 mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
616 mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
617 mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
619 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
620 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
621 sbp->sb_inopblock);
622 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
623}
David Chinner92821e22007-05-24 15:26:31 +1000624
625/*
626 * xfs_initialize_perag_data
627 *
628 * Read in each per-ag structure so we can count up the number of
629 * allocated inodes, free inodes and used filesystem blocks as this
630 * information is no longer persistent in the superblock. Once we have
631 * this information, write it into the in-core superblock structure.
632 */
633STATIC int
634xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
635{
636 xfs_agnumber_t index;
637 xfs_perag_t *pag;
638 xfs_sb_t *sbp = &mp->m_sb;
639 uint64_t ifree = 0;
640 uint64_t ialloc = 0;
641 uint64_t bfree = 0;
642 uint64_t bfreelst = 0;
643 uint64_t btree = 0;
644 int error;
David Chinner92821e22007-05-24 15:26:31 +1000645
646 for (index = 0; index < agcount; index++) {
647 /*
648 * read the agf, then the agi. This gets us
649 * all the inforamtion we need and populates the
650 * per-ag structures for us.
651 */
652 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
653 if (error)
654 return error;
655
656 error = xfs_ialloc_pagi_init(mp, NULL, index);
657 if (error)
658 return error;
659 pag = &mp->m_perag[index];
660 ifree += pag->pagi_freecount;
661 ialloc += pag->pagi_count;
662 bfree += pag->pagf_freeblks;
663 bfreelst += pag->pagf_flcount;
664 btree += pag->pagf_btreeblks;
665 }
666 /*
667 * Overwrite incore superblock counters with just-read data
668 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +1000669 spin_lock(&mp->m_sb_lock);
David Chinner92821e22007-05-24 15:26:31 +1000670 sbp->sb_ifree = ifree;
671 sbp->sb_icount = ialloc;
672 sbp->sb_fdblocks = bfree + bfreelst + btree;
Eric Sandeen3685c2a2007-10-11 17:42:32 +1000673 spin_unlock(&mp->m_sb_lock);
David Chinner92821e22007-05-24 15:26:31 +1000674
675 /* Fixup the per-cpu counters as well. */
676 xfs_icsb_reinit_counters(mp);
677
678 return 0;
679}
680
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681/*
Eric Sandeen0771fb42007-10-12 11:03:40 +1000682 * Update alignment values based on mount options and sb values
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 */
Eric Sandeen0771fb42007-10-12 11:03:40 +1000684STATIC int
Christoph Hellwig42490232008-08-13 16:49:32 +1000685xfs_update_alignment(xfs_mount_t *mp, __uint64_t *update_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 xfs_sb_t *sbp = &(mp->m_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688
Christoph Hellwig42490232008-08-13 16:49:32 +1000689 if (mp->m_dalign) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 /*
691 * If stripe unit and stripe width are not multiples
692 * of the fs blocksize turn off alignment.
693 */
694 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
695 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
696 if (mp->m_flags & XFS_MOUNT_RETERR) {
697 cmn_err(CE_WARN,
698 "XFS: alignment check 1 failed");
Eric Sandeen0771fb42007-10-12 11:03:40 +1000699 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 }
701 mp->m_dalign = mp->m_swidth = 0;
702 } else {
703 /*
704 * Convert the stripe unit and width to FSBs.
705 */
706 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
707 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
708 if (mp->m_flags & XFS_MOUNT_RETERR) {
Eric Sandeen0771fb42007-10-12 11:03:40 +1000709 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 }
711 xfs_fs_cmn_err(CE_WARN, mp,
712"stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
713 mp->m_dalign, mp->m_swidth,
714 sbp->sb_agblocks);
715
716 mp->m_dalign = 0;
717 mp->m_swidth = 0;
718 } else if (mp->m_dalign) {
719 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
720 } else {
721 if (mp->m_flags & XFS_MOUNT_RETERR) {
722 xfs_fs_cmn_err(CE_WARN, mp,
723"stripe alignment turned off: sunit(%d) less than bsize(%d)",
724 mp->m_dalign,
725 mp->m_blockmask +1);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000726 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 }
728 mp->m_swidth = 0;
729 }
730 }
731
732 /*
733 * Update superblock with new values
734 * and log changes
735 */
Eric Sandeen62118702008-03-06 13:44:28 +1100736 if (xfs_sb_version_hasdalign(sbp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 if (sbp->sb_unit != mp->m_dalign) {
738 sbp->sb_unit = mp->m_dalign;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000739 *update_flags |= XFS_SB_UNIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 }
741 if (sbp->sb_width != mp->m_swidth) {
742 sbp->sb_width = mp->m_swidth;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000743 *update_flags |= XFS_SB_WIDTH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 }
745 }
746 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
Eric Sandeen62118702008-03-06 13:44:28 +1100747 xfs_sb_version_hasdalign(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 mp->m_dalign = sbp->sb_unit;
749 mp->m_swidth = sbp->sb_width;
750 }
751
Eric Sandeen0771fb42007-10-12 11:03:40 +1000752 return 0;
753}
754
755/*
756 * Set the maximum inode count for this filesystem
757 */
758STATIC void
759xfs_set_maxicount(xfs_mount_t *mp)
760{
761 xfs_sb_t *sbp = &(mp->m_sb);
762 __uint64_t icount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763
764 if (sbp->sb_imax_pct) {
Eric Sandeen0771fb42007-10-12 11:03:40 +1000765 /*
766 * Make sure the maximum inode count is a multiple
767 * of the units we allocate inodes in.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
770 do_div(icount, 100);
771 do_div(icount, mp->m_ialloc_blks);
772 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
773 sbp->sb_inopblog;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000774 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 mp->m_maxicount = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 }
Eric Sandeen0771fb42007-10-12 11:03:40 +1000777}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778
Eric Sandeen0771fb42007-10-12 11:03:40 +1000779/*
780 * Set the default minimum read and write sizes unless
781 * already specified in a mount option.
782 * We use smaller I/O sizes when the file system
783 * is being used for NFS service (wsync mount option).
784 */
785STATIC void
786xfs_set_rw_sizes(xfs_mount_t *mp)
787{
788 xfs_sb_t *sbp = &(mp->m_sb);
789 int readio_log, writeio_log;
790
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
792 if (mp->m_flags & XFS_MOUNT_WSYNC) {
793 readio_log = XFS_WSYNC_READIO_LOG;
794 writeio_log = XFS_WSYNC_WRITEIO_LOG;
795 } else {
796 readio_log = XFS_READIO_LOG_LARGE;
797 writeio_log = XFS_WRITEIO_LOG_LARGE;
798 }
799 } else {
800 readio_log = mp->m_readio_log;
801 writeio_log = mp->m_writeio_log;
802 }
803
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 if (sbp->sb_blocklog > readio_log) {
805 mp->m_readio_log = sbp->sb_blocklog;
806 } else {
807 mp->m_readio_log = readio_log;
808 }
809 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
810 if (sbp->sb_blocklog > writeio_log) {
811 mp->m_writeio_log = sbp->sb_blocklog;
812 } else {
813 mp->m_writeio_log = writeio_log;
814 }
815 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000816}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817
Eric Sandeen0771fb42007-10-12 11:03:40 +1000818/*
819 * Set whether we're using inode alignment.
820 */
821STATIC void
822xfs_set_inoalignment(xfs_mount_t *mp)
823{
Eric Sandeen62118702008-03-06 13:44:28 +1100824 if (xfs_sb_version_hasalign(&mp->m_sb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 mp->m_sb.sb_inoalignmt >=
826 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
827 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
828 else
829 mp->m_inoalign_mask = 0;
830 /*
831 * If we are using stripe alignment, check whether
832 * the stripe unit is a multiple of the inode alignment
833 */
834 if (mp->m_dalign && mp->m_inoalign_mask &&
835 !(mp->m_dalign & mp->m_inoalign_mask))
836 mp->m_sinoalign = mp->m_dalign;
837 else
838 mp->m_sinoalign = 0;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000839}
840
841/*
842 * Check that the data (and log if separate) are an ok size.
843 */
844STATIC int
Christoph Hellwig42490232008-08-13 16:49:32 +1000845xfs_check_sizes(xfs_mount_t *mp)
Eric Sandeen0771fb42007-10-12 11:03:40 +1000846{
847 xfs_buf_t *bp;
848 xfs_daddr_t d;
849 int error;
850
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
852 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
853 cmn_err(CE_WARN, "XFS: size check 1 failed");
Eric Sandeen0771fb42007-10-12 11:03:40 +1000854 return XFS_ERROR(E2BIG);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 }
856 error = xfs_read_buf(mp, mp->m_ddev_targp,
857 d - XFS_FSS_TO_BB(mp, 1),
858 XFS_FSS_TO_BB(mp, 1), 0, &bp);
859 if (!error) {
860 xfs_buf_relse(bp);
861 } else {
862 cmn_err(CE_WARN, "XFS: size check 2 failed");
Eric Sandeen0771fb42007-10-12 11:03:40 +1000863 if (error == ENOSPC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 error = XFS_ERROR(E2BIG);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000865 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 }
867
Christoph Hellwig42490232008-08-13 16:49:32 +1000868 if (mp->m_logdev_targp != mp->m_ddev_targp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
870 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
871 cmn_err(CE_WARN, "XFS: size check 3 failed");
Eric Sandeen0771fb42007-10-12 11:03:40 +1000872 return XFS_ERROR(E2BIG);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 }
874 error = xfs_read_buf(mp, mp->m_logdev_targp,
875 d - XFS_FSB_TO_BB(mp, 1),
876 XFS_FSB_TO_BB(mp, 1), 0, &bp);
877 if (!error) {
878 xfs_buf_relse(bp);
879 } else {
880 cmn_err(CE_WARN, "XFS: size check 3 failed");
Eric Sandeen0771fb42007-10-12 11:03:40 +1000881 if (error == ENOSPC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 error = XFS_ERROR(E2BIG);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000883 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 }
885 }
Eric Sandeen0771fb42007-10-12 11:03:40 +1000886 return 0;
887}
888
889/*
890 * xfs_mountfs
891 *
892 * This function does the following on an initial mount of a file system:
893 * - reads the superblock from disk and init the mount struct
894 * - if we're a 32-bit kernel, do a size check on the superblock
895 * so we don't mount terabyte filesystems
896 * - init mount struct realtime fields
897 * - allocate inode hash table for fs
898 * - init directory manager
899 * - perform recovery and init the log manager
900 */
901int
902xfs_mountfs(
Christoph Hellwig42490232008-08-13 16:49:32 +1000903 xfs_mount_t *mp)
Eric Sandeen0771fb42007-10-12 11:03:40 +1000904{
905 xfs_sb_t *sbp = &(mp->m_sb);
906 xfs_inode_t *rip;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000907 __uint64_t resblks;
908 __int64_t update_flags = 0LL;
909 uint quotamount, quotaflags;
Eric Sandeen0771fb42007-10-12 11:03:40 +1000910 int uuid_mounted = 0;
911 int error = 0;
912
Eric Sandeen0771fb42007-10-12 11:03:40 +1000913 xfs_mount_common(mp, sbp);
914
915 /*
Eric Sandeene6957ea2008-04-10 12:19:34 +1000916 * Check for a mismatched features2 values. Older kernels
917 * read & wrote into the wrong sb offset for sb_features2
918 * on some platforms due to xfs_sb_t not being 64bit size aligned
919 * when sb_features2 was added, which made older superblock
920 * reading/writing routines swap it as a 64-bit value.
David Chinneree1c0902008-03-06 13:45:50 +1100921 *
Eric Sandeene6957ea2008-04-10 12:19:34 +1000922 * For backwards compatibility, we make both slots equal.
923 *
924 * If we detect a mismatched field, we OR the set bits into the
925 * existing features2 field in case it has already been modified; we
926 * don't want to lose any features. We then update the bad location
927 * with the ORed value so that older kernels will see any features2
928 * flags, and mark the two fields as needing updates once the
929 * transaction subsystem is online.
David Chinneree1c0902008-03-06 13:45:50 +1100930 */
Eric Sandeene6957ea2008-04-10 12:19:34 +1000931 if (xfs_sb_has_mismatched_features2(sbp)) {
David Chinneree1c0902008-03-06 13:45:50 +1100932 cmn_err(CE_WARN,
933 "XFS: correcting sb_features alignment problem");
934 sbp->sb_features2 |= sbp->sb_bad_features2;
Eric Sandeene6957ea2008-04-10 12:19:34 +1000935 sbp->sb_bad_features2 = sbp->sb_features2;
David Chinneree1c0902008-03-06 13:45:50 +1100936 update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
Eric Sandeene6957ea2008-04-10 12:19:34 +1000937
938 /*
939 * Re-check for ATTR2 in case it was found in bad_features2
940 * slot.
941 */
Tim Shimmin7c12f292008-04-30 18:15:28 +1000942 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
943 !(mp->m_flags & XFS_MOUNT_NOATTR2))
Eric Sandeene6957ea2008-04-10 12:19:34 +1000944 mp->m_flags |= XFS_MOUNT_ATTR2;
Tim Shimmin7c12f292008-04-30 18:15:28 +1000945 }
Eric Sandeene6957ea2008-04-10 12:19:34 +1000946
Tim Shimmin7c12f292008-04-30 18:15:28 +1000947 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
948 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
949 xfs_sb_version_removeattr2(&mp->m_sb);
950 update_flags |= XFS_SB_FEATURES2;
951
952 /* update sb_versionnum for the clearing of the morebits */
953 if (!sbp->sb_features2)
954 update_flags |= XFS_SB_VERSIONNUM;
David Chinneree1c0902008-03-06 13:45:50 +1100955 }
956
957 /*
Eric Sandeen0771fb42007-10-12 11:03:40 +1000958 * Check if sb_agblocks is aligned at stripe boundary
959 * If sb_agblocks is NOT aligned turn off m_dalign since
960 * allocator alignment is within an ag, therefore ag has
961 * to be aligned at stripe boundary.
962 */
Christoph Hellwig42490232008-08-13 16:49:32 +1000963 error = xfs_update_alignment(mp, &update_flags);
Eric Sandeen0771fb42007-10-12 11:03:40 +1000964 if (error)
965 goto error1;
966
967 xfs_alloc_compute_maxlevels(mp);
968 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
969 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
970 xfs_ialloc_compute_maxlevels(mp);
971
972 xfs_set_maxicount(mp);
973
974 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
975
976 /*
977 * XFS uses the uuid from the superblock as the unique
978 * identifier for fsid. We can not use the uuid from the volume
979 * since a single partition filesystem is identical to a single
980 * partition volume/filesystem.
981 */
Christoph Hellwig42490232008-08-13 16:49:32 +1000982 if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0) {
Eric Sandeen0771fb42007-10-12 11:03:40 +1000983 if (xfs_uuid_mount(mp)) {
984 error = XFS_ERROR(EINVAL);
985 goto error1;
986 }
987 uuid_mounted=1;
988 }
989
990 /*
991 * Set the minimum read and write sizes
992 */
993 xfs_set_rw_sizes(mp);
994
995 /*
996 * Set the inode cluster size.
997 * This may still be overridden by the file system
998 * block size if it is larger than the chosen cluster size.
999 */
1000 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1001
1002 /*
1003 * Set inode alignment fields
1004 */
1005 xfs_set_inoalignment(mp);
1006
1007 /*
1008 * Check that the data (and log if separate) are an ok size.
1009 */
Christoph Hellwig42490232008-08-13 16:49:32 +10001010 error = xfs_check_sizes(mp);
Eric Sandeen0771fb42007-10-12 11:03:40 +10001011 if (error)
1012 goto error1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013
1014 /*
1015 * Initialize realtime fields in the mount structure
1016 */
Eric Sandeen0771fb42007-10-12 11:03:40 +10001017 error = xfs_rtmount_init(mp);
1018 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 cmn_err(CE_WARN, "XFS: RT mount failed");
1020 goto error1;
1021 }
1022
1023 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 * Copies the low order bits of the timestamp and the randomly
1025 * set "sequence" number out of a UUID.
1026 */
1027 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1028
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 mp->m_dmevmask = 0; /* not persistent; set after each mount */
1030
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001031 xfs_dir_mount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032
1033 /*
1034 * Initialize the attribute manager's entries.
1035 */
1036 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1037
1038 /*
1039 * Initialize the precomputed transaction reservations values.
1040 */
1041 xfs_trans_init(mp);
1042
1043 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 * Allocate and initialize the per-ag data.
1045 */
1046 init_rwsem(&mp->m_peraglock);
Christoph Hellwigd62c2512008-08-13 16:51:29 +10001047 mp->m_perag = kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t),
1048 KM_MAYFAIL);
1049 if (!mp->m_perag)
1050 goto error1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051
Christoph Hellwigb267ce92007-08-30 17:21:30 +10001052 mp->m_maxagi = xfs_initialize_perag(mp, sbp->sb_agcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053
1054 /*
1055 * log's mount-time initialization. Perform 1st part recovery if needed
1056 */
1057 if (likely(sbp->sb_logblocks > 0)) { /* check for volume case */
1058 error = xfs_log_mount(mp, mp->m_logdev_targp,
1059 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1060 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1061 if (error) {
1062 cmn_err(CE_WARN, "XFS: log mount failed");
1063 goto error2;
1064 }
1065 } else { /* No log has been defined */
1066 cmn_err(CE_WARN, "XFS: no log defined");
1067 XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW, mp);
1068 error = XFS_ERROR(EFSCORRUPTED);
1069 goto error2;
1070 }
1071
1072 /*
David Chinner92821e22007-05-24 15:26:31 +10001073 * Now the log is mounted, we know if it was an unclean shutdown or
1074 * not. If it was, with the first phase of recovery has completed, we
1075 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1076 * but they are recovered transactionally in the second recovery phase
1077 * later.
1078 *
1079 * Hence we can safely re-initialise incore superblock counters from
1080 * the per-ag data. These may not be correct if the filesystem was not
1081 * cleanly unmounted, so we need to wait for recovery to finish before
1082 * doing this.
1083 *
1084 * If the filesystem was cleanly unmounted, then we can trust the
1085 * values in the superblock to be correct and we don't need to do
1086 * anything here.
1087 *
1088 * If we are currently making the filesystem, the initialisation will
1089 * fail as the perag data is in an undefined state.
1090 */
1091
1092 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1093 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1094 !mp->m_sb.sb_inprogress) {
1095 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
1096 if (error) {
1097 goto error2;
1098 }
1099 }
1100 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 * Get and sanity-check the root inode.
1102 * Save the pointer to it in the mount structure.
1103 */
1104 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0);
1105 if (error) {
1106 cmn_err(CE_WARN, "XFS: failed to read root inode");
1107 goto error3;
1108 }
1109
1110 ASSERT(rip != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111
1112 if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
1113 cmn_err(CE_WARN, "XFS: corrupted root inode");
Nathan Scottb6574522006-06-09 15:29:40 +10001114 cmn_err(CE_WARN, "Device %s - root %llu is not a directory",
1115 XFS_BUFTARG_NAME(mp->m_ddev_targp),
1116 (unsigned long long)rip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1118 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1119 mp);
1120 error = XFS_ERROR(EFSCORRUPTED);
1121 goto error4;
1122 }
1123 mp->m_rootip = rip; /* save it */
1124
1125 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1126
1127 /*
1128 * Initialize realtime inode pointers in the mount structure
1129 */
Eric Sandeen0771fb42007-10-12 11:03:40 +10001130 error = xfs_rtmount_inodes(mp);
1131 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 /*
1133 * Free up the root inode.
1134 */
1135 cmn_err(CE_WARN, "XFS: failed to read RT inodes");
1136 goto error4;
1137 }
1138
1139 /*
David Chinneree1c0902008-03-06 13:45:50 +11001140 * If fs is not mounted readonly, then update the superblock changes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 */
David Chinnere5720ee2008-04-10 12:21:18 +10001142 if (update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1143 error = xfs_mount_log_sb(mp, update_flags);
1144 if (error) {
1145 cmn_err(CE_WARN, "XFS: failed to write sb changes");
1146 goto error4;
1147 }
1148 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149
1150 /*
1151 * Initialise the XFS quota management subsystem for this mount
1152 */
Eric Sandeen0771fb42007-10-12 11:03:40 +10001153 error = XFS_QM_INIT(mp, &quotamount, &quotaflags);
1154 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 goto error4;
1156
1157 /*
1158 * Finish recovering the file system. This part needed to be
1159 * delayed until after the root and real-time bitmap inodes
1160 * were consistently read in.
1161 */
Christoph Hellwig42490232008-08-13 16:49:32 +10001162 error = xfs_log_mount_finish(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 if (error) {
1164 cmn_err(CE_WARN, "XFS: log mount finish failed");
1165 goto error4;
1166 }
1167
1168 /*
1169 * Complete the quota initialisation, post-log-replay component.
1170 */
Christoph Hellwig42490232008-08-13 16:49:32 +10001171 error = XFS_QM_MOUNT(mp, quotamount, quotaflags);
Eric Sandeen0771fb42007-10-12 11:03:40 +10001172 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 goto error4;
1174
David Chinner84e1e992007-06-18 16:50:27 +10001175 /*
1176 * Now we are mounted, reserve a small amount of unused space for
1177 * privileged transactions. This is needed so that transaction
1178 * space required for critical operations can dip into this pool
1179 * when at ENOSPC. This is needed for operations like create with
1180 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1181 * are not allowed to use this reserved space.
1182 *
1183 * We default to 5% or 1024 fsbs of space reserved, whichever is smaller.
1184 * This may drive us straight to ENOSPC on mount, but that implies
David Chinner714082b2008-04-10 12:20:03 +10001185 * we were already there on the last unmount. Warn if this occurs.
David Chinner84e1e992007-06-18 16:50:27 +10001186 */
Christoph Hellwig39726be2007-06-18 17:57:45 +10001187 resblks = mp->m_sb.sb_dblocks;
1188 do_div(resblks, 20);
1189 resblks = min_t(__uint64_t, resblks, 1024);
David Chinner714082b2008-04-10 12:20:03 +10001190 error = xfs_reserve_blocks(mp, &resblks, NULL);
1191 if (error)
1192 cmn_err(CE_WARN, "XFS: Unable to allocate reserve blocks. "
1193 "Continuing without a reserve pool.");
David Chinner84e1e992007-06-18 16:50:27 +10001194
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 return 0;
1196
1197 error4:
1198 /*
1199 * Free up the root inode.
1200 */
Christoph Hellwig43355092008-03-27 18:01:08 +11001201 IRELE(rip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 error3:
1203 xfs_log_unmount_dealloc(mp);
1204 error2:
Christoph Hellwigff4f0382008-08-13 16:50:47 +10001205 xfs_free_perag(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 error1:
1207 if (uuid_mounted)
Christoph Hellwigfa6adbe2008-05-20 11:30:33 +10001208 uuid_table_remove(&mp->m_sb.sb_uuid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 return error;
1210}
1211
1212/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 * This flushes out the inodes,dquots and the superblock, unmounts the
1214 * log and makes sure that incore structures are freed.
1215 */
Christoph Hellwig41b5c2e2008-08-13 16:49:57 +10001216void
1217xfs_unmountfs(
1218 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219{
Christoph Hellwig41b5c2e2008-08-13 16:49:57 +10001220 __uint64_t resblks;
1221 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222
Christoph Hellwigf95099b2008-12-03 12:20:37 +01001223 /*
1224 * Release dquot that rootinode, rbmino and rsumino might be holding,
1225 * and release the quota inodes.
1226 */
1227 XFS_QM_UNMOUNT(mp);
1228
1229 if (mp->m_rbmip)
1230 IRELE(mp->m_rbmip);
1231 if (mp->m_rsumip)
1232 IRELE(mp->m_rsumip);
Christoph Hellwig77508ec2008-08-13 16:49:04 +10001233 IRELE(mp->m_rootip);
1234
David Chinner641c56f2007-06-18 16:50:17 +10001235 /*
1236 * We can potentially deadlock here if we have an inode cluster
1237 * that has been freed has it's buffer still pinned in memory because
1238 * the transaction is still sitting in a iclog. The stale inodes
1239 * on that buffer will have their flush locks held until the
1240 * transaction hits the disk and the callbacks run. the inode
1241 * flush takes the flush lock unconditionally and with nothing to
1242 * push out the iclog we will never get that unlocked. hence we
1243 * need to force the log first.
1244 */
1245 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
David Chinner1dc33182008-10-30 17:37:15 +11001246 xfs_reclaim_inodes(mp, 0, XFS_IFLUSH_ASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247
Nathan Scottee2a4f72006-01-11 15:33:36 +11001248 XFS_QM_DQPURGEALL(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_UMOUNTING);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249
Lachlan McIlroya357a122008-10-30 16:53:25 +11001250 if (mp->m_quotainfo)
1251 XFS_QM_DONE(mp);
1252
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 /*
1254 * Flush out the log synchronously so that we know for sure
1255 * that nothing is pinned. This is important because bflush()
1256 * will skip pinned buffers.
1257 */
1258 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1259
1260 xfs_binval(mp->m_ddev_targp);
1261 if (mp->m_rtdev_targp) {
1262 xfs_binval(mp->m_rtdev_targp);
1263 }
1264
David Chinner84e1e992007-06-18 16:50:27 +10001265 /*
1266 * Unreserve any blocks we have so that when we unmount we don't account
1267 * the reserved free space as used. This is really only necessary for
1268 * lazy superblock counting because it trusts the incore superblock
1269 * counters to be aboslutely correct on clean unmount.
1270 *
1271 * We don't bother correcting this elsewhere for lazy superblock
1272 * counting because on mount of an unclean filesystem we reconstruct the
1273 * correct counter value and this is irrelevant.
1274 *
1275 * For non-lazy counter filesystems, this doesn't matter at all because
1276 * we only every apply deltas to the superblock and hence the incore
1277 * value does not matter....
1278 */
1279 resblks = 0;
David Chinner714082b2008-04-10 12:20:03 +10001280 error = xfs_reserve_blocks(mp, &resblks, NULL);
1281 if (error)
1282 cmn_err(CE_WARN, "XFS: Unable to free reserved block pool. "
1283 "Freespace may not be correct on next mount.");
1284
David Chinnere5720ee2008-04-10 12:21:18 +10001285 error = xfs_log_sbcount(mp, 1);
1286 if (error)
1287 cmn_err(CE_WARN, "XFS: Unable to update superblock counters. "
1288 "Freespace may not be correct on next mount.");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 xfs_unmountfs_writesb(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 xfs_unmountfs_wait(mp); /* wait for async bufs */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 xfs_log_unmount(mp); /* Done! No more fs ops. */
1292
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0)
Christoph Hellwigfa6adbe2008-05-20 11:30:33 +10001294 uuid_table_remove(&mp->m_sb.sb_uuid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
Christoph Hellwig1550d0b2008-08-13 16:17:37 +10001296#if defined(DEBUG)
Christoph Hellwig0ce4cfd2007-08-30 17:20:53 +10001297 xfs_errortag_clearall(mp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298#endif
Christoph Hellwigff4f0382008-08-13 16:50:47 +10001299 xfs_free_perag(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300}
1301
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10001302STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303xfs_unmountfs_wait(xfs_mount_t *mp)
1304{
1305 if (mp->m_logdev_targp != mp->m_ddev_targp)
1306 xfs_wait_buftarg(mp->m_logdev_targp);
1307 if (mp->m_rtdev_targp)
1308 xfs_wait_buftarg(mp->m_rtdev_targp);
1309 xfs_wait_buftarg(mp->m_ddev_targp);
1310}
1311
1312int
David Chinner92821e22007-05-24 15:26:31 +10001313xfs_fs_writable(xfs_mount_t *mp)
1314{
Christoph Hellwigb267ce92007-08-30 17:21:30 +10001315 return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001316 (mp->m_flags & XFS_MOUNT_RDONLY));
David Chinner92821e22007-05-24 15:26:31 +10001317}
1318
1319/*
1320 * xfs_log_sbcount
1321 *
1322 * Called either periodically to keep the on disk superblock values
1323 * roughly up to date or from unmount to make sure the values are
1324 * correct on a clean unmount.
1325 *
1326 * Note this code can be called during the process of freezing, so
1327 * we may need to use the transaction allocator which does not not
1328 * block when the transaction subsystem is in its frozen state.
1329 */
1330int
1331xfs_log_sbcount(
1332 xfs_mount_t *mp,
1333 uint sync)
1334{
1335 xfs_trans_t *tp;
1336 int error;
1337
1338 if (!xfs_fs_writable(mp))
1339 return 0;
1340
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10001341 xfs_icsb_sync_counters(mp, 0);
David Chinner92821e22007-05-24 15:26:31 +10001342
1343 /*
1344 * we don't need to do this if we are updating the superblock
1345 * counters on every modification.
1346 */
1347 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1348 return 0;
1349
1350 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT);
1351 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1352 XFS_DEFAULT_LOG_COUNT);
1353 if (error) {
1354 xfs_trans_cancel(tp, 0);
1355 return error;
1356 }
1357
1358 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1359 if (sync)
1360 xfs_trans_set_sync(tp);
David Chinnere5720ee2008-04-10 12:21:18 +10001361 error = xfs_trans_commit(tp, 0);
1362 return error;
David Chinner92821e22007-05-24 15:26:31 +10001363}
1364
1365int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366xfs_unmountfs_writesb(xfs_mount_t *mp)
1367{
1368 xfs_buf_t *sbp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 int error = 0;
1370
1371 /*
1372 * skip superblock write if fs is read-only, or
1373 * if we are doing a forced umount.
1374 */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001375 if (!((mp->m_flags & XFS_MOUNT_RDONLY) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 XFS_FORCED_SHUTDOWN(mp))) {
David Chinner8d280b92006-03-14 13:13:09 +11001377
David Chinner92821e22007-05-24 15:26:31 +10001378 sbp = xfs_getsb(mp, 0);
David Chinner8d280b92006-03-14 13:13:09 +11001379
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 XFS_BUF_UNDONE(sbp);
1381 XFS_BUF_UNREAD(sbp);
1382 XFS_BUF_UNDELAYWRITE(sbp);
1383 XFS_BUF_WRITE(sbp);
1384 XFS_BUF_UNASYNC(sbp);
1385 ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1386 xfsbdstrat(mp, sbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 error = xfs_iowait(sbp);
1388 if (error)
1389 xfs_ioerror_alert("xfs_unmountfs_writesb",
1390 mp, sbp, XFS_BUF_ADDR(sbp));
David Chinner92821e22007-05-24 15:26:31 +10001391 xfs_buf_relse(sbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001393 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394}
1395
1396/*
1397 * xfs_mod_sb() can be used to copy arbitrary changes to the
1398 * in-core superblock into the superblock buffer to be logged.
1399 * It does not provide the higher level of locking that is
1400 * needed to protect the in-core superblock from concurrent
1401 * access.
1402 */
1403void
1404xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1405{
1406 xfs_buf_t *bp;
1407 int first;
1408 int last;
1409 xfs_mount_t *mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 xfs_sb_field_t f;
1411
1412 ASSERT(fields);
1413 if (!fields)
1414 return;
1415 mp = tp->t_mountp;
1416 bp = xfs_trans_getsb(tp, mp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 first = sizeof(xfs_sb_t);
1418 last = 0;
1419
1420 /* translate/copy */
1421
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +10001422 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
1424 /* find modified range */
1425
1426 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1427 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1428 first = xfs_sb_info[f].offset;
1429
1430 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1431 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1432 last = xfs_sb_info[f + 1].offset - 1;
1433
1434 xfs_trans_log_buf(tp, bp, first, last);
1435}
Yingping Lud210a282006-06-09 14:55:18 +10001436
Yingping Lud210a282006-06-09 14:55:18 +10001437
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438/*
1439 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1440 * a delta to a specified field in the in-core superblock. Simply
1441 * switch on the field indicated and apply the delta to that field.
1442 * Fields are not allowed to dip below zero, so if the delta would
1443 * do this do not apply it and return EINVAL.
1444 *
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001445 * The m_sb_lock must be held when this routine is called.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 */
David Chinner8d280b92006-03-14 13:13:09 +11001447int
David Chinner20f4ebf2007-02-10 18:36:10 +11001448xfs_mod_incore_sb_unlocked(
1449 xfs_mount_t *mp,
1450 xfs_sb_field_t field,
1451 int64_t delta,
1452 int rsvd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453{
1454 int scounter; /* short counter for 32 bit fields */
1455 long long lcounter; /* long counter for 64 bit fields */
1456 long long res_used, rem;
1457
1458 /*
1459 * With the in-core superblock spin lock held, switch
1460 * on the indicated field. Apply the delta to the
1461 * proper field. If the fields value would dip below
1462 * 0, then do not apply the delta and return EINVAL.
1463 */
1464 switch (field) {
1465 case XFS_SBS_ICOUNT:
1466 lcounter = (long long)mp->m_sb.sb_icount;
1467 lcounter += delta;
1468 if (lcounter < 0) {
1469 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001470 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 }
1472 mp->m_sb.sb_icount = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001473 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 case XFS_SBS_IFREE:
1475 lcounter = (long long)mp->m_sb.sb_ifree;
1476 lcounter += delta;
1477 if (lcounter < 0) {
1478 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001479 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 }
1481 mp->m_sb.sb_ifree = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001482 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 case XFS_SBS_FDBLOCKS:
David Chinner4be536d2006-09-07 14:26:50 +10001484 lcounter = (long long)
1485 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1487
1488 if (delta > 0) { /* Putting blocks back */
1489 if (res_used > delta) {
1490 mp->m_resblks_avail += delta;
1491 } else {
1492 rem = delta - res_used;
1493 mp->m_resblks_avail = mp->m_resblks;
1494 lcounter += rem;
1495 }
1496 } else { /* Taking blocks away */
1497
1498 lcounter += delta;
1499
1500 /*
1501 * If were out of blocks, use any available reserved blocks if
1502 * were allowed to.
1503 */
1504
1505 if (lcounter < 0) {
1506 if (rsvd) {
1507 lcounter = (long long)mp->m_resblks_avail + delta;
1508 if (lcounter < 0) {
Jesper Juhl014c2542006-01-15 02:37:08 +01001509 return XFS_ERROR(ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 }
1511 mp->m_resblks_avail = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001512 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 } else { /* not reserved */
Jesper Juhl014c2542006-01-15 02:37:08 +01001514 return XFS_ERROR(ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 }
1516 }
1517 }
1518
David Chinner4be536d2006-09-07 14:26:50 +10001519 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
Jesper Juhl014c2542006-01-15 02:37:08 +01001520 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 case XFS_SBS_FREXTENTS:
1522 lcounter = (long long)mp->m_sb.sb_frextents;
1523 lcounter += delta;
1524 if (lcounter < 0) {
Jesper Juhl014c2542006-01-15 02:37:08 +01001525 return XFS_ERROR(ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 }
1527 mp->m_sb.sb_frextents = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001528 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 case XFS_SBS_DBLOCKS:
1530 lcounter = (long long)mp->m_sb.sb_dblocks;
1531 lcounter += delta;
1532 if (lcounter < 0) {
1533 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001534 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 }
1536 mp->m_sb.sb_dblocks = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001537 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 case XFS_SBS_AGCOUNT:
1539 scounter = mp->m_sb.sb_agcount;
1540 scounter += delta;
1541 if (scounter < 0) {
1542 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001543 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 }
1545 mp->m_sb.sb_agcount = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001546 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 case XFS_SBS_IMAX_PCT:
1548 scounter = mp->m_sb.sb_imax_pct;
1549 scounter += delta;
1550 if (scounter < 0) {
1551 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001552 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 }
1554 mp->m_sb.sb_imax_pct = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001555 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 case XFS_SBS_REXTSIZE:
1557 scounter = mp->m_sb.sb_rextsize;
1558 scounter += delta;
1559 if (scounter < 0) {
1560 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001561 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 }
1563 mp->m_sb.sb_rextsize = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001564 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 case XFS_SBS_RBMBLOCKS:
1566 scounter = mp->m_sb.sb_rbmblocks;
1567 scounter += delta;
1568 if (scounter < 0) {
1569 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001570 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 }
1572 mp->m_sb.sb_rbmblocks = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001573 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 case XFS_SBS_RBLOCKS:
1575 lcounter = (long long)mp->m_sb.sb_rblocks;
1576 lcounter += delta;
1577 if (lcounter < 0) {
1578 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001579 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 }
1581 mp->m_sb.sb_rblocks = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001582 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 case XFS_SBS_REXTENTS:
1584 lcounter = (long long)mp->m_sb.sb_rextents;
1585 lcounter += delta;
1586 if (lcounter < 0) {
1587 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001588 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 }
1590 mp->m_sb.sb_rextents = lcounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001591 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 case XFS_SBS_REXTSLOG:
1593 scounter = mp->m_sb.sb_rextslog;
1594 scounter += delta;
1595 if (scounter < 0) {
1596 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001597 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 }
1599 mp->m_sb.sb_rextslog = scounter;
Jesper Juhl014c2542006-01-15 02:37:08 +01001600 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 default:
1602 ASSERT(0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001603 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 }
1605}
1606
1607/*
1608 * xfs_mod_incore_sb() is used to change a field in the in-core
1609 * superblock structure by the specified delta. This modification
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001610 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 * routine to do the work.
1612 */
1613int
David Chinner20f4ebf2007-02-10 18:36:10 +11001614xfs_mod_incore_sb(
1615 xfs_mount_t *mp,
1616 xfs_sb_field_t field,
1617 int64_t delta,
1618 int rsvd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 int status;
1621
David Chinner8d280b92006-03-14 13:13:09 +11001622 /* check for per-cpu counters */
1623 switch (field) {
1624#ifdef HAVE_PERCPU_SB
1625 case XFS_SBS_ICOUNT:
1626 case XFS_SBS_IFREE:
1627 case XFS_SBS_FDBLOCKS:
1628 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1629 status = xfs_icsb_modify_counters(mp, field,
1630 delta, rsvd);
1631 break;
1632 }
1633 /* FALLTHROUGH */
1634#endif
1635 default:
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001636 spin_lock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001637 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001638 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001639 break;
1640 }
1641
Jesper Juhl014c2542006-01-15 02:37:08 +01001642 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643}
1644
1645/*
1646 * xfs_mod_incore_sb_batch() is used to change more than one field
1647 * in the in-core superblock structure at a time. This modification
1648 * is protected by a lock internal to this module. The fields and
1649 * changes to those fields are specified in the array of xfs_mod_sb
1650 * structures passed in.
1651 *
1652 * Either all of the specified deltas will be applied or none of
1653 * them will. If any modified field dips below 0, then all modifications
1654 * will be backed out and EINVAL will be returned.
1655 */
1656int
1657xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1658{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 int status=0;
1660 xfs_mod_sb_t *msbp;
1661
1662 /*
1663 * Loop through the array of mod structures and apply each
1664 * individually. If any fail, then back out all those
1665 * which have already been applied. Do all of this within
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001666 * the scope of the m_sb_lock so that all of the changes will
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 * be atomic.
1668 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001669 spin_lock(&mp->m_sb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 msbp = &msb[0];
1671 for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1672 /*
1673 * Apply the delta at index n. If it fails, break
1674 * from the loop so we'll fall into the undo loop
1675 * below.
1676 */
David Chinner8d280b92006-03-14 13:13:09 +11001677 switch (msbp->msb_field) {
1678#ifdef HAVE_PERCPU_SB
1679 case XFS_SBS_ICOUNT:
1680 case XFS_SBS_IFREE:
1681 case XFS_SBS_FDBLOCKS:
1682 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001683 spin_unlock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11001684 status = xfs_icsb_modify_counters(mp,
David Chinner8d280b92006-03-14 13:13:09 +11001685 msbp->msb_field,
1686 msbp->msb_delta, rsvd);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001687 spin_lock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001688 break;
1689 }
1690 /* FALLTHROUGH */
1691#endif
1692 default:
1693 status = xfs_mod_incore_sb_unlocked(mp,
1694 msbp->msb_field,
1695 msbp->msb_delta, rsvd);
1696 break;
1697 }
1698
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 if (status != 0) {
1700 break;
1701 }
1702 }
1703
1704 /*
1705 * If we didn't complete the loop above, then back out
1706 * any changes made to the superblock. If you add code
1707 * between the loop above and here, make sure that you
1708 * preserve the value of status. Loop back until
1709 * we step below the beginning of the array. Make sure
1710 * we don't touch anything back there.
1711 */
1712 if (status != 0) {
1713 msbp--;
1714 while (msbp >= msb) {
David Chinner8d280b92006-03-14 13:13:09 +11001715 switch (msbp->msb_field) {
1716#ifdef HAVE_PERCPU_SB
1717 case XFS_SBS_ICOUNT:
1718 case XFS_SBS_IFREE:
1719 case XFS_SBS_FDBLOCKS:
1720 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001721 spin_unlock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11001722 status = xfs_icsb_modify_counters(mp,
David Chinner8d280b92006-03-14 13:13:09 +11001723 msbp->msb_field,
1724 -(msbp->msb_delta),
1725 rsvd);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001726 spin_lock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11001727 break;
1728 }
1729 /* FALLTHROUGH */
1730#endif
1731 default:
1732 status = xfs_mod_incore_sb_unlocked(mp,
1733 msbp->msb_field,
1734 -(msbp->msb_delta),
1735 rsvd);
1736 break;
1737 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 ASSERT(status == 0);
1739 msbp--;
1740 }
1741 }
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001742 spin_unlock(&mp->m_sb_lock);
Jesper Juhl014c2542006-01-15 02:37:08 +01001743 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744}
1745
1746/*
1747 * xfs_getsb() is called to obtain the buffer for the superblock.
1748 * The buffer is returned locked and read in from disk.
1749 * The buffer should be released with a call to xfs_brelse().
1750 *
1751 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1752 * the superblock buffer if it can be locked without sleeping.
1753 * If it can't then we'll return NULL.
1754 */
1755xfs_buf_t *
1756xfs_getsb(
1757 xfs_mount_t *mp,
1758 int flags)
1759{
1760 xfs_buf_t *bp;
1761
1762 ASSERT(mp->m_sb_bp != NULL);
1763 bp = mp->m_sb_bp;
1764 if (flags & XFS_BUF_TRYLOCK) {
1765 if (!XFS_BUF_CPSEMA(bp)) {
1766 return NULL;
1767 }
1768 } else {
1769 XFS_BUF_PSEMA(bp, PRIBIO);
1770 }
1771 XFS_BUF_HOLD(bp);
1772 ASSERT(XFS_BUF_ISDONE(bp));
Jesper Juhl014c2542006-01-15 02:37:08 +01001773 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774}
1775
1776/*
1777 * Used to free the superblock along various error paths.
1778 */
1779void
1780xfs_freesb(
1781 xfs_mount_t *mp)
1782{
1783 xfs_buf_t *bp;
1784
1785 /*
1786 * Use xfs_getsb() so that the buffer will be locked
1787 * when we call xfs_buf_relse().
1788 */
1789 bp = xfs_getsb(mp, 0);
1790 XFS_BUF_UNMANAGE(bp);
1791 xfs_buf_relse(bp);
1792 mp->m_sb_bp = NULL;
1793}
1794
1795/*
1796 * See if the UUID is unique among mounted XFS filesystems.
1797 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1798 */
1799STATIC int
1800xfs_uuid_mount(
1801 xfs_mount_t *mp)
1802{
1803 if (uuid_is_nil(&mp->m_sb.sb_uuid)) {
1804 cmn_err(CE_WARN,
1805 "XFS: Filesystem %s has nil UUID - can't mount",
1806 mp->m_fsname);
1807 return -1;
1808 }
1809 if (!uuid_table_insert(&mp->m_sb.sb_uuid)) {
1810 cmn_err(CE_WARN,
1811 "XFS: Filesystem %s has duplicate UUID - can't mount",
1812 mp->m_fsname);
1813 return -1;
1814 }
1815 return 0;
1816}
1817
1818/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 * Used to log changes to the superblock unit and width fields which could
Eric Sandeene6957ea2008-04-10 12:19:34 +10001820 * be altered by the mount options, as well as any potential sb_features2
1821 * fixup. Only the first superblock is updated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 */
David Chinnere5720ee2008-04-10 12:21:18 +10001823STATIC int
David Chinneree1c0902008-03-06 13:45:50 +11001824xfs_mount_log_sb(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 xfs_mount_t *mp,
1826 __int64_t fields)
1827{
1828 xfs_trans_t *tp;
David Chinnere5720ee2008-04-10 12:21:18 +10001829 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830
David Chinneree1c0902008-03-06 13:45:50 +11001831 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
David Chinner4b166de2008-05-20 11:30:27 +10001832 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1833 XFS_SB_VERSIONNUM));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834
1835 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
David Chinnere5720ee2008-04-10 12:21:18 +10001836 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1837 XFS_DEFAULT_LOG_COUNT);
1838 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 xfs_trans_cancel(tp, 0);
David Chinnere5720ee2008-04-10 12:21:18 +10001840 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 }
1842 xfs_mod_sb(tp, fields);
David Chinnere5720ee2008-04-10 12:21:18 +10001843 error = xfs_trans_commit(tp, 0);
1844 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845}
David Chinner8d280b92006-03-14 13:13:09 +11001846
1847
1848#ifdef HAVE_PERCPU_SB
1849/*
1850 * Per-cpu incore superblock counters
1851 *
1852 * Simple concept, difficult implementation
1853 *
1854 * Basically, replace the incore superblock counters with a distributed per cpu
1855 * counter for contended fields (e.g. free block count).
1856 *
1857 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1858 * hence needs to be accurately read when we are running low on space. Hence
1859 * there is a method to enable and disable the per-cpu counters based on how
1860 * much "stuff" is available in them.
1861 *
1862 * Basically, a counter is enabled if there is enough free resource to justify
1863 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1864 * ENOSPC), then we disable the counters to synchronise all callers and
1865 * re-distribute the available resources.
1866 *
1867 * If, once we redistributed the available resources, we still get a failure,
1868 * we disable the per-cpu counter and go through the slow path.
1869 *
1870 * The slow path is the current xfs_mod_incore_sb() function. This means that
1871 * when we disable a per-cpu counter, we need to drain it's resources back to
1872 * the global superblock. We do this after disabling the counter to prevent
1873 * more threads from queueing up on the counter.
1874 *
1875 * Essentially, this means that we still need a lock in the fast path to enable
1876 * synchronisation between the global counters and the per-cpu counters. This
1877 * is not a problem because the lock will be local to a CPU almost all the time
1878 * and have little contention except when we get to ENOSPC conditions.
1879 *
1880 * Basically, this lock becomes a barrier that enables us to lock out the fast
1881 * path while we do things like enabling and disabling counters and
1882 * synchronising the counters.
1883 *
1884 * Locking rules:
1885 *
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001886 * 1. m_sb_lock before picking up per-cpu locks
David Chinner8d280b92006-03-14 13:13:09 +11001887 * 2. per-cpu locks always picked up via for_each_online_cpu() order
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001888 * 3. accurate counter sync requires m_sb_lock + per cpu locks
David Chinner8d280b92006-03-14 13:13:09 +11001889 * 4. modifying per-cpu counters requires holding per-cpu lock
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001890 * 5. modifying global counters requires holding m_sb_lock
1891 * 6. enabling or disabling a counter requires holding the m_sb_lock
David Chinner8d280b92006-03-14 13:13:09 +11001892 * and _none_ of the per-cpu locks.
1893 *
1894 * Disabled counters are only ever re-enabled by a balance operation
1895 * that results in more free resources per CPU than a given threshold.
1896 * To ensure counters don't remain disabled, they are rebalanced when
1897 * the global resource goes above a higher threshold (i.e. some hysteresis
1898 * is present to prevent thrashing).
David Chinner8d280b92006-03-14 13:13:09 +11001899 */
David Chinnere8234a62006-03-14 13:23:52 +11001900
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001901#ifdef CONFIG_HOTPLUG_CPU
David Chinnere8234a62006-03-14 13:23:52 +11001902/*
1903 * hot-plug CPU notifier support.
1904 *
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001905 * We need a notifier per filesystem as we need to be able to identify
1906 * the filesystem to balance the counters out. This is achieved by
1907 * having a notifier block embedded in the xfs_mount_t and doing pointer
1908 * magic to get the mount pointer from the notifier block address.
David Chinnere8234a62006-03-14 13:23:52 +11001909 */
1910STATIC int
1911xfs_icsb_cpu_notify(
1912 struct notifier_block *nfb,
1913 unsigned long action,
1914 void *hcpu)
1915{
1916 xfs_icsb_cnts_t *cntp;
1917 xfs_mount_t *mp;
David Chinnere8234a62006-03-14 13:23:52 +11001918
1919 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
1920 cntp = (xfs_icsb_cnts_t *)
1921 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
1922 switch (action) {
1923 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001924 case CPU_UP_PREPARE_FROZEN:
David Chinnere8234a62006-03-14 13:23:52 +11001925 /* Easy Case - initialize the area and locks, and
1926 * then rebalance when online does everything else for us. */
David Chinner01e1b692006-03-14 13:29:16 +11001927 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinnere8234a62006-03-14 13:23:52 +11001928 break;
1929 case CPU_ONLINE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001930 case CPU_ONLINE_FROZEN:
David Chinner03135cf2007-02-10 18:35:15 +11001931 xfs_icsb_lock(mp);
Christoph Hellwig45af6c62008-04-22 17:34:44 +10001932 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
1933 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
1934 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
David Chinner03135cf2007-02-10 18:35:15 +11001935 xfs_icsb_unlock(mp);
David Chinnere8234a62006-03-14 13:23:52 +11001936 break;
1937 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001938 case CPU_DEAD_FROZEN:
David Chinnere8234a62006-03-14 13:23:52 +11001939 /* Disable all the counters, then fold the dead cpu's
1940 * count into the total on the global superblock and
1941 * re-enable the counters. */
David Chinner03135cf2007-02-10 18:35:15 +11001942 xfs_icsb_lock(mp);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001943 spin_lock(&mp->m_sb_lock);
David Chinnere8234a62006-03-14 13:23:52 +11001944 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
1945 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
1946 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
1947
1948 mp->m_sb.sb_icount += cntp->icsb_icount;
1949 mp->m_sb.sb_ifree += cntp->icsb_ifree;
1950 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
1951
David Chinner01e1b692006-03-14 13:29:16 +11001952 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinnere8234a62006-03-14 13:23:52 +11001953
Christoph Hellwig45af6c62008-04-22 17:34:44 +10001954 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
1955 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
1956 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10001957 spin_unlock(&mp->m_sb_lock);
David Chinner03135cf2007-02-10 18:35:15 +11001958 xfs_icsb_unlock(mp);
David Chinnere8234a62006-03-14 13:23:52 +11001959 break;
1960 }
1961
1962 return NOTIFY_OK;
1963}
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001964#endif /* CONFIG_HOTPLUG_CPU */
David Chinnere8234a62006-03-14 13:23:52 +11001965
David Chinner8d280b92006-03-14 13:13:09 +11001966int
1967xfs_icsb_init_counters(
1968 xfs_mount_t *mp)
1969{
1970 xfs_icsb_cnts_t *cntp;
1971 int i;
1972
1973 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
1974 if (mp->m_sb_cnts == NULL)
1975 return -ENOMEM;
1976
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001977#ifdef CONFIG_HOTPLUG_CPU
David Chinnere8234a62006-03-14 13:23:52 +11001978 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
1979 mp->m_icsb_notifier.priority = 0;
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07001980 register_hotcpu_notifier(&mp->m_icsb_notifier);
1981#endif /* CONFIG_HOTPLUG_CPU */
David Chinnere8234a62006-03-14 13:23:52 +11001982
David Chinner8d280b92006-03-14 13:13:09 +11001983 for_each_online_cpu(i) {
1984 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11001985 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
David Chinner8d280b92006-03-14 13:13:09 +11001986 }
David Chinner20b64282007-02-10 18:35:09 +11001987
1988 mutex_init(&mp->m_icsb_mutex);
1989
David Chinner8d280b92006-03-14 13:13:09 +11001990 /*
1991 * start with all counters disabled so that the
1992 * initial balance kicks us off correctly
1993 */
1994 mp->m_icsb_counters = -1;
1995 return 0;
1996}
1997
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11001998void
1999xfs_icsb_reinit_counters(
2000 xfs_mount_t *mp)
2001{
2002 xfs_icsb_lock(mp);
2003 /*
2004 * start with all counters disabled so that the
2005 * initial balance kicks us off correctly
2006 */
2007 mp->m_icsb_counters = -1;
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002008 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2009 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2010 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11002011 xfs_icsb_unlock(mp);
2012}
2013
Christoph Hellwigc962fb72008-05-20 15:10:52 +10002014void
David Chinner8d280b92006-03-14 13:13:09 +11002015xfs_icsb_destroy_counters(
2016 xfs_mount_t *mp)
2017{
David Chinnere8234a62006-03-14 13:23:52 +11002018 if (mp->m_sb_cnts) {
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07002019 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
David Chinner8d280b92006-03-14 13:13:09 +11002020 free_percpu(mp->m_sb_cnts);
David Chinnere8234a62006-03-14 13:23:52 +11002021 }
David Chinner03135cf2007-02-10 18:35:15 +11002022 mutex_destroy(&mp->m_icsb_mutex);
David Chinner8d280b92006-03-14 13:13:09 +11002023}
2024
David Chinner7989cb82007-02-10 18:34:56 +11002025STATIC_INLINE void
David Chinner01e1b692006-03-14 13:29:16 +11002026xfs_icsb_lock_cntr(
2027 xfs_icsb_cnts_t *icsbp)
2028{
2029 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2030 ndelay(1000);
2031 }
2032}
2033
David Chinner7989cb82007-02-10 18:34:56 +11002034STATIC_INLINE void
David Chinner01e1b692006-03-14 13:29:16 +11002035xfs_icsb_unlock_cntr(
2036 xfs_icsb_cnts_t *icsbp)
2037{
2038 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2039}
2040
David Chinner8d280b92006-03-14 13:13:09 +11002041
David Chinner7989cb82007-02-10 18:34:56 +11002042STATIC_INLINE void
David Chinner8d280b92006-03-14 13:13:09 +11002043xfs_icsb_lock_all_counters(
2044 xfs_mount_t *mp)
2045{
2046 xfs_icsb_cnts_t *cntp;
2047 int i;
2048
2049 for_each_online_cpu(i) {
2050 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11002051 xfs_icsb_lock_cntr(cntp);
David Chinner8d280b92006-03-14 13:13:09 +11002052 }
2053}
2054
David Chinner7989cb82007-02-10 18:34:56 +11002055STATIC_INLINE void
David Chinner8d280b92006-03-14 13:13:09 +11002056xfs_icsb_unlock_all_counters(
2057 xfs_mount_t *mp)
2058{
2059 xfs_icsb_cnts_t *cntp;
2060 int i;
2061
2062 for_each_online_cpu(i) {
2063 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
David Chinner01e1b692006-03-14 13:29:16 +11002064 xfs_icsb_unlock_cntr(cntp);
David Chinner8d280b92006-03-14 13:13:09 +11002065 }
2066}
2067
2068STATIC void
2069xfs_icsb_count(
2070 xfs_mount_t *mp,
2071 xfs_icsb_cnts_t *cnt,
2072 int flags)
2073{
2074 xfs_icsb_cnts_t *cntp;
2075 int i;
2076
2077 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2078
2079 if (!(flags & XFS_ICSB_LAZY_COUNT))
2080 xfs_icsb_lock_all_counters(mp);
2081
2082 for_each_online_cpu(i) {
2083 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2084 cnt->icsb_icount += cntp->icsb_icount;
2085 cnt->icsb_ifree += cntp->icsb_ifree;
2086 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2087 }
2088
2089 if (!(flags & XFS_ICSB_LAZY_COUNT))
2090 xfs_icsb_unlock_all_counters(mp);
2091}
2092
2093STATIC int
2094xfs_icsb_counter_disabled(
2095 xfs_mount_t *mp,
2096 xfs_sb_field_t field)
2097{
2098 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2099 return test_bit(field, &mp->m_icsb_counters);
2100}
2101
David Chinner36fbe6e2008-04-10 12:19:56 +10002102STATIC void
David Chinner8d280b92006-03-14 13:13:09 +11002103xfs_icsb_disable_counter(
2104 xfs_mount_t *mp,
2105 xfs_sb_field_t field)
2106{
2107 xfs_icsb_cnts_t cnt;
2108
2109 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2110
David Chinner20b64282007-02-10 18:35:09 +11002111 /*
2112 * If we are already disabled, then there is nothing to do
2113 * here. We check before locking all the counters to avoid
2114 * the expensive lock operation when being called in the
2115 * slow path and the counter is already disabled. This is
2116 * safe because the only time we set or clear this state is under
2117 * the m_icsb_mutex.
2118 */
2119 if (xfs_icsb_counter_disabled(mp, field))
David Chinner36fbe6e2008-04-10 12:19:56 +10002120 return;
David Chinner20b64282007-02-10 18:35:09 +11002121
David Chinner8d280b92006-03-14 13:13:09 +11002122 xfs_icsb_lock_all_counters(mp);
2123 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2124 /* drain back to superblock */
2125
Christoph Hellwigce461932008-04-22 17:34:50 +10002126 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
David Chinner8d280b92006-03-14 13:13:09 +11002127 switch(field) {
2128 case XFS_SBS_ICOUNT:
2129 mp->m_sb.sb_icount = cnt.icsb_icount;
2130 break;
2131 case XFS_SBS_IFREE:
2132 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2133 break;
2134 case XFS_SBS_FDBLOCKS:
2135 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2136 break;
2137 default:
2138 BUG();
2139 }
2140 }
2141
2142 xfs_icsb_unlock_all_counters(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002143}
2144
2145STATIC void
2146xfs_icsb_enable_counter(
2147 xfs_mount_t *mp,
2148 xfs_sb_field_t field,
2149 uint64_t count,
2150 uint64_t resid)
2151{
2152 xfs_icsb_cnts_t *cntp;
2153 int i;
2154
2155 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2156
2157 xfs_icsb_lock_all_counters(mp);
2158 for_each_online_cpu(i) {
2159 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2160 switch (field) {
2161 case XFS_SBS_ICOUNT:
2162 cntp->icsb_icount = count + resid;
2163 break;
2164 case XFS_SBS_IFREE:
2165 cntp->icsb_ifree = count + resid;
2166 break;
2167 case XFS_SBS_FDBLOCKS:
2168 cntp->icsb_fdblocks = count + resid;
2169 break;
2170 default:
2171 BUG();
2172 break;
2173 }
2174 resid = 0;
2175 }
2176 clear_bit(field, &mp->m_icsb_counters);
2177 xfs_icsb_unlock_all_counters(mp);
2178}
2179
David Chinnerdbcabad2007-02-10 18:36:17 +11002180void
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002181xfs_icsb_sync_counters_locked(
David Chinner8d280b92006-03-14 13:13:09 +11002182 xfs_mount_t *mp,
2183 int flags)
2184{
2185 xfs_icsb_cnts_t cnt;
David Chinner8d280b92006-03-14 13:13:09 +11002186
David Chinner8d280b92006-03-14 13:13:09 +11002187 xfs_icsb_count(mp, &cnt, flags);
2188
David Chinner8d280b92006-03-14 13:13:09 +11002189 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2190 mp->m_sb.sb_icount = cnt.icsb_icount;
2191 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2192 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2193 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2194 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
David Chinner8d280b92006-03-14 13:13:09 +11002195}
2196
2197/*
2198 * Accurate update of per-cpu counters to incore superblock
2199 */
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002200void
David Chinner8d280b92006-03-14 13:13:09 +11002201xfs_icsb_sync_counters(
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002202 xfs_mount_t *mp,
2203 int flags)
David Chinner8d280b92006-03-14 13:13:09 +11002204{
Christoph Hellwigd4d90b52008-04-22 17:34:37 +10002205 spin_lock(&mp->m_sb_lock);
2206 xfs_icsb_sync_counters_locked(mp, flags);
2207 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11002208}
2209
2210/*
2211 * Balance and enable/disable counters as necessary.
2212 *
David Chinner20b64282007-02-10 18:35:09 +11002213 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2214 * chosen to be the same number as single on disk allocation chunk per CPU, and
2215 * free blocks is something far enough zero that we aren't going thrash when we
2216 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2217 * prevent looping endlessly when xfs_alloc_space asks for more than will
2218 * be distributed to a single CPU but each CPU has enough blocks to be
2219 * reenabled.
2220 *
2221 * Note that we can be called when counters are already disabled.
2222 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2223 * prevent locking every per-cpu counter needlessly.
David Chinner8d280b92006-03-14 13:13:09 +11002224 */
David Chinner20b64282007-02-10 18:35:09 +11002225
2226#define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
David Chinner4be536d2006-09-07 14:26:50 +10002227#define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
David Chinner20b64282007-02-10 18:35:09 +11002228 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
David Chinner8d280b92006-03-14 13:13:09 +11002229STATIC void
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002230xfs_icsb_balance_counter_locked(
David Chinner8d280b92006-03-14 13:13:09 +11002231 xfs_mount_t *mp,
2232 xfs_sb_field_t field,
David Chinner20b64282007-02-10 18:35:09 +11002233 int min_per_cpu)
David Chinner8d280b92006-03-14 13:13:09 +11002234{
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002235 uint64_t count, resid;
David Chinner8d280b92006-03-14 13:13:09 +11002236 int weight = num_online_cpus();
David Chinner20b64282007-02-10 18:35:09 +11002237 uint64_t min = (uint64_t)min_per_cpu;
David Chinner8d280b92006-03-14 13:13:09 +11002238
David Chinner8d280b92006-03-14 13:13:09 +11002239 /* disable counter and sync counter */
2240 xfs_icsb_disable_counter(mp, field);
2241
2242 /* update counters - first CPU gets residual*/
2243 switch (field) {
2244 case XFS_SBS_ICOUNT:
2245 count = mp->m_sb.sb_icount;
2246 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002247 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002248 return;
David Chinner8d280b92006-03-14 13:13:09 +11002249 break;
2250 case XFS_SBS_IFREE:
2251 count = mp->m_sb.sb_ifree;
2252 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002253 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002254 return;
David Chinner8d280b92006-03-14 13:13:09 +11002255 break;
2256 case XFS_SBS_FDBLOCKS:
2257 count = mp->m_sb.sb_fdblocks;
2258 resid = do_div(count, weight);
David Chinner20b64282007-02-10 18:35:09 +11002259 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002260 return;
David Chinner8d280b92006-03-14 13:13:09 +11002261 break;
2262 default:
2263 BUG();
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002264 count = resid = 0; /* quiet, gcc */
David Chinner8d280b92006-03-14 13:13:09 +11002265 break;
2266 }
2267
2268 xfs_icsb_enable_counter(mp, field, count, resid);
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002269}
2270
2271STATIC void
2272xfs_icsb_balance_counter(
2273 xfs_mount_t *mp,
2274 xfs_sb_field_t fields,
2275 int min_per_cpu)
2276{
2277 spin_lock(&mp->m_sb_lock);
2278 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2279 spin_unlock(&mp->m_sb_lock);
David Chinner8d280b92006-03-14 13:13:09 +11002280}
2281
David Chinnera8272ce2007-11-23 16:28:09 +11002282STATIC int
David Chinner20b64282007-02-10 18:35:09 +11002283xfs_icsb_modify_counters(
David Chinner8d280b92006-03-14 13:13:09 +11002284 xfs_mount_t *mp,
2285 xfs_sb_field_t field,
David Chinner20f4ebf2007-02-10 18:36:10 +11002286 int64_t delta,
David Chinner20b64282007-02-10 18:35:09 +11002287 int rsvd)
David Chinner8d280b92006-03-14 13:13:09 +11002288{
2289 xfs_icsb_cnts_t *icsbp;
2290 long long lcounter; /* long counter for 64 bit fields */
Eric Sandeen007c61c2007-10-11 17:43:56 +10002291 int cpu, ret = 0;
David Chinner8d280b92006-03-14 13:13:09 +11002292
David Chinner20b64282007-02-10 18:35:09 +11002293 might_sleep();
David Chinner8d280b92006-03-14 13:13:09 +11002294again:
2295 cpu = get_cpu();
David Chinner20b64282007-02-10 18:35:09 +11002296 icsbp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, cpu);
2297
2298 /*
2299 * if the counter is disabled, go to slow path
2300 */
David Chinner8d280b92006-03-14 13:13:09 +11002301 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2302 goto slow_path;
David Chinner20b64282007-02-10 18:35:09 +11002303 xfs_icsb_lock_cntr(icsbp);
2304 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2305 xfs_icsb_unlock_cntr(icsbp);
2306 goto slow_path;
2307 }
David Chinner8d280b92006-03-14 13:13:09 +11002308
2309 switch (field) {
2310 case XFS_SBS_ICOUNT:
2311 lcounter = icsbp->icsb_icount;
2312 lcounter += delta;
2313 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002314 goto balance_counter;
David Chinner8d280b92006-03-14 13:13:09 +11002315 icsbp->icsb_icount = lcounter;
2316 break;
2317
2318 case XFS_SBS_IFREE:
2319 lcounter = icsbp->icsb_ifree;
2320 lcounter += delta;
2321 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002322 goto balance_counter;
David Chinner8d280b92006-03-14 13:13:09 +11002323 icsbp->icsb_ifree = lcounter;
2324 break;
2325
2326 case XFS_SBS_FDBLOCKS:
2327 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2328
David Chinner4be536d2006-09-07 14:26:50 +10002329 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002330 lcounter += delta;
2331 if (unlikely(lcounter < 0))
David Chinner20b64282007-02-10 18:35:09 +11002332 goto balance_counter;
David Chinner4be536d2006-09-07 14:26:50 +10002333 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002334 break;
2335 default:
2336 BUG();
2337 break;
2338 }
David Chinner01e1b692006-03-14 13:29:16 +11002339 xfs_icsb_unlock_cntr(icsbp);
David Chinner8d280b92006-03-14 13:13:09 +11002340 put_cpu();
David Chinner8d280b92006-03-14 13:13:09 +11002341 return 0;
2342
David Chinner8d280b92006-03-14 13:13:09 +11002343slow_path:
David Chinner20b64282007-02-10 18:35:09 +11002344 put_cpu();
2345
2346 /*
2347 * serialise with a mutex so we don't burn lots of cpu on
2348 * the superblock lock. We still need to hold the superblock
2349 * lock, however, when we modify the global structures.
2350 */
David Chinner03135cf2007-02-10 18:35:15 +11002351 xfs_icsb_lock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002352
2353 /*
2354 * Now running atomically.
2355 *
2356 * If the counter is enabled, someone has beaten us to rebalancing.
2357 * Drop the lock and try again in the fast path....
2358 */
2359 if (!(xfs_icsb_counter_disabled(mp, field))) {
David Chinner03135cf2007-02-10 18:35:15 +11002360 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002361 goto again;
2362 }
2363
2364 /*
2365 * The counter is currently disabled. Because we are
2366 * running atomically here, we know a rebalance cannot
2367 * be in progress. Hence we can go straight to operating
2368 * on the global superblock. We do not call xfs_mod_incore_sb()
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002369 * here even though we need to get the m_sb_lock. Doing so
David Chinner20b64282007-02-10 18:35:09 +11002370 * will cause us to re-enter this function and deadlock.
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002371 * Hence we get the m_sb_lock ourselves and then call
David Chinner20b64282007-02-10 18:35:09 +11002372 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2373 * directly on the global counters.
2374 */
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002375 spin_lock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11002376 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
Eric Sandeen3685c2a2007-10-11 17:42:32 +10002377 spin_unlock(&mp->m_sb_lock);
David Chinner20b64282007-02-10 18:35:09 +11002378
2379 /*
2380 * Now that we've modified the global superblock, we
2381 * may be able to re-enable the distributed counters
2382 * (e.g. lots of space just got freed). After that
2383 * we are done.
2384 */
2385 if (ret != ENOSPC)
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002386 xfs_icsb_balance_counter(mp, field, 0);
David Chinner03135cf2007-02-10 18:35:15 +11002387 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002388 return ret;
2389
2390balance_counter:
David Chinner01e1b692006-03-14 13:29:16 +11002391 xfs_icsb_unlock_cntr(icsbp);
David Chinner8d280b92006-03-14 13:13:09 +11002392 put_cpu();
2393
David Chinner20b64282007-02-10 18:35:09 +11002394 /*
2395 * We may have multiple threads here if multiple per-cpu
2396 * counters run dry at the same time. This will mean we can
2397 * do more balances than strictly necessary but it is not
2398 * the common slowpath case.
2399 */
David Chinner03135cf2007-02-10 18:35:15 +11002400 xfs_icsb_lock(mp);
David Chinner8d280b92006-03-14 13:13:09 +11002401
David Chinner20b64282007-02-10 18:35:09 +11002402 /*
2403 * running atomically.
2404 *
2405 * This will leave the counter in the correct state for future
2406 * accesses. After the rebalance, we simply try again and our retry
2407 * will either succeed through the fast path or slow path without
2408 * another balance operation being required.
2409 */
Christoph Hellwig45af6c62008-04-22 17:34:44 +10002410 xfs_icsb_balance_counter(mp, field, delta);
David Chinner03135cf2007-02-10 18:35:15 +11002411 xfs_icsb_unlock(mp);
David Chinner20b64282007-02-10 18:35:09 +11002412 goto again;
David Chinner8d280b92006-03-14 13:13:09 +11002413}
2414
David Chinner8d280b92006-03-14 13:13:09 +11002415#endif