]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * 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. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * 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 | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
1da177e4 | 20 | #include "xfs_types.h" |
a844f451 | 21 | #include "xfs_bit.h" |
1da177e4 | 22 | #include "xfs_log.h" |
a844f451 | 23 | #include "xfs_inum.h" |
1da177e4 LT |
24 | #include "xfs_trans.h" |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_ag.h" | |
1da177e4 LT |
27 | #include "xfs_dir2.h" |
28 | #include "xfs_dmapi.h" | |
29 | #include "xfs_mount.h" | |
1da177e4 | 30 | #include "xfs_bmap_btree.h" |
a844f451 | 31 | #include "xfs_alloc_btree.h" |
1da177e4 | 32 | #include "xfs_ialloc_btree.h" |
1da177e4 | 33 | #include "xfs_dir2_sf.h" |
a844f451 | 34 | #include "xfs_attr_sf.h" |
1da177e4 LT |
35 | #include "xfs_dinode.h" |
36 | #include "xfs_inode.h" | |
a844f451 NS |
37 | #include "xfs_btree.h" |
38 | #include "xfs_ialloc.h" | |
1da177e4 LT |
39 | #include "xfs_alloc.h" |
40 | #include "xfs_rtalloc.h" | |
41 | #include "xfs_bmap.h" | |
42 | #include "xfs_error.h" | |
1da177e4 LT |
43 | #include "xfs_rw.h" |
44 | #include "xfs_quota.h" | |
45 | #include "xfs_fsops.h" | |
43355099 | 46 | #include "xfs_utils.h" |
1da177e4 | 47 | |
e5720eec | 48 | STATIC int xfs_mount_log_sb(xfs_mount_t *, __int64_t); |
1da177e4 | 49 | STATIC int xfs_uuid_mount(xfs_mount_t *); |
ba0f32d4 | 50 | STATIC void xfs_unmountfs_wait(xfs_mount_t *); |
1da177e4 | 51 | |
8d280b98 DC |
52 | |
53 | #ifdef HAVE_PERCPU_SB | |
54 | STATIC void xfs_icsb_destroy_counters(xfs_mount_t *); | |
20f4ebf2 | 55 | STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t, |
45af6c6d CH |
56 | int); |
57 | STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t, | |
58 | int); | |
8d280b98 | 59 | STATIC int xfs_icsb_modify_counters(xfs_mount_t *, xfs_sb_field_t, |
20f4ebf2 | 60 | int64_t, int); |
36fbe6e6 | 61 | STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t); |
8d280b98 DC |
62 | |
63 | #else | |
64 | ||
65 | #define xfs_icsb_destroy_counters(mp) do { } while (0) | |
45af6c6d CH |
66 | #define xfs_icsb_balance_counter(mp, a, b) do { } while (0) |
67 | #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0) | |
8d280b98 | 68 | #define xfs_icsb_modify_counters(mp, a, b, c) do { } while (0) |
8d280b98 DC |
69 | |
70 | #endif | |
71 | ||
1df84c93 | 72 | static const struct { |
8d280b98 DC |
73 | short offset; |
74 | short type; /* 0 = integer | |
75 | * 1 = binary / string (no translation) | |
76 | */ | |
1da177e4 LT |
77 | } xfs_sb_info[] = { |
78 | { offsetof(xfs_sb_t, sb_magicnum), 0 }, | |
79 | { offsetof(xfs_sb_t, sb_blocksize), 0 }, | |
80 | { offsetof(xfs_sb_t, sb_dblocks), 0 }, | |
81 | { offsetof(xfs_sb_t, sb_rblocks), 0 }, | |
82 | { offsetof(xfs_sb_t, sb_rextents), 0 }, | |
83 | { offsetof(xfs_sb_t, sb_uuid), 1 }, | |
84 | { offsetof(xfs_sb_t, sb_logstart), 0 }, | |
85 | { offsetof(xfs_sb_t, sb_rootino), 0 }, | |
86 | { offsetof(xfs_sb_t, sb_rbmino), 0 }, | |
87 | { offsetof(xfs_sb_t, sb_rsumino), 0 }, | |
88 | { offsetof(xfs_sb_t, sb_rextsize), 0 }, | |
89 | { offsetof(xfs_sb_t, sb_agblocks), 0 }, | |
90 | { offsetof(xfs_sb_t, sb_agcount), 0 }, | |
91 | { offsetof(xfs_sb_t, sb_rbmblocks), 0 }, | |
92 | { offsetof(xfs_sb_t, sb_logblocks), 0 }, | |
93 | { offsetof(xfs_sb_t, sb_versionnum), 0 }, | |
94 | { offsetof(xfs_sb_t, sb_sectsize), 0 }, | |
95 | { offsetof(xfs_sb_t, sb_inodesize), 0 }, | |
96 | { offsetof(xfs_sb_t, sb_inopblock), 0 }, | |
97 | { offsetof(xfs_sb_t, sb_fname[0]), 1 }, | |
98 | { offsetof(xfs_sb_t, sb_blocklog), 0 }, | |
99 | { offsetof(xfs_sb_t, sb_sectlog), 0 }, | |
100 | { offsetof(xfs_sb_t, sb_inodelog), 0 }, | |
101 | { offsetof(xfs_sb_t, sb_inopblog), 0 }, | |
102 | { offsetof(xfs_sb_t, sb_agblklog), 0 }, | |
103 | { offsetof(xfs_sb_t, sb_rextslog), 0 }, | |
104 | { offsetof(xfs_sb_t, sb_inprogress), 0 }, | |
105 | { offsetof(xfs_sb_t, sb_imax_pct), 0 }, | |
106 | { offsetof(xfs_sb_t, sb_icount), 0 }, | |
107 | { offsetof(xfs_sb_t, sb_ifree), 0 }, | |
108 | { offsetof(xfs_sb_t, sb_fdblocks), 0 }, | |
109 | { offsetof(xfs_sb_t, sb_frextents), 0 }, | |
110 | { offsetof(xfs_sb_t, sb_uquotino), 0 }, | |
111 | { offsetof(xfs_sb_t, sb_gquotino), 0 }, | |
112 | { offsetof(xfs_sb_t, sb_qflags), 0 }, | |
113 | { offsetof(xfs_sb_t, sb_flags), 0 }, | |
114 | { offsetof(xfs_sb_t, sb_shared_vn), 0 }, | |
115 | { offsetof(xfs_sb_t, sb_inoalignmt), 0 }, | |
116 | { offsetof(xfs_sb_t, sb_unit), 0 }, | |
117 | { offsetof(xfs_sb_t, sb_width), 0 }, | |
118 | { offsetof(xfs_sb_t, sb_dirblklog), 0 }, | |
119 | { offsetof(xfs_sb_t, sb_logsectlog), 0 }, | |
120 | { offsetof(xfs_sb_t, sb_logsectsize),0 }, | |
121 | { offsetof(xfs_sb_t, sb_logsunit), 0 }, | |
122 | { offsetof(xfs_sb_t, sb_features2), 0 }, | |
ee1c0908 | 123 | { offsetof(xfs_sb_t, sb_bad_features2), 0 }, |
1da177e4 LT |
124 | { sizeof(xfs_sb_t), 0 } |
125 | }; | |
126 | ||
127 | /* | |
128 | * Return a pointer to an initialized xfs_mount structure. | |
129 | */ | |
130 | xfs_mount_t * | |
131 | xfs_mount_init(void) | |
132 | { | |
133 | xfs_mount_t *mp; | |
134 | ||
8d280b98 DC |
135 | mp = kmem_zalloc(sizeof(xfs_mount_t), KM_SLEEP); |
136 | ||
137 | if (xfs_icsb_init_counters(mp)) { | |
138 | mp->m_flags |= XFS_MOUNT_NO_PERCPU_SB; | |
139 | } | |
1da177e4 | 140 | |
007c61c6 | 141 | spin_lock_init(&mp->m_sb_lock); |
794ee1ba | 142 | mutex_init(&mp->m_ilock); |
cc92e7ac | 143 | mutex_init(&mp->m_growlock); |
1da177e4 LT |
144 | atomic_set(&mp->m_active_trans, 0); |
145 | ||
146 | return mp; | |
147 | } | |
148 | ||
149 | /* | |
150 | * Free up the resources associated with a mount structure. Assume that | |
151 | * the structure was initially zeroed, so we can tell which fields got | |
152 | * initialized. | |
153 | */ | |
154 | void | |
155 | xfs_mount_free( | |
745f6919 | 156 | xfs_mount_t *mp) |
1da177e4 | 157 | { |
1da177e4 LT |
158 | if (mp->m_perag) { |
159 | int agno; | |
160 | ||
161 | for (agno = 0; agno < mp->m_maxagi; agno++) | |
162 | if (mp->m_perag[agno].pagb_list) | |
f0e2d93c DV |
163 | kmem_free(mp->m_perag[agno].pagb_list); |
164 | kmem_free(mp->m_perag); | |
1da177e4 LT |
165 | } |
166 | ||
287f3dad | 167 | spinlock_destroy(&mp->m_ail_lock); |
1da177e4 LT |
168 | spinlock_destroy(&mp->m_sb_lock); |
169 | mutex_destroy(&mp->m_ilock); | |
cc92e7ac | 170 | mutex_destroy(&mp->m_growlock); |
1da177e4 LT |
171 | if (mp->m_quotainfo) |
172 | XFS_QM_DONE(mp); | |
173 | ||
174 | if (mp->m_fsname != NULL) | |
f0e2d93c | 175 | kmem_free(mp->m_fsname); |
fc1f8c1c | 176 | if (mp->m_rtname != NULL) |
f0e2d93c | 177 | kmem_free(mp->m_rtname); |
fc1f8c1c | 178 | if (mp->m_logname != NULL) |
f0e2d93c | 179 | kmem_free(mp->m_logname); |
1da177e4 | 180 | |
8d280b98 | 181 | xfs_icsb_destroy_counters(mp); |
1da177e4 LT |
182 | } |
183 | ||
4cc929ee NS |
184 | /* |
185 | * Check size of device based on the (data/realtime) block count. | |
186 | * Note: this check is used by the growfs code as well as mount. | |
187 | */ | |
188 | int | |
189 | xfs_sb_validate_fsb_count( | |
190 | xfs_sb_t *sbp, | |
191 | __uint64_t nblocks) | |
192 | { | |
193 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
194 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
195 | ||
196 | #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */ | |
197 | if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) | |
198 | return E2BIG; | |
199 | #else /* Limited by UINT_MAX of sectors */ | |
200 | if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX) | |
201 | return E2BIG; | |
202 | #endif | |
203 | return 0; | |
204 | } | |
1da177e4 LT |
205 | |
206 | /* | |
207 | * Check the validity of the SB found. | |
208 | */ | |
209 | STATIC int | |
210 | xfs_mount_validate_sb( | |
211 | xfs_mount_t *mp, | |
764d1f89 NS |
212 | xfs_sb_t *sbp, |
213 | int flags) | |
1da177e4 LT |
214 | { |
215 | /* | |
216 | * If the log device and data device have the | |
217 | * same device number, the log is internal. | |
218 | * Consequently, the sb_logstart should be non-zero. If | |
219 | * we have a zero sb_logstart in this case, we may be trying to mount | |
220 | * a volume filesystem in a non-volume manner. | |
221 | */ | |
222 | if (sbp->sb_magicnum != XFS_SB_MAGIC) { | |
764d1f89 | 223 | xfs_fs_mount_cmn_err(flags, "bad magic number"); |
1da177e4 LT |
224 | return XFS_ERROR(EWRONGFS); |
225 | } | |
226 | ||
62118709 | 227 | if (!xfs_sb_good_version(sbp)) { |
764d1f89 | 228 | xfs_fs_mount_cmn_err(flags, "bad version"); |
1da177e4 LT |
229 | return XFS_ERROR(EWRONGFS); |
230 | } | |
231 | ||
232 | if (unlikely( | |
233 | sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) { | |
764d1f89 NS |
234 | xfs_fs_mount_cmn_err(flags, |
235 | "filesystem is marked as having an external log; " | |
236 | "specify logdev on the\nmount command line."); | |
237 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
238 | } |
239 | ||
240 | if (unlikely( | |
241 | sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) { | |
764d1f89 NS |
242 | xfs_fs_mount_cmn_err(flags, |
243 | "filesystem is marked as having an internal log; " | |
244 | "do not specify logdev on\nthe mount command line."); | |
245 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
246 | } |
247 | ||
248 | /* | |
249 | * More sanity checking. These were stolen directly from | |
250 | * xfs_repair. | |
251 | */ | |
252 | if (unlikely( | |
253 | sbp->sb_agcount <= 0 || | |
254 | sbp->sb_sectsize < XFS_MIN_SECTORSIZE || | |
255 | sbp->sb_sectsize > XFS_MAX_SECTORSIZE || | |
256 | sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG || | |
257 | sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG || | |
258 | sbp->sb_blocksize < XFS_MIN_BLOCKSIZE || | |
259 | sbp->sb_blocksize > XFS_MAX_BLOCKSIZE || | |
260 | sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG || | |
261 | sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG || | |
262 | sbp->sb_inodesize < XFS_DINODE_MIN_SIZE || | |
263 | sbp->sb_inodesize > XFS_DINODE_MAX_SIZE || | |
9f989c94 NS |
264 | sbp->sb_inodelog < XFS_DINODE_MIN_LOG || |
265 | sbp->sb_inodelog > XFS_DINODE_MAX_LOG || | |
266 | (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) || | |
1da177e4 LT |
267 | (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) || |
268 | (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) || | |
e50bd16f | 269 | (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) { |
764d1f89 | 270 | xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed"); |
1da177e4 LT |
271 | return XFS_ERROR(EFSCORRUPTED); |
272 | } | |
273 | ||
274 | /* | |
275 | * Sanity check AG count, size fields against data size field | |
276 | */ | |
277 | if (unlikely( | |
278 | sbp->sb_dblocks == 0 || | |
279 | sbp->sb_dblocks > | |
280 | (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks || | |
281 | sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) * | |
282 | sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) { | |
764d1f89 | 283 | xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed"); |
1da177e4 LT |
284 | return XFS_ERROR(EFSCORRUPTED); |
285 | } | |
286 | ||
4cc929ee NS |
287 | if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) || |
288 | xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) { | |
764d1f89 NS |
289 | xfs_fs_mount_cmn_err(flags, |
290 | "file system too large to be mounted on this system."); | |
1da177e4 LT |
291 | return XFS_ERROR(E2BIG); |
292 | } | |
293 | ||
294 | if (unlikely(sbp->sb_inprogress)) { | |
764d1f89 | 295 | xfs_fs_mount_cmn_err(flags, "file system busy"); |
1da177e4 LT |
296 | return XFS_ERROR(EFSCORRUPTED); |
297 | } | |
298 | ||
de20614b NS |
299 | /* |
300 | * Version 1 directory format has never worked on Linux. | |
301 | */ | |
62118709 | 302 | if (unlikely(!xfs_sb_version_hasdirv2(sbp))) { |
764d1f89 NS |
303 | xfs_fs_mount_cmn_err(flags, |
304 | "file system using version 1 directory format"); | |
de20614b NS |
305 | return XFS_ERROR(ENOSYS); |
306 | } | |
307 | ||
1da177e4 LT |
308 | /* |
309 | * Until this is fixed only page-sized or smaller data blocks work. | |
310 | */ | |
311 | if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) { | |
764d1f89 NS |
312 | xfs_fs_mount_cmn_err(flags, |
313 | "file system with blocksize %d bytes", | |
1da177e4 | 314 | sbp->sb_blocksize); |
764d1f89 NS |
315 | xfs_fs_mount_cmn_err(flags, |
316 | "only pagesize (%ld) or less will currently work.", | |
1da177e4 LT |
317 | PAGE_SIZE); |
318 | return XFS_ERROR(ENOSYS); | |
319 | } | |
320 | ||
321 | return 0; | |
322 | } | |
323 | ||
da353b0d DC |
324 | STATIC void |
325 | xfs_initialize_perag_icache( | |
326 | xfs_perag_t *pag) | |
327 | { | |
328 | if (!pag->pag_ici_init) { | |
329 | rwlock_init(&pag->pag_ici_lock); | |
330 | INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); | |
331 | pag->pag_ici_init = 1; | |
332 | } | |
333 | } | |
334 | ||
1da177e4 | 335 | xfs_agnumber_t |
c11e2c36 | 336 | xfs_initialize_perag( |
c11e2c36 NS |
337 | xfs_mount_t *mp, |
338 | xfs_agnumber_t agcount) | |
1da177e4 LT |
339 | { |
340 | xfs_agnumber_t index, max_metadata; | |
341 | xfs_perag_t *pag; | |
342 | xfs_agino_t agino; | |
343 | xfs_ino_t ino; | |
344 | xfs_sb_t *sbp = &mp->m_sb; | |
345 | xfs_ino_t max_inum = XFS_MAXINUMBER_32; | |
346 | ||
347 | /* Check to see if the filesystem can overflow 32 bit inodes */ | |
348 | agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0); | |
349 | ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino); | |
350 | ||
351 | /* Clear the mount flag if no inode can overflow 32 bits | |
352 | * on this filesystem, or if specifically requested.. | |
353 | */ | |
bd186aa9 | 354 | if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > max_inum) { |
1da177e4 LT |
355 | mp->m_flags |= XFS_MOUNT_32BITINODES; |
356 | } else { | |
357 | mp->m_flags &= ~XFS_MOUNT_32BITINODES; | |
358 | } | |
359 | ||
360 | /* If we can overflow then setup the ag headers accordingly */ | |
361 | if (mp->m_flags & XFS_MOUNT_32BITINODES) { | |
362 | /* Calculate how much should be reserved for inodes to | |
363 | * meet the max inode percentage. | |
364 | */ | |
365 | if (mp->m_maxicount) { | |
366 | __uint64_t icount; | |
367 | ||
368 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; | |
369 | do_div(icount, 100); | |
370 | icount += sbp->sb_agblocks - 1; | |
a749ee86 | 371 | do_div(icount, sbp->sb_agblocks); |
1da177e4 LT |
372 | max_metadata = icount; |
373 | } else { | |
374 | max_metadata = agcount; | |
375 | } | |
376 | for (index = 0; index < agcount; index++) { | |
377 | ino = XFS_AGINO_TO_INO(mp, index, agino); | |
378 | if (ino > max_inum) { | |
379 | index++; | |
380 | break; | |
381 | } | |
382 | ||
c41564b5 | 383 | /* This ag is preferred for inodes */ |
1da177e4 LT |
384 | pag = &mp->m_perag[index]; |
385 | pag->pagi_inodeok = 1; | |
386 | if (index < max_metadata) | |
387 | pag->pagf_metadata = 1; | |
da353b0d | 388 | xfs_initialize_perag_icache(pag); |
1da177e4 LT |
389 | } |
390 | } else { | |
391 | /* Setup default behavior for smaller filesystems */ | |
392 | for (index = 0; index < agcount; index++) { | |
393 | pag = &mp->m_perag[index]; | |
394 | pag->pagi_inodeok = 1; | |
da353b0d | 395 | xfs_initialize_perag_icache(pag); |
1da177e4 LT |
396 | } |
397 | } | |
398 | return index; | |
399 | } | |
400 | ||
2bdf7cd0 CH |
401 | void |
402 | xfs_sb_from_disk( | |
403 | xfs_sb_t *to, | |
404 | xfs_dsb_t *from) | |
405 | { | |
406 | to->sb_magicnum = be32_to_cpu(from->sb_magicnum); | |
407 | to->sb_blocksize = be32_to_cpu(from->sb_blocksize); | |
408 | to->sb_dblocks = be64_to_cpu(from->sb_dblocks); | |
409 | to->sb_rblocks = be64_to_cpu(from->sb_rblocks); | |
410 | to->sb_rextents = be64_to_cpu(from->sb_rextents); | |
411 | memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid)); | |
412 | to->sb_logstart = be64_to_cpu(from->sb_logstart); | |
413 | to->sb_rootino = be64_to_cpu(from->sb_rootino); | |
414 | to->sb_rbmino = be64_to_cpu(from->sb_rbmino); | |
415 | to->sb_rsumino = be64_to_cpu(from->sb_rsumino); | |
416 | to->sb_rextsize = be32_to_cpu(from->sb_rextsize); | |
417 | to->sb_agblocks = be32_to_cpu(from->sb_agblocks); | |
418 | to->sb_agcount = be32_to_cpu(from->sb_agcount); | |
419 | to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks); | |
420 | to->sb_logblocks = be32_to_cpu(from->sb_logblocks); | |
421 | to->sb_versionnum = be16_to_cpu(from->sb_versionnum); | |
422 | to->sb_sectsize = be16_to_cpu(from->sb_sectsize); | |
423 | to->sb_inodesize = be16_to_cpu(from->sb_inodesize); | |
424 | to->sb_inopblock = be16_to_cpu(from->sb_inopblock); | |
425 | memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname)); | |
426 | to->sb_blocklog = from->sb_blocklog; | |
427 | to->sb_sectlog = from->sb_sectlog; | |
428 | to->sb_inodelog = from->sb_inodelog; | |
429 | to->sb_inopblog = from->sb_inopblog; | |
430 | to->sb_agblklog = from->sb_agblklog; | |
431 | to->sb_rextslog = from->sb_rextslog; | |
432 | to->sb_inprogress = from->sb_inprogress; | |
433 | to->sb_imax_pct = from->sb_imax_pct; | |
434 | to->sb_icount = be64_to_cpu(from->sb_icount); | |
435 | to->sb_ifree = be64_to_cpu(from->sb_ifree); | |
436 | to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks); | |
437 | to->sb_frextents = be64_to_cpu(from->sb_frextents); | |
438 | to->sb_uquotino = be64_to_cpu(from->sb_uquotino); | |
439 | to->sb_gquotino = be64_to_cpu(from->sb_gquotino); | |
440 | to->sb_qflags = be16_to_cpu(from->sb_qflags); | |
441 | to->sb_flags = from->sb_flags; | |
442 | to->sb_shared_vn = from->sb_shared_vn; | |
443 | to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt); | |
444 | to->sb_unit = be32_to_cpu(from->sb_unit); | |
445 | to->sb_width = be32_to_cpu(from->sb_width); | |
446 | to->sb_dirblklog = from->sb_dirblklog; | |
447 | to->sb_logsectlog = from->sb_logsectlog; | |
448 | to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize); | |
449 | to->sb_logsunit = be32_to_cpu(from->sb_logsunit); | |
450 | to->sb_features2 = be32_to_cpu(from->sb_features2); | |
ee1c0908 | 451 | to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2); |
2bdf7cd0 CH |
452 | } |
453 | ||
1da177e4 | 454 | /* |
2bdf7cd0 | 455 | * Copy in core superblock to ondisk one. |
1da177e4 | 456 | * |
2bdf7cd0 | 457 | * The fields argument is mask of superblock fields to copy. |
1da177e4 LT |
458 | */ |
459 | void | |
2bdf7cd0 CH |
460 | xfs_sb_to_disk( |
461 | xfs_dsb_t *to, | |
462 | xfs_sb_t *from, | |
1da177e4 LT |
463 | __int64_t fields) |
464 | { | |
2bdf7cd0 CH |
465 | xfs_caddr_t to_ptr = (xfs_caddr_t)to; |
466 | xfs_caddr_t from_ptr = (xfs_caddr_t)from; | |
1da177e4 LT |
467 | xfs_sb_field_t f; |
468 | int first; | |
469 | int size; | |
470 | ||
1da177e4 | 471 | ASSERT(fields); |
1da177e4 LT |
472 | if (!fields) |
473 | return; | |
474 | ||
1da177e4 LT |
475 | while (fields) { |
476 | f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields); | |
477 | first = xfs_sb_info[f].offset; | |
478 | size = xfs_sb_info[f + 1].offset - first; | |
479 | ||
480 | ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1); | |
481 | ||
482 | if (size == 1 || xfs_sb_info[f].type == 1) { | |
2bdf7cd0 | 483 | memcpy(to_ptr + first, from_ptr + first, size); |
1da177e4 LT |
484 | } else { |
485 | switch (size) { | |
486 | case 2: | |
2bdf7cd0 CH |
487 | *(__be16 *)(to_ptr + first) = |
488 | cpu_to_be16(*(__u16 *)(from_ptr + first)); | |
1da177e4 LT |
489 | break; |
490 | case 4: | |
2bdf7cd0 CH |
491 | *(__be32 *)(to_ptr + first) = |
492 | cpu_to_be32(*(__u32 *)(from_ptr + first)); | |
1da177e4 LT |
493 | break; |
494 | case 8: | |
2bdf7cd0 CH |
495 | *(__be64 *)(to_ptr + first) = |
496 | cpu_to_be64(*(__u64 *)(from_ptr + first)); | |
1da177e4 LT |
497 | break; |
498 | default: | |
499 | ASSERT(0); | |
500 | } | |
501 | } | |
502 | ||
503 | fields &= ~(1LL << f); | |
504 | } | |
505 | } | |
506 | ||
507 | /* | |
508 | * xfs_readsb | |
509 | * | |
510 | * Does the initial read of the superblock. | |
511 | */ | |
512 | int | |
764d1f89 | 513 | xfs_readsb(xfs_mount_t *mp, int flags) |
1da177e4 LT |
514 | { |
515 | unsigned int sector_size; | |
516 | unsigned int extra_flags; | |
517 | xfs_buf_t *bp; | |
1da177e4 LT |
518 | int error; |
519 | ||
520 | ASSERT(mp->m_sb_bp == NULL); | |
521 | ASSERT(mp->m_ddev_targp != NULL); | |
522 | ||
523 | /* | |
524 | * Allocate a (locked) buffer to hold the superblock. | |
525 | * This will be kept around at all times to optimize | |
526 | * access to the superblock. | |
527 | */ | |
528 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
529 | extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED; | |
530 | ||
531 | bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR, | |
532 | BTOBB(sector_size), extra_flags); | |
533 | if (!bp || XFS_BUF_ISERROR(bp)) { | |
764d1f89 | 534 | xfs_fs_mount_cmn_err(flags, "SB read failed"); |
1da177e4 LT |
535 | error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM; |
536 | goto fail; | |
537 | } | |
538 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
539 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); | |
540 | ||
541 | /* | |
542 | * Initialize the mount structure from the superblock. | |
543 | * But first do some basic consistency checking. | |
544 | */ | |
2bdf7cd0 | 545 | xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp)); |
1da177e4 | 546 | |
764d1f89 | 547 | error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags); |
1da177e4 | 548 | if (error) { |
764d1f89 | 549 | xfs_fs_mount_cmn_err(flags, "SB validate failed"); |
1da177e4 LT |
550 | goto fail; |
551 | } | |
552 | ||
553 | /* | |
554 | * We must be able to do sector-sized and sector-aligned IO. | |
555 | */ | |
556 | if (sector_size > mp->m_sb.sb_sectsize) { | |
764d1f89 NS |
557 | xfs_fs_mount_cmn_err(flags, |
558 | "device supports only %u byte sectors (not %u)", | |
1da177e4 LT |
559 | sector_size, mp->m_sb.sb_sectsize); |
560 | error = ENOSYS; | |
561 | goto fail; | |
562 | } | |
563 | ||
564 | /* | |
565 | * If device sector size is smaller than the superblock size, | |
566 | * re-read the superblock so the buffer is correctly sized. | |
567 | */ | |
568 | if (sector_size < mp->m_sb.sb_sectsize) { | |
569 | XFS_BUF_UNMANAGE(bp); | |
570 | xfs_buf_relse(bp); | |
571 | sector_size = mp->m_sb.sb_sectsize; | |
572 | bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR, | |
573 | BTOBB(sector_size), extra_flags); | |
574 | if (!bp || XFS_BUF_ISERROR(bp)) { | |
764d1f89 | 575 | xfs_fs_mount_cmn_err(flags, "SB re-read failed"); |
1da177e4 LT |
576 | error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM; |
577 | goto fail; | |
578 | } | |
579 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
580 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); | |
581 | } | |
582 | ||
5478eead LM |
583 | /* Initialize per-cpu counters */ |
584 | xfs_icsb_reinit_counters(mp); | |
8d280b98 | 585 | |
1da177e4 LT |
586 | mp->m_sb_bp = bp; |
587 | xfs_buf_relse(bp); | |
588 | ASSERT(XFS_BUF_VALUSEMA(bp) > 0); | |
589 | return 0; | |
590 | ||
591 | fail: | |
592 | if (bp) { | |
593 | XFS_BUF_UNMANAGE(bp); | |
594 | xfs_buf_relse(bp); | |
595 | } | |
596 | return error; | |
597 | } | |
598 | ||
599 | ||
600 | /* | |
601 | * xfs_mount_common | |
602 | * | |
603 | * Mount initialization code establishing various mount | |
604 | * fields from the superblock associated with the given | |
605 | * mount structure | |
606 | */ | |
ba0f32d4 | 607 | STATIC void |
1da177e4 LT |
608 | xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp) |
609 | { | |
610 | int i; | |
611 | ||
612 | mp->m_agfrotor = mp->m_agirotor = 0; | |
007c61c6 | 613 | spin_lock_init(&mp->m_agirotor_lock); |
1da177e4 LT |
614 | mp->m_maxagi = mp->m_sb.sb_agcount; |
615 | mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG; | |
616 | mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT; | |
617 | mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT; | |
618 | mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1; | |
619 | mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog; | |
620 | mp->m_litino = sbp->sb_inodesize - | |
621 | ((uint)sizeof(xfs_dinode_core_t) + (uint)sizeof(xfs_agino_t)); | |
622 | mp->m_blockmask = sbp->sb_blocksize - 1; | |
623 | mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG; | |
624 | mp->m_blockwmask = mp->m_blockwsize - 1; | |
625 | INIT_LIST_HEAD(&mp->m_del_inodes); | |
626 | ||
627 | /* | |
628 | * Setup for attributes, in case they get created. | |
629 | * This value is for inodes getting attributes for the first time, | |
630 | * the per-inode value is for old attribute values. | |
631 | */ | |
632 | ASSERT(sbp->sb_inodesize >= 256 && sbp->sb_inodesize <= 2048); | |
633 | switch (sbp->sb_inodesize) { | |
634 | case 256: | |
d8cc890d NS |
635 | mp->m_attroffset = XFS_LITINO(mp) - |
636 | XFS_BMDR_SPACE_CALC(MINABTPTRS); | |
1da177e4 LT |
637 | break; |
638 | case 512: | |
639 | case 1024: | |
640 | case 2048: | |
d8cc890d | 641 | mp->m_attroffset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS); |
1da177e4 LT |
642 | break; |
643 | default: | |
644 | ASSERT(0); | |
645 | } | |
646 | ASSERT(mp->m_attroffset < XFS_LITINO(mp)); | |
647 | ||
648 | for (i = 0; i < 2; i++) { | |
649 | mp->m_alloc_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize, | |
650 | xfs_alloc, i == 0); | |
651 | mp->m_alloc_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize, | |
652 | xfs_alloc, i == 0); | |
653 | } | |
654 | for (i = 0; i < 2; i++) { | |
655 | mp->m_bmap_dmxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize, | |
656 | xfs_bmbt, i == 0); | |
657 | mp->m_bmap_dmnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize, | |
658 | xfs_bmbt, i == 0); | |
659 | } | |
660 | for (i = 0; i < 2; i++) { | |
661 | mp->m_inobt_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize, | |
662 | xfs_inobt, i == 0); | |
663 | mp->m_inobt_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize, | |
664 | xfs_inobt, i == 0); | |
665 | } | |
666 | ||
667 | mp->m_bsize = XFS_FSB_TO_BB(mp, 1); | |
668 | mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK, | |
669 | sbp->sb_inopblock); | |
670 | mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog; | |
671 | } | |
92821e2b DC |
672 | |
673 | /* | |
674 | * xfs_initialize_perag_data | |
675 | * | |
676 | * Read in each per-ag structure so we can count up the number of | |
677 | * allocated inodes, free inodes and used filesystem blocks as this | |
678 | * information is no longer persistent in the superblock. Once we have | |
679 | * this information, write it into the in-core superblock structure. | |
680 | */ | |
681 | STATIC int | |
682 | xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount) | |
683 | { | |
684 | xfs_agnumber_t index; | |
685 | xfs_perag_t *pag; | |
686 | xfs_sb_t *sbp = &mp->m_sb; | |
687 | uint64_t ifree = 0; | |
688 | uint64_t ialloc = 0; | |
689 | uint64_t bfree = 0; | |
690 | uint64_t bfreelst = 0; | |
691 | uint64_t btree = 0; | |
692 | int error; | |
92821e2b DC |
693 | |
694 | for (index = 0; index < agcount; index++) { | |
695 | /* | |
696 | * read the agf, then the agi. This gets us | |
697 | * all the inforamtion we need and populates the | |
698 | * per-ag structures for us. | |
699 | */ | |
700 | error = xfs_alloc_pagf_init(mp, NULL, index, 0); | |
701 | if (error) | |
702 | return error; | |
703 | ||
704 | error = xfs_ialloc_pagi_init(mp, NULL, index); | |
705 | if (error) | |
706 | return error; | |
707 | pag = &mp->m_perag[index]; | |
708 | ifree += pag->pagi_freecount; | |
709 | ialloc += pag->pagi_count; | |
710 | bfree += pag->pagf_freeblks; | |
711 | bfreelst += pag->pagf_flcount; | |
712 | btree += pag->pagf_btreeblks; | |
713 | } | |
714 | /* | |
715 | * Overwrite incore superblock counters with just-read data | |
716 | */ | |
3685c2a1 | 717 | spin_lock(&mp->m_sb_lock); |
92821e2b DC |
718 | sbp->sb_ifree = ifree; |
719 | sbp->sb_icount = ialloc; | |
720 | sbp->sb_fdblocks = bfree + bfreelst + btree; | |
3685c2a1 | 721 | spin_unlock(&mp->m_sb_lock); |
92821e2b DC |
722 | |
723 | /* Fixup the per-cpu counters as well. */ | |
724 | xfs_icsb_reinit_counters(mp); | |
725 | ||
726 | return 0; | |
727 | } | |
728 | ||
1da177e4 | 729 | /* |
0771fb45 | 730 | * Update alignment values based on mount options and sb values |
1da177e4 | 731 | */ |
0771fb45 ES |
732 | STATIC int |
733 | xfs_update_alignment(xfs_mount_t *mp, int mfsi_flags, __uint64_t *update_flags) | |
1da177e4 | 734 | { |
1da177e4 | 735 | xfs_sb_t *sbp = &(mp->m_sb); |
1da177e4 | 736 | |
1da177e4 LT |
737 | if (mp->m_dalign && !(mfsi_flags & XFS_MFSI_SECOND)) { |
738 | /* | |
739 | * If stripe unit and stripe width are not multiples | |
740 | * of the fs blocksize turn off alignment. | |
741 | */ | |
742 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
743 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
744 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
745 | cmn_err(CE_WARN, | |
746 | "XFS: alignment check 1 failed"); | |
0771fb45 | 747 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
748 | } |
749 | mp->m_dalign = mp->m_swidth = 0; | |
750 | } else { | |
751 | /* | |
752 | * Convert the stripe unit and width to FSBs. | |
753 | */ | |
754 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
755 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
756 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
0771fb45 | 757 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
758 | } |
759 | xfs_fs_cmn_err(CE_WARN, mp, | |
760 | "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)", | |
761 | mp->m_dalign, mp->m_swidth, | |
762 | sbp->sb_agblocks); | |
763 | ||
764 | mp->m_dalign = 0; | |
765 | mp->m_swidth = 0; | |
766 | } else if (mp->m_dalign) { | |
767 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
768 | } else { | |
769 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
770 | xfs_fs_cmn_err(CE_WARN, mp, | |
771 | "stripe alignment turned off: sunit(%d) less than bsize(%d)", | |
772 | mp->m_dalign, | |
773 | mp->m_blockmask +1); | |
0771fb45 | 774 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
775 | } |
776 | mp->m_swidth = 0; | |
777 | } | |
778 | } | |
779 | ||
780 | /* | |
781 | * Update superblock with new values | |
782 | * and log changes | |
783 | */ | |
62118709 | 784 | if (xfs_sb_version_hasdalign(sbp)) { |
1da177e4 LT |
785 | if (sbp->sb_unit != mp->m_dalign) { |
786 | sbp->sb_unit = mp->m_dalign; | |
0771fb45 | 787 | *update_flags |= XFS_SB_UNIT; |
1da177e4 LT |
788 | } |
789 | if (sbp->sb_width != mp->m_swidth) { | |
790 | sbp->sb_width = mp->m_swidth; | |
0771fb45 | 791 | *update_flags |= XFS_SB_WIDTH; |
1da177e4 LT |
792 | } |
793 | } | |
794 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
62118709 | 795 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
1da177e4 LT |
796 | mp->m_dalign = sbp->sb_unit; |
797 | mp->m_swidth = sbp->sb_width; | |
798 | } | |
799 | ||
0771fb45 ES |
800 | return 0; |
801 | } | |
1da177e4 | 802 | |
0771fb45 ES |
803 | /* |
804 | * Set the maximum inode count for this filesystem | |
805 | */ | |
806 | STATIC void | |
807 | xfs_set_maxicount(xfs_mount_t *mp) | |
808 | { | |
809 | xfs_sb_t *sbp = &(mp->m_sb); | |
810 | __uint64_t icount; | |
1da177e4 | 811 | |
0771fb45 ES |
812 | if (sbp->sb_imax_pct) { |
813 | /* | |
814 | * Make sure the maximum inode count is a multiple | |
815 | * of the units we allocate inodes in. | |
1da177e4 | 816 | */ |
1da177e4 LT |
817 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; |
818 | do_div(icount, 100); | |
819 | do_div(icount, mp->m_ialloc_blks); | |
820 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << | |
821 | sbp->sb_inopblog; | |
0771fb45 | 822 | } else { |
1da177e4 | 823 | mp->m_maxicount = 0; |
1da177e4 | 824 | } |
0771fb45 ES |
825 | } |
826 | ||
827 | /* | |
828 | * Set the default minimum read and write sizes unless | |
829 | * already specified in a mount option. | |
830 | * We use smaller I/O sizes when the file system | |
831 | * is being used for NFS service (wsync mount option). | |
832 | */ | |
833 | STATIC void | |
834 | xfs_set_rw_sizes(xfs_mount_t *mp) | |
835 | { | |
836 | xfs_sb_t *sbp = &(mp->m_sb); | |
837 | int readio_log, writeio_log; | |
1da177e4 | 838 | |
1da177e4 LT |
839 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
840 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
841 | readio_log = XFS_WSYNC_READIO_LOG; | |
842 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
843 | } else { | |
844 | readio_log = XFS_READIO_LOG_LARGE; | |
845 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
846 | } | |
847 | } else { | |
848 | readio_log = mp->m_readio_log; | |
849 | writeio_log = mp->m_writeio_log; | |
850 | } | |
851 | ||
1da177e4 LT |
852 | if (sbp->sb_blocklog > readio_log) { |
853 | mp->m_readio_log = sbp->sb_blocklog; | |
854 | } else { | |
855 | mp->m_readio_log = readio_log; | |
856 | } | |
857 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
858 | if (sbp->sb_blocklog > writeio_log) { | |
859 | mp->m_writeio_log = sbp->sb_blocklog; | |
860 | } else { | |
861 | mp->m_writeio_log = writeio_log; | |
862 | } | |
863 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
0771fb45 | 864 | } |
1da177e4 | 865 | |
0771fb45 ES |
866 | /* |
867 | * Set whether we're using inode alignment. | |
868 | */ | |
869 | STATIC void | |
870 | xfs_set_inoalignment(xfs_mount_t *mp) | |
871 | { | |
62118709 | 872 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
1da177e4 LT |
873 | mp->m_sb.sb_inoalignmt >= |
874 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
875 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; | |
876 | else | |
877 | mp->m_inoalign_mask = 0; | |
878 | /* | |
879 | * If we are using stripe alignment, check whether | |
880 | * the stripe unit is a multiple of the inode alignment | |
881 | */ | |
882 | if (mp->m_dalign && mp->m_inoalign_mask && | |
883 | !(mp->m_dalign & mp->m_inoalign_mask)) | |
884 | mp->m_sinoalign = mp->m_dalign; | |
885 | else | |
886 | mp->m_sinoalign = 0; | |
0771fb45 ES |
887 | } |
888 | ||
889 | /* | |
890 | * Check that the data (and log if separate) are an ok size. | |
891 | */ | |
892 | STATIC int | |
893 | xfs_check_sizes(xfs_mount_t *mp, int mfsi_flags) | |
894 | { | |
895 | xfs_buf_t *bp; | |
896 | xfs_daddr_t d; | |
897 | int error; | |
898 | ||
1da177e4 LT |
899 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
900 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
901 | cmn_err(CE_WARN, "XFS: size check 1 failed"); | |
0771fb45 | 902 | return XFS_ERROR(E2BIG); |
1da177e4 LT |
903 | } |
904 | error = xfs_read_buf(mp, mp->m_ddev_targp, | |
905 | d - XFS_FSS_TO_BB(mp, 1), | |
906 | XFS_FSS_TO_BB(mp, 1), 0, &bp); | |
907 | if (!error) { | |
908 | xfs_buf_relse(bp); | |
909 | } else { | |
910 | cmn_err(CE_WARN, "XFS: size check 2 failed"); | |
0771fb45 | 911 | if (error == ENOSPC) |
1da177e4 | 912 | error = XFS_ERROR(E2BIG); |
0771fb45 | 913 | return error; |
1da177e4 LT |
914 | } |
915 | ||
916 | if (((mfsi_flags & XFS_MFSI_CLIENT) == 0) && | |
917 | mp->m_logdev_targp != mp->m_ddev_targp) { | |
918 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); | |
919 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
920 | cmn_err(CE_WARN, "XFS: size check 3 failed"); | |
0771fb45 | 921 | return XFS_ERROR(E2BIG); |
1da177e4 LT |
922 | } |
923 | error = xfs_read_buf(mp, mp->m_logdev_targp, | |
924 | d - XFS_FSB_TO_BB(mp, 1), | |
925 | XFS_FSB_TO_BB(mp, 1), 0, &bp); | |
926 | if (!error) { | |
927 | xfs_buf_relse(bp); | |
928 | } else { | |
929 | cmn_err(CE_WARN, "XFS: size check 3 failed"); | |
0771fb45 | 930 | if (error == ENOSPC) |
1da177e4 | 931 | error = XFS_ERROR(E2BIG); |
0771fb45 ES |
932 | return error; |
933 | } | |
934 | } | |
935 | return 0; | |
936 | } | |
937 | ||
938 | /* | |
939 | * xfs_mountfs | |
940 | * | |
941 | * This function does the following on an initial mount of a file system: | |
942 | * - reads the superblock from disk and init the mount struct | |
943 | * - if we're a 32-bit kernel, do a size check on the superblock | |
944 | * so we don't mount terabyte filesystems | |
945 | * - init mount struct realtime fields | |
946 | * - allocate inode hash table for fs | |
947 | * - init directory manager | |
948 | * - perform recovery and init the log manager | |
949 | */ | |
950 | int | |
951 | xfs_mountfs( | |
952 | xfs_mount_t *mp, | |
953 | int mfsi_flags) | |
954 | { | |
955 | xfs_sb_t *sbp = &(mp->m_sb); | |
956 | xfs_inode_t *rip; | |
0771fb45 ES |
957 | __uint64_t resblks; |
958 | __int64_t update_flags = 0LL; | |
959 | uint quotamount, quotaflags; | |
960 | int agno; | |
961 | int uuid_mounted = 0; | |
962 | int error = 0; | |
963 | ||
0771fb45 ES |
964 | xfs_mount_common(mp, sbp); |
965 | ||
ee1c0908 | 966 | /* |
e6957ea4 ES |
967 | * Check for a mismatched features2 values. Older kernels |
968 | * read & wrote into the wrong sb offset for sb_features2 | |
969 | * on some platforms due to xfs_sb_t not being 64bit size aligned | |
970 | * when sb_features2 was added, which made older superblock | |
971 | * reading/writing routines swap it as a 64-bit value. | |
ee1c0908 | 972 | * |
e6957ea4 ES |
973 | * For backwards compatibility, we make both slots equal. |
974 | * | |
975 | * If we detect a mismatched field, we OR the set bits into the | |
976 | * existing features2 field in case it has already been modified; we | |
977 | * don't want to lose any features. We then update the bad location | |
978 | * with the ORed value so that older kernels will see any features2 | |
979 | * flags, and mark the two fields as needing updates once the | |
980 | * transaction subsystem is online. | |
ee1c0908 | 981 | */ |
e6957ea4 | 982 | if (xfs_sb_has_mismatched_features2(sbp)) { |
ee1c0908 DC |
983 | cmn_err(CE_WARN, |
984 | "XFS: correcting sb_features alignment problem"); | |
985 | sbp->sb_features2 |= sbp->sb_bad_features2; | |
e6957ea4 | 986 | sbp->sb_bad_features2 = sbp->sb_features2; |
ee1c0908 | 987 | update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2; |
e6957ea4 ES |
988 | |
989 | /* | |
990 | * Re-check for ATTR2 in case it was found in bad_features2 | |
991 | * slot. | |
992 | */ | |
7c12f296 TS |
993 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
994 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 995 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
996 | } |
997 | ||
998 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
999 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
1000 | xfs_sb_version_removeattr2(&mp->m_sb); | |
1001 | update_flags |= XFS_SB_FEATURES2; | |
e6957ea4 | 1002 | |
7c12f296 TS |
1003 | /* update sb_versionnum for the clearing of the morebits */ |
1004 | if (!sbp->sb_features2) | |
1005 | update_flags |= XFS_SB_VERSIONNUM; | |
ee1c0908 DC |
1006 | } |
1007 | ||
0771fb45 ES |
1008 | /* |
1009 | * Check if sb_agblocks is aligned at stripe boundary | |
1010 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
1011 | * allocator alignment is within an ag, therefore ag has | |
1012 | * to be aligned at stripe boundary. | |
1013 | */ | |
1014 | error = xfs_update_alignment(mp, mfsi_flags, &update_flags); | |
1015 | if (error) | |
1016 | goto error1; | |
1017 | ||
1018 | xfs_alloc_compute_maxlevels(mp); | |
1019 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
1020 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
1021 | xfs_ialloc_compute_maxlevels(mp); | |
1022 | ||
1023 | xfs_set_maxicount(mp); | |
1024 | ||
1025 | mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog); | |
1026 | ||
1027 | /* | |
1028 | * XFS uses the uuid from the superblock as the unique | |
1029 | * identifier for fsid. We can not use the uuid from the volume | |
1030 | * since a single partition filesystem is identical to a single | |
1031 | * partition volume/filesystem. | |
1032 | */ | |
1033 | if ((mfsi_flags & XFS_MFSI_SECOND) == 0 && | |
1034 | (mp->m_flags & XFS_MOUNT_NOUUID) == 0) { | |
1035 | if (xfs_uuid_mount(mp)) { | |
1036 | error = XFS_ERROR(EINVAL); | |
1da177e4 LT |
1037 | goto error1; |
1038 | } | |
0771fb45 | 1039 | uuid_mounted=1; |
1da177e4 LT |
1040 | } |
1041 | ||
0771fb45 ES |
1042 | /* |
1043 | * Set the minimum read and write sizes | |
1044 | */ | |
1045 | xfs_set_rw_sizes(mp); | |
1046 | ||
1047 | /* | |
1048 | * Set the inode cluster size. | |
1049 | * This may still be overridden by the file system | |
1050 | * block size if it is larger than the chosen cluster size. | |
1051 | */ | |
1052 | mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; | |
1053 | ||
1054 | /* | |
1055 | * Set inode alignment fields | |
1056 | */ | |
1057 | xfs_set_inoalignment(mp); | |
1058 | ||
1059 | /* | |
1060 | * Check that the data (and log if separate) are an ok size. | |
1061 | */ | |
1062 | error = xfs_check_sizes(mp, mfsi_flags); | |
1063 | if (error) | |
1064 | goto error1; | |
1065 | ||
1da177e4 LT |
1066 | /* |
1067 | * Initialize realtime fields in the mount structure | |
1068 | */ | |
0771fb45 ES |
1069 | error = xfs_rtmount_init(mp); |
1070 | if (error) { | |
1da177e4 LT |
1071 | cmn_err(CE_WARN, "XFS: RT mount failed"); |
1072 | goto error1; | |
1073 | } | |
1074 | ||
1075 | /* | |
1076 | * For client case we are done now | |
1077 | */ | |
1078 | if (mfsi_flags & XFS_MFSI_CLIENT) { | |
014c2544 | 1079 | return 0; |
1da177e4 LT |
1080 | } |
1081 | ||
1082 | /* | |
1083 | * Copies the low order bits of the timestamp and the randomly | |
1084 | * set "sequence" number out of a UUID. | |
1085 | */ | |
1086 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); | |
1087 | ||
1da177e4 LT |
1088 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ |
1089 | ||
f6c2d1fa | 1090 | xfs_dir_mount(mp); |
1da177e4 LT |
1091 | |
1092 | /* | |
1093 | * Initialize the attribute manager's entries. | |
1094 | */ | |
1095 | mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100; | |
1096 | ||
1097 | /* | |
1098 | * Initialize the precomputed transaction reservations values. | |
1099 | */ | |
1100 | xfs_trans_init(mp); | |
1101 | ||
1da177e4 LT |
1102 | /* |
1103 | * Allocate and initialize the per-ag data. | |
1104 | */ | |
1105 | init_rwsem(&mp->m_peraglock); | |
1106 | mp->m_perag = | |
1107 | kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t), KM_SLEEP); | |
1108 | ||
b267ce99 | 1109 | mp->m_maxagi = xfs_initialize_perag(mp, sbp->sb_agcount); |
1da177e4 LT |
1110 | |
1111 | /* | |
1112 | * log's mount-time initialization. Perform 1st part recovery if needed | |
1113 | */ | |
1114 | if (likely(sbp->sb_logblocks > 0)) { /* check for volume case */ | |
1115 | error = xfs_log_mount(mp, mp->m_logdev_targp, | |
1116 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
1117 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
1118 | if (error) { | |
1119 | cmn_err(CE_WARN, "XFS: log mount failed"); | |
1120 | goto error2; | |
1121 | } | |
1122 | } else { /* No log has been defined */ | |
1123 | cmn_err(CE_WARN, "XFS: no log defined"); | |
1124 | XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW, mp); | |
1125 | error = XFS_ERROR(EFSCORRUPTED); | |
1126 | goto error2; | |
1127 | } | |
1128 | ||
92821e2b DC |
1129 | /* |
1130 | * Now the log is mounted, we know if it was an unclean shutdown or | |
1131 | * not. If it was, with the first phase of recovery has completed, we | |
1132 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
1133 | * but they are recovered transactionally in the second recovery phase | |
1134 | * later. | |
1135 | * | |
1136 | * Hence we can safely re-initialise incore superblock counters from | |
1137 | * the per-ag data. These may not be correct if the filesystem was not | |
1138 | * cleanly unmounted, so we need to wait for recovery to finish before | |
1139 | * doing this. | |
1140 | * | |
1141 | * If the filesystem was cleanly unmounted, then we can trust the | |
1142 | * values in the superblock to be correct and we don't need to do | |
1143 | * anything here. | |
1144 | * | |
1145 | * If we are currently making the filesystem, the initialisation will | |
1146 | * fail as the perag data is in an undefined state. | |
1147 | */ | |
1148 | ||
1149 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && | |
1150 | !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
1151 | !mp->m_sb.sb_inprogress) { | |
1152 | error = xfs_initialize_perag_data(mp, sbp->sb_agcount); | |
1153 | if (error) { | |
1154 | goto error2; | |
1155 | } | |
1156 | } | |
1da177e4 LT |
1157 | /* |
1158 | * Get and sanity-check the root inode. | |
1159 | * Save the pointer to it in the mount structure. | |
1160 | */ | |
1161 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0); | |
1162 | if (error) { | |
1163 | cmn_err(CE_WARN, "XFS: failed to read root inode"); | |
1164 | goto error3; | |
1165 | } | |
1166 | ||
1167 | ASSERT(rip != NULL); | |
1da177e4 LT |
1168 | |
1169 | if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) { | |
1170 | cmn_err(CE_WARN, "XFS: corrupted root inode"); | |
b6574520 NS |
1171 | cmn_err(CE_WARN, "Device %s - root %llu is not a directory", |
1172 | XFS_BUFTARG_NAME(mp->m_ddev_targp), | |
1173 | (unsigned long long)rip->i_ino); | |
1da177e4 LT |
1174 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
1175 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
1176 | mp); | |
1177 | error = XFS_ERROR(EFSCORRUPTED); | |
1178 | goto error4; | |
1179 | } | |
1180 | mp->m_rootip = rip; /* save it */ | |
1181 | ||
1182 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
1183 | ||
1184 | /* | |
1185 | * Initialize realtime inode pointers in the mount structure | |
1186 | */ | |
0771fb45 ES |
1187 | error = xfs_rtmount_inodes(mp); |
1188 | if (error) { | |
1da177e4 LT |
1189 | /* |
1190 | * Free up the root inode. | |
1191 | */ | |
1192 | cmn_err(CE_WARN, "XFS: failed to read RT inodes"); | |
1193 | goto error4; | |
1194 | } | |
1195 | ||
1196 | /* | |
ee1c0908 | 1197 | * If fs is not mounted readonly, then update the superblock changes. |
1da177e4 | 1198 | */ |
e5720eec DC |
1199 | if (update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
1200 | error = xfs_mount_log_sb(mp, update_flags); | |
1201 | if (error) { | |
1202 | cmn_err(CE_WARN, "XFS: failed to write sb changes"); | |
1203 | goto error4; | |
1204 | } | |
1205 | } | |
1da177e4 LT |
1206 | |
1207 | /* | |
1208 | * Initialise the XFS quota management subsystem for this mount | |
1209 | */ | |
0771fb45 ES |
1210 | error = XFS_QM_INIT(mp, "amount, "aflags); |
1211 | if (error) | |
1da177e4 LT |
1212 | goto error4; |
1213 | ||
1214 | /* | |
1215 | * Finish recovering the file system. This part needed to be | |
1216 | * delayed until after the root and real-time bitmap inodes | |
1217 | * were consistently read in. | |
1218 | */ | |
1219 | error = xfs_log_mount_finish(mp, mfsi_flags); | |
1220 | if (error) { | |
1221 | cmn_err(CE_WARN, "XFS: log mount finish failed"); | |
1222 | goto error4; | |
1223 | } | |
1224 | ||
1225 | /* | |
1226 | * Complete the quota initialisation, post-log-replay component. | |
1227 | */ | |
0771fb45 ES |
1228 | error = XFS_QM_MOUNT(mp, quotamount, quotaflags, mfsi_flags); |
1229 | if (error) | |
1da177e4 LT |
1230 | goto error4; |
1231 | ||
84e1e99f DC |
1232 | /* |
1233 | * Now we are mounted, reserve a small amount of unused space for | |
1234 | * privileged transactions. This is needed so that transaction | |
1235 | * space required for critical operations can dip into this pool | |
1236 | * when at ENOSPC. This is needed for operations like create with | |
1237 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
1238 | * are not allowed to use this reserved space. | |
1239 | * | |
1240 | * We default to 5% or 1024 fsbs of space reserved, whichever is smaller. | |
1241 | * This may drive us straight to ENOSPC on mount, but that implies | |
714082bc | 1242 | * we were already there on the last unmount. Warn if this occurs. |
84e1e99f | 1243 | */ |
39726be2 CH |
1244 | resblks = mp->m_sb.sb_dblocks; |
1245 | do_div(resblks, 20); | |
1246 | resblks = min_t(__uint64_t, resblks, 1024); | |
714082bc DC |
1247 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1248 | if (error) | |
1249 | cmn_err(CE_WARN, "XFS: Unable to allocate reserve blocks. " | |
1250 | "Continuing without a reserve pool."); | |
84e1e99f | 1251 | |
1da177e4 LT |
1252 | return 0; |
1253 | ||
1254 | error4: | |
1255 | /* | |
1256 | * Free up the root inode. | |
1257 | */ | |
43355099 | 1258 | IRELE(rip); |
1da177e4 LT |
1259 | error3: |
1260 | xfs_log_unmount_dealloc(mp); | |
1261 | error2: | |
1da177e4 LT |
1262 | for (agno = 0; agno < sbp->sb_agcount; agno++) |
1263 | if (mp->m_perag[agno].pagb_list) | |
f0e2d93c DV |
1264 | kmem_free(mp->m_perag[agno].pagb_list); |
1265 | kmem_free(mp->m_perag); | |
1da177e4 LT |
1266 | mp->m_perag = NULL; |
1267 | /* FALLTHROUGH */ | |
1268 | error1: | |
1269 | if (uuid_mounted) | |
fa6adbe0 | 1270 | uuid_table_remove(&mp->m_sb.sb_uuid); |
1da177e4 LT |
1271 | return error; |
1272 | } | |
1273 | ||
1274 | /* | |
1275 | * xfs_unmountfs | |
1276 | * | |
1277 | * This flushes out the inodes,dquots and the superblock, unmounts the | |
1278 | * log and makes sure that incore structures are freed. | |
1279 | */ | |
1280 | int | |
19f354d4 | 1281 | xfs_unmountfs(xfs_mount_t *mp) |
1da177e4 | 1282 | { |
84e1e99f | 1283 | __uint64_t resblks; |
714082bc | 1284 | int error = 0; |
1da177e4 | 1285 | |
641c56fb DC |
1286 | /* |
1287 | * We can potentially deadlock here if we have an inode cluster | |
1288 | * that has been freed has it's buffer still pinned in memory because | |
1289 | * the transaction is still sitting in a iclog. The stale inodes | |
1290 | * on that buffer will have their flush locks held until the | |
1291 | * transaction hits the disk and the callbacks run. the inode | |
1292 | * flush takes the flush lock unconditionally and with nothing to | |
1293 | * push out the iclog we will never get that unlocked. hence we | |
1294 | * need to force the log first. | |
1295 | */ | |
1296 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC); | |
efa80278 | 1297 | xfs_iflush_all(mp); |
1da177e4 | 1298 | |
ee2a4f7c | 1299 | XFS_QM_DQPURGEALL(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_UMOUNTING); |
1da177e4 LT |
1300 | |
1301 | /* | |
1302 | * Flush out the log synchronously so that we know for sure | |
1303 | * that nothing is pinned. This is important because bflush() | |
1304 | * will skip pinned buffers. | |
1305 | */ | |
1306 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC); | |
1307 | ||
1308 | xfs_binval(mp->m_ddev_targp); | |
1309 | if (mp->m_rtdev_targp) { | |
1310 | xfs_binval(mp->m_rtdev_targp); | |
1311 | } | |
1312 | ||
84e1e99f DC |
1313 | /* |
1314 | * Unreserve any blocks we have so that when we unmount we don't account | |
1315 | * the reserved free space as used. This is really only necessary for | |
1316 | * lazy superblock counting because it trusts the incore superblock | |
1317 | * counters to be aboslutely correct on clean unmount. | |
1318 | * | |
1319 | * We don't bother correcting this elsewhere for lazy superblock | |
1320 | * counting because on mount of an unclean filesystem we reconstruct the | |
1321 | * correct counter value and this is irrelevant. | |
1322 | * | |
1323 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1324 | * we only every apply deltas to the superblock and hence the incore | |
1325 | * value does not matter.... | |
1326 | */ | |
1327 | resblks = 0; | |
714082bc DC |
1328 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1329 | if (error) | |
1330 | cmn_err(CE_WARN, "XFS: Unable to free reserved block pool. " | |
1331 | "Freespace may not be correct on next mount."); | |
1332 | ||
e5720eec DC |
1333 | error = xfs_log_sbcount(mp, 1); |
1334 | if (error) | |
1335 | cmn_err(CE_WARN, "XFS: Unable to update superblock counters. " | |
1336 | "Freespace may not be correct on next mount."); | |
1da177e4 | 1337 | xfs_unmountfs_writesb(mp); |
1da177e4 | 1338 | xfs_unmountfs_wait(mp); /* wait for async bufs */ |
1da177e4 LT |
1339 | xfs_log_unmount(mp); /* Done! No more fs ops. */ |
1340 | ||
1341 | xfs_freesb(mp); | |
1342 | ||
1343 | /* | |
1344 | * All inodes from this mount point should be freed. | |
1345 | */ | |
1346 | ASSERT(mp->m_inodes == NULL); | |
1347 | ||
1da177e4 | 1348 | if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0) |
fa6adbe0 | 1349 | uuid_table_remove(&mp->m_sb.sb_uuid); |
1da177e4 LT |
1350 | |
1351 | #if defined(DEBUG) || defined(INDUCE_IO_ERROR) | |
0ce4cfd4 | 1352 | xfs_errortag_clearall(mp, 0); |
1da177e4 | 1353 | #endif |
745f6919 | 1354 | xfs_mount_free(mp); |
1da177e4 LT |
1355 | return 0; |
1356 | } | |
1357 | ||
ba0f32d4 | 1358 | STATIC void |
1da177e4 LT |
1359 | xfs_unmountfs_wait(xfs_mount_t *mp) |
1360 | { | |
1361 | if (mp->m_logdev_targp != mp->m_ddev_targp) | |
1362 | xfs_wait_buftarg(mp->m_logdev_targp); | |
1363 | if (mp->m_rtdev_targp) | |
1364 | xfs_wait_buftarg(mp->m_rtdev_targp); | |
1365 | xfs_wait_buftarg(mp->m_ddev_targp); | |
1366 | } | |
1367 | ||
92821e2b DC |
1368 | int |
1369 | xfs_fs_writable(xfs_mount_t *mp) | |
1370 | { | |
b267ce99 | 1371 | return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) || |
bd186aa9 | 1372 | (mp->m_flags & XFS_MOUNT_RDONLY)); |
92821e2b DC |
1373 | } |
1374 | ||
1375 | /* | |
1376 | * xfs_log_sbcount | |
1377 | * | |
1378 | * Called either periodically to keep the on disk superblock values | |
1379 | * roughly up to date or from unmount to make sure the values are | |
1380 | * correct on a clean unmount. | |
1381 | * | |
1382 | * Note this code can be called during the process of freezing, so | |
1383 | * we may need to use the transaction allocator which does not not | |
1384 | * block when the transaction subsystem is in its frozen state. | |
1385 | */ | |
1386 | int | |
1387 | xfs_log_sbcount( | |
1388 | xfs_mount_t *mp, | |
1389 | uint sync) | |
1390 | { | |
1391 | xfs_trans_t *tp; | |
1392 | int error; | |
1393 | ||
1394 | if (!xfs_fs_writable(mp)) | |
1395 | return 0; | |
1396 | ||
d4d90b57 | 1397 | xfs_icsb_sync_counters(mp, 0); |
92821e2b DC |
1398 | |
1399 | /* | |
1400 | * we don't need to do this if we are updating the superblock | |
1401 | * counters on every modification. | |
1402 | */ | |
1403 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1404 | return 0; | |
1405 | ||
1406 | tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT); | |
1407 | error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0, | |
1408 | XFS_DEFAULT_LOG_COUNT); | |
1409 | if (error) { | |
1410 | xfs_trans_cancel(tp, 0); | |
1411 | return error; | |
1412 | } | |
1413 | ||
1414 | xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS); | |
1415 | if (sync) | |
1416 | xfs_trans_set_sync(tp); | |
e5720eec DC |
1417 | error = xfs_trans_commit(tp, 0); |
1418 | return error; | |
92821e2b DC |
1419 | } |
1420 | ||
2bdf7cd0 CH |
1421 | STATIC void |
1422 | xfs_mark_shared_ro( | |
1423 | xfs_mount_t *mp, | |
1424 | xfs_buf_t *bp) | |
1425 | { | |
1426 | xfs_dsb_t *sb = XFS_BUF_TO_SBP(bp); | |
1427 | __uint16_t version; | |
1428 | ||
1429 | if (!(sb->sb_flags & XFS_SBF_READONLY)) | |
1430 | sb->sb_flags |= XFS_SBF_READONLY; | |
1431 | ||
1432 | version = be16_to_cpu(sb->sb_versionnum); | |
1433 | if ((version & XFS_SB_VERSION_NUMBITS) != XFS_SB_VERSION_4 || | |
1434 | !(version & XFS_SB_VERSION_SHAREDBIT)) | |
1435 | version |= XFS_SB_VERSION_SHAREDBIT; | |
1436 | sb->sb_versionnum = cpu_to_be16(version); | |
1437 | } | |
1438 | ||
1da177e4 LT |
1439 | int |
1440 | xfs_unmountfs_writesb(xfs_mount_t *mp) | |
1441 | { | |
1442 | xfs_buf_t *sbp; | |
1da177e4 LT |
1443 | int error = 0; |
1444 | ||
1445 | /* | |
1446 | * skip superblock write if fs is read-only, or | |
1447 | * if we are doing a forced umount. | |
1448 | */ | |
bd186aa9 | 1449 | if (!((mp->m_flags & XFS_MOUNT_RDONLY) || |
1da177e4 | 1450 | XFS_FORCED_SHUTDOWN(mp))) { |
8d280b98 | 1451 | |
92821e2b | 1452 | sbp = xfs_getsb(mp, 0); |
8d280b98 | 1453 | |
1da177e4 LT |
1454 | /* |
1455 | * mark shared-readonly if desired | |
1456 | */ | |
2bdf7cd0 CH |
1457 | if (mp->m_mk_sharedro) |
1458 | xfs_mark_shared_ro(mp, sbp); | |
92821e2b | 1459 | |
1da177e4 LT |
1460 | XFS_BUF_UNDONE(sbp); |
1461 | XFS_BUF_UNREAD(sbp); | |
1462 | XFS_BUF_UNDELAYWRITE(sbp); | |
1463 | XFS_BUF_WRITE(sbp); | |
1464 | XFS_BUF_UNASYNC(sbp); | |
1465 | ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp); | |
1466 | xfsbdstrat(mp, sbp); | |
1da177e4 LT |
1467 | error = xfs_iowait(sbp); |
1468 | if (error) | |
1469 | xfs_ioerror_alert("xfs_unmountfs_writesb", | |
1470 | mp, sbp, XFS_BUF_ADDR(sbp)); | |
1471 | if (error && mp->m_mk_sharedro) | |
1472 | xfs_fs_cmn_err(CE_ALERT, mp, "Superblock write error detected while unmounting. Filesystem may not be marked shared readonly"); | |
92821e2b | 1473 | xfs_buf_relse(sbp); |
1da177e4 | 1474 | } |
014c2544 | 1475 | return error; |
1da177e4 LT |
1476 | } |
1477 | ||
1478 | /* | |
1479 | * xfs_mod_sb() can be used to copy arbitrary changes to the | |
1480 | * in-core superblock into the superblock buffer to be logged. | |
1481 | * It does not provide the higher level of locking that is | |
1482 | * needed to protect the in-core superblock from concurrent | |
1483 | * access. | |
1484 | */ | |
1485 | void | |
1486 | xfs_mod_sb(xfs_trans_t *tp, __int64_t fields) | |
1487 | { | |
1488 | xfs_buf_t *bp; | |
1489 | int first; | |
1490 | int last; | |
1491 | xfs_mount_t *mp; | |
1da177e4 LT |
1492 | xfs_sb_field_t f; |
1493 | ||
1494 | ASSERT(fields); | |
1495 | if (!fields) | |
1496 | return; | |
1497 | mp = tp->t_mountp; | |
1498 | bp = xfs_trans_getsb(tp, mp, 0); | |
1da177e4 LT |
1499 | first = sizeof(xfs_sb_t); |
1500 | last = 0; | |
1501 | ||
1502 | /* translate/copy */ | |
1503 | ||
2bdf7cd0 | 1504 | xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields); |
1da177e4 LT |
1505 | |
1506 | /* find modified range */ | |
1507 | ||
1508 | f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields); | |
1509 | ASSERT((1LL << f) & XFS_SB_MOD_BITS); | |
1510 | first = xfs_sb_info[f].offset; | |
1511 | ||
1512 | f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields); | |
1513 | ASSERT((1LL << f) & XFS_SB_MOD_BITS); | |
1514 | last = xfs_sb_info[f + 1].offset - 1; | |
1515 | ||
1516 | xfs_trans_log_buf(tp, bp, first, last); | |
1517 | } | |
d210a28c | 1518 | |
d210a28c | 1519 | |
1da177e4 LT |
1520 | /* |
1521 | * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply | |
1522 | * a delta to a specified field in the in-core superblock. Simply | |
1523 | * switch on the field indicated and apply the delta to that field. | |
1524 | * Fields are not allowed to dip below zero, so if the delta would | |
1525 | * do this do not apply it and return EINVAL. | |
1526 | * | |
3685c2a1 | 1527 | * The m_sb_lock must be held when this routine is called. |
1da177e4 | 1528 | */ |
8d280b98 | 1529 | int |
20f4ebf2 DC |
1530 | xfs_mod_incore_sb_unlocked( |
1531 | xfs_mount_t *mp, | |
1532 | xfs_sb_field_t field, | |
1533 | int64_t delta, | |
1534 | int rsvd) | |
1da177e4 LT |
1535 | { |
1536 | int scounter; /* short counter for 32 bit fields */ | |
1537 | long long lcounter; /* long counter for 64 bit fields */ | |
1538 | long long res_used, rem; | |
1539 | ||
1540 | /* | |
1541 | * With the in-core superblock spin lock held, switch | |
1542 | * on the indicated field. Apply the delta to the | |
1543 | * proper field. If the fields value would dip below | |
1544 | * 0, then do not apply the delta and return EINVAL. | |
1545 | */ | |
1546 | switch (field) { | |
1547 | case XFS_SBS_ICOUNT: | |
1548 | lcounter = (long long)mp->m_sb.sb_icount; | |
1549 | lcounter += delta; | |
1550 | if (lcounter < 0) { | |
1551 | ASSERT(0); | |
014c2544 | 1552 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1553 | } |
1554 | mp->m_sb.sb_icount = lcounter; | |
014c2544 | 1555 | return 0; |
1da177e4 LT |
1556 | case XFS_SBS_IFREE: |
1557 | lcounter = (long long)mp->m_sb.sb_ifree; | |
1558 | lcounter += delta; | |
1559 | if (lcounter < 0) { | |
1560 | ASSERT(0); | |
014c2544 | 1561 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1562 | } |
1563 | mp->m_sb.sb_ifree = lcounter; | |
014c2544 | 1564 | return 0; |
1da177e4 | 1565 | case XFS_SBS_FDBLOCKS: |
4be536de DC |
1566 | lcounter = (long long) |
1567 | mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); | |
1da177e4 LT |
1568 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); |
1569 | ||
1570 | if (delta > 0) { /* Putting blocks back */ | |
1571 | if (res_used > delta) { | |
1572 | mp->m_resblks_avail += delta; | |
1573 | } else { | |
1574 | rem = delta - res_used; | |
1575 | mp->m_resblks_avail = mp->m_resblks; | |
1576 | lcounter += rem; | |
1577 | } | |
1578 | } else { /* Taking blocks away */ | |
1579 | ||
1580 | lcounter += delta; | |
1581 | ||
1582 | /* | |
1583 | * If were out of blocks, use any available reserved blocks if | |
1584 | * were allowed to. | |
1585 | */ | |
1586 | ||
1587 | if (lcounter < 0) { | |
1588 | if (rsvd) { | |
1589 | lcounter = (long long)mp->m_resblks_avail + delta; | |
1590 | if (lcounter < 0) { | |
014c2544 | 1591 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1592 | } |
1593 | mp->m_resblks_avail = lcounter; | |
014c2544 | 1594 | return 0; |
1da177e4 | 1595 | } else { /* not reserved */ |
014c2544 | 1596 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1597 | } |
1598 | } | |
1599 | } | |
1600 | ||
4be536de | 1601 | mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
014c2544 | 1602 | return 0; |
1da177e4 LT |
1603 | case XFS_SBS_FREXTENTS: |
1604 | lcounter = (long long)mp->m_sb.sb_frextents; | |
1605 | lcounter += delta; | |
1606 | if (lcounter < 0) { | |
014c2544 | 1607 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1608 | } |
1609 | mp->m_sb.sb_frextents = lcounter; | |
014c2544 | 1610 | return 0; |
1da177e4 LT |
1611 | case XFS_SBS_DBLOCKS: |
1612 | lcounter = (long long)mp->m_sb.sb_dblocks; | |
1613 | lcounter += delta; | |
1614 | if (lcounter < 0) { | |
1615 | ASSERT(0); | |
014c2544 | 1616 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1617 | } |
1618 | mp->m_sb.sb_dblocks = lcounter; | |
014c2544 | 1619 | return 0; |
1da177e4 LT |
1620 | case XFS_SBS_AGCOUNT: |
1621 | scounter = mp->m_sb.sb_agcount; | |
1622 | scounter += delta; | |
1623 | if (scounter < 0) { | |
1624 | ASSERT(0); | |
014c2544 | 1625 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1626 | } |
1627 | mp->m_sb.sb_agcount = scounter; | |
014c2544 | 1628 | return 0; |
1da177e4 LT |
1629 | case XFS_SBS_IMAX_PCT: |
1630 | scounter = mp->m_sb.sb_imax_pct; | |
1631 | scounter += delta; | |
1632 | if (scounter < 0) { | |
1633 | ASSERT(0); | |
014c2544 | 1634 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1635 | } |
1636 | mp->m_sb.sb_imax_pct = scounter; | |
014c2544 | 1637 | return 0; |
1da177e4 LT |
1638 | case XFS_SBS_REXTSIZE: |
1639 | scounter = mp->m_sb.sb_rextsize; | |
1640 | scounter += delta; | |
1641 | if (scounter < 0) { | |
1642 | ASSERT(0); | |
014c2544 | 1643 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1644 | } |
1645 | mp->m_sb.sb_rextsize = scounter; | |
014c2544 | 1646 | return 0; |
1da177e4 LT |
1647 | case XFS_SBS_RBMBLOCKS: |
1648 | scounter = mp->m_sb.sb_rbmblocks; | |
1649 | scounter += delta; | |
1650 | if (scounter < 0) { | |
1651 | ASSERT(0); | |
014c2544 | 1652 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1653 | } |
1654 | mp->m_sb.sb_rbmblocks = scounter; | |
014c2544 | 1655 | return 0; |
1da177e4 LT |
1656 | case XFS_SBS_RBLOCKS: |
1657 | lcounter = (long long)mp->m_sb.sb_rblocks; | |
1658 | lcounter += delta; | |
1659 | if (lcounter < 0) { | |
1660 | ASSERT(0); | |
014c2544 | 1661 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1662 | } |
1663 | mp->m_sb.sb_rblocks = lcounter; | |
014c2544 | 1664 | return 0; |
1da177e4 LT |
1665 | case XFS_SBS_REXTENTS: |
1666 | lcounter = (long long)mp->m_sb.sb_rextents; | |
1667 | lcounter += delta; | |
1668 | if (lcounter < 0) { | |
1669 | ASSERT(0); | |
014c2544 | 1670 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1671 | } |
1672 | mp->m_sb.sb_rextents = lcounter; | |
014c2544 | 1673 | return 0; |
1da177e4 LT |
1674 | case XFS_SBS_REXTSLOG: |
1675 | scounter = mp->m_sb.sb_rextslog; | |
1676 | scounter += delta; | |
1677 | if (scounter < 0) { | |
1678 | ASSERT(0); | |
014c2544 | 1679 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1680 | } |
1681 | mp->m_sb.sb_rextslog = scounter; | |
014c2544 | 1682 | return 0; |
1da177e4 LT |
1683 | default: |
1684 | ASSERT(0); | |
014c2544 | 1685 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1686 | } |
1687 | } | |
1688 | ||
1689 | /* | |
1690 | * xfs_mod_incore_sb() is used to change a field in the in-core | |
1691 | * superblock structure by the specified delta. This modification | |
3685c2a1 | 1692 | * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked() |
1da177e4 LT |
1693 | * routine to do the work. |
1694 | */ | |
1695 | int | |
20f4ebf2 DC |
1696 | xfs_mod_incore_sb( |
1697 | xfs_mount_t *mp, | |
1698 | xfs_sb_field_t field, | |
1699 | int64_t delta, | |
1700 | int rsvd) | |
1da177e4 | 1701 | { |
1da177e4 LT |
1702 | int status; |
1703 | ||
8d280b98 DC |
1704 | /* check for per-cpu counters */ |
1705 | switch (field) { | |
1706 | #ifdef HAVE_PERCPU_SB | |
1707 | case XFS_SBS_ICOUNT: | |
1708 | case XFS_SBS_IFREE: | |
1709 | case XFS_SBS_FDBLOCKS: | |
1710 | if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) { | |
1711 | status = xfs_icsb_modify_counters(mp, field, | |
1712 | delta, rsvd); | |
1713 | break; | |
1714 | } | |
1715 | /* FALLTHROUGH */ | |
1716 | #endif | |
1717 | default: | |
3685c2a1 | 1718 | spin_lock(&mp->m_sb_lock); |
8d280b98 | 1719 | status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); |
3685c2a1 | 1720 | spin_unlock(&mp->m_sb_lock); |
8d280b98 DC |
1721 | break; |
1722 | } | |
1723 | ||
014c2544 | 1724 | return status; |
1da177e4 LT |
1725 | } |
1726 | ||
1727 | /* | |
1728 | * xfs_mod_incore_sb_batch() is used to change more than one field | |
1729 | * in the in-core superblock structure at a time. This modification | |
1730 | * is protected by a lock internal to this module. The fields and | |
1731 | * changes to those fields are specified in the array of xfs_mod_sb | |
1732 | * structures passed in. | |
1733 | * | |
1734 | * Either all of the specified deltas will be applied or none of | |
1735 | * them will. If any modified field dips below 0, then all modifications | |
1736 | * will be backed out and EINVAL will be returned. | |
1737 | */ | |
1738 | int | |
1739 | xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd) | |
1740 | { | |
1da177e4 LT |
1741 | int status=0; |
1742 | xfs_mod_sb_t *msbp; | |
1743 | ||
1744 | /* | |
1745 | * Loop through the array of mod structures and apply each | |
1746 | * individually. If any fail, then back out all those | |
1747 | * which have already been applied. Do all of this within | |
3685c2a1 | 1748 | * the scope of the m_sb_lock so that all of the changes will |
1da177e4 LT |
1749 | * be atomic. |
1750 | */ | |
3685c2a1 | 1751 | spin_lock(&mp->m_sb_lock); |
1da177e4 LT |
1752 | msbp = &msb[0]; |
1753 | for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) { | |
1754 | /* | |
1755 | * Apply the delta at index n. If it fails, break | |
1756 | * from the loop so we'll fall into the undo loop | |
1757 | * below. | |
1758 | */ | |
8d280b98 DC |
1759 | switch (msbp->msb_field) { |
1760 | #ifdef HAVE_PERCPU_SB | |
1761 | case XFS_SBS_ICOUNT: | |
1762 | case XFS_SBS_IFREE: | |
1763 | case XFS_SBS_FDBLOCKS: | |
1764 | if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) { | |
3685c2a1 | 1765 | spin_unlock(&mp->m_sb_lock); |
20b64285 | 1766 | status = xfs_icsb_modify_counters(mp, |
8d280b98 DC |
1767 | msbp->msb_field, |
1768 | msbp->msb_delta, rsvd); | |
3685c2a1 | 1769 | spin_lock(&mp->m_sb_lock); |
8d280b98 DC |
1770 | break; |
1771 | } | |
1772 | /* FALLTHROUGH */ | |
1773 | #endif | |
1774 | default: | |
1775 | status = xfs_mod_incore_sb_unlocked(mp, | |
1776 | msbp->msb_field, | |
1777 | msbp->msb_delta, rsvd); | |
1778 | break; | |
1779 | } | |
1780 | ||
1da177e4 LT |
1781 | if (status != 0) { |
1782 | break; | |
1783 | } | |
1784 | } | |
1785 | ||
1786 | /* | |
1787 | * If we didn't complete the loop above, then back out | |
1788 | * any changes made to the superblock. If you add code | |
1789 | * between the loop above and here, make sure that you | |
1790 | * preserve the value of status. Loop back until | |
1791 | * we step below the beginning of the array. Make sure | |
1792 | * we don't touch anything back there. | |
1793 | */ | |
1794 | if (status != 0) { | |
1795 | msbp--; | |
1796 | while (msbp >= msb) { | |
8d280b98 DC |
1797 | switch (msbp->msb_field) { |
1798 | #ifdef HAVE_PERCPU_SB | |
1799 | case XFS_SBS_ICOUNT: | |
1800 | case XFS_SBS_IFREE: | |
1801 | case XFS_SBS_FDBLOCKS: | |
1802 | if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) { | |
3685c2a1 | 1803 | spin_unlock(&mp->m_sb_lock); |
20b64285 | 1804 | status = xfs_icsb_modify_counters(mp, |
8d280b98 DC |
1805 | msbp->msb_field, |
1806 | -(msbp->msb_delta), | |
1807 | rsvd); | |
3685c2a1 | 1808 | spin_lock(&mp->m_sb_lock); |
8d280b98 DC |
1809 | break; |
1810 | } | |
1811 | /* FALLTHROUGH */ | |
1812 | #endif | |
1813 | default: | |
1814 | status = xfs_mod_incore_sb_unlocked(mp, | |
1815 | msbp->msb_field, | |
1816 | -(msbp->msb_delta), | |
1817 | rsvd); | |
1818 | break; | |
1819 | } | |
1da177e4 LT |
1820 | ASSERT(status == 0); |
1821 | msbp--; | |
1822 | } | |
1823 | } | |
3685c2a1 | 1824 | spin_unlock(&mp->m_sb_lock); |
014c2544 | 1825 | return status; |
1da177e4 LT |
1826 | } |
1827 | ||
1828 | /* | |
1829 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
1830 | * The buffer is returned locked and read in from disk. | |
1831 | * The buffer should be released with a call to xfs_brelse(). | |
1832 | * | |
1833 | * If the flags parameter is BUF_TRYLOCK, then we'll only return | |
1834 | * the superblock buffer if it can be locked without sleeping. | |
1835 | * If it can't then we'll return NULL. | |
1836 | */ | |
1837 | xfs_buf_t * | |
1838 | xfs_getsb( | |
1839 | xfs_mount_t *mp, | |
1840 | int flags) | |
1841 | { | |
1842 | xfs_buf_t *bp; | |
1843 | ||
1844 | ASSERT(mp->m_sb_bp != NULL); | |
1845 | bp = mp->m_sb_bp; | |
1846 | if (flags & XFS_BUF_TRYLOCK) { | |
1847 | if (!XFS_BUF_CPSEMA(bp)) { | |
1848 | return NULL; | |
1849 | } | |
1850 | } else { | |
1851 | XFS_BUF_PSEMA(bp, PRIBIO); | |
1852 | } | |
1853 | XFS_BUF_HOLD(bp); | |
1854 | ASSERT(XFS_BUF_ISDONE(bp)); | |
014c2544 | 1855 | return bp; |
1da177e4 LT |
1856 | } |
1857 | ||
1858 | /* | |
1859 | * Used to free the superblock along various error paths. | |
1860 | */ | |
1861 | void | |
1862 | xfs_freesb( | |
1863 | xfs_mount_t *mp) | |
1864 | { | |
1865 | xfs_buf_t *bp; | |
1866 | ||
1867 | /* | |
1868 | * Use xfs_getsb() so that the buffer will be locked | |
1869 | * when we call xfs_buf_relse(). | |
1870 | */ | |
1871 | bp = xfs_getsb(mp, 0); | |
1872 | XFS_BUF_UNMANAGE(bp); | |
1873 | xfs_buf_relse(bp); | |
1874 | mp->m_sb_bp = NULL; | |
1875 | } | |
1876 | ||
1877 | /* | |
1878 | * See if the UUID is unique among mounted XFS filesystems. | |
1879 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. | |
1880 | */ | |
1881 | STATIC int | |
1882 | xfs_uuid_mount( | |
1883 | xfs_mount_t *mp) | |
1884 | { | |
1885 | if (uuid_is_nil(&mp->m_sb.sb_uuid)) { | |
1886 | cmn_err(CE_WARN, | |
1887 | "XFS: Filesystem %s has nil UUID - can't mount", | |
1888 | mp->m_fsname); | |
1889 | return -1; | |
1890 | } | |
1891 | if (!uuid_table_insert(&mp->m_sb.sb_uuid)) { | |
1892 | cmn_err(CE_WARN, | |
1893 | "XFS: Filesystem %s has duplicate UUID - can't mount", | |
1894 | mp->m_fsname); | |
1895 | return -1; | |
1896 | } | |
1897 | return 0; | |
1898 | } | |
1899 | ||
1da177e4 LT |
1900 | /* |
1901 | * Used to log changes to the superblock unit and width fields which could | |
e6957ea4 ES |
1902 | * be altered by the mount options, as well as any potential sb_features2 |
1903 | * fixup. Only the first superblock is updated. | |
1da177e4 | 1904 | */ |
e5720eec | 1905 | STATIC int |
ee1c0908 | 1906 | xfs_mount_log_sb( |
1da177e4 LT |
1907 | xfs_mount_t *mp, |
1908 | __int64_t fields) | |
1909 | { | |
1910 | xfs_trans_t *tp; | |
e5720eec | 1911 | int error; |
1da177e4 | 1912 | |
ee1c0908 | 1913 | ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID | |
4b166de0 DC |
1914 | XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 | |
1915 | XFS_SB_VERSIONNUM)); | |
1da177e4 LT |
1916 | |
1917 | tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT); | |
e5720eec DC |
1918 | error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0, |
1919 | XFS_DEFAULT_LOG_COUNT); | |
1920 | if (error) { | |
1da177e4 | 1921 | xfs_trans_cancel(tp, 0); |
e5720eec | 1922 | return error; |
1da177e4 LT |
1923 | } |
1924 | xfs_mod_sb(tp, fields); | |
e5720eec DC |
1925 | error = xfs_trans_commit(tp, 0); |
1926 | return error; | |
1da177e4 | 1927 | } |
8d280b98 DC |
1928 | |
1929 | ||
1930 | #ifdef HAVE_PERCPU_SB | |
1931 | /* | |
1932 | * Per-cpu incore superblock counters | |
1933 | * | |
1934 | * Simple concept, difficult implementation | |
1935 | * | |
1936 | * Basically, replace the incore superblock counters with a distributed per cpu | |
1937 | * counter for contended fields (e.g. free block count). | |
1938 | * | |
1939 | * Difficulties arise in that the incore sb is used for ENOSPC checking, and | |
1940 | * hence needs to be accurately read when we are running low on space. Hence | |
1941 | * there is a method to enable and disable the per-cpu counters based on how | |
1942 | * much "stuff" is available in them. | |
1943 | * | |
1944 | * Basically, a counter is enabled if there is enough free resource to justify | |
1945 | * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local | |
1946 | * ENOSPC), then we disable the counters to synchronise all callers and | |
1947 | * re-distribute the available resources. | |
1948 | * | |
1949 | * If, once we redistributed the available resources, we still get a failure, | |
1950 | * we disable the per-cpu counter and go through the slow path. | |
1951 | * | |
1952 | * The slow path is the current xfs_mod_incore_sb() function. This means that | |
1953 | * when we disable a per-cpu counter, we need to drain it's resources back to | |
1954 | * the global superblock. We do this after disabling the counter to prevent | |
1955 | * more threads from queueing up on the counter. | |
1956 | * | |
1957 | * Essentially, this means that we still need a lock in the fast path to enable | |
1958 | * synchronisation between the global counters and the per-cpu counters. This | |
1959 | * is not a problem because the lock will be local to a CPU almost all the time | |
1960 | * and have little contention except when we get to ENOSPC conditions. | |
1961 | * | |
1962 | * Basically, this lock becomes a barrier that enables us to lock out the fast | |
1963 | * path while we do things like enabling and disabling counters and | |
1964 | * synchronising the counters. | |
1965 | * | |
1966 | * Locking rules: | |
1967 | * | |
3685c2a1 | 1968 | * 1. m_sb_lock before picking up per-cpu locks |
8d280b98 | 1969 | * 2. per-cpu locks always picked up via for_each_online_cpu() order |
3685c2a1 | 1970 | * 3. accurate counter sync requires m_sb_lock + per cpu locks |
8d280b98 | 1971 | * 4. modifying per-cpu counters requires holding per-cpu lock |
3685c2a1 ES |
1972 | * 5. modifying global counters requires holding m_sb_lock |
1973 | * 6. enabling or disabling a counter requires holding the m_sb_lock | |
8d280b98 DC |
1974 | * and _none_ of the per-cpu locks. |
1975 | * | |
1976 | * Disabled counters are only ever re-enabled by a balance operation | |
1977 | * that results in more free resources per CPU than a given threshold. | |
1978 | * To ensure counters don't remain disabled, they are rebalanced when | |
1979 | * the global resource goes above a higher threshold (i.e. some hysteresis | |
1980 | * is present to prevent thrashing). | |
e8234a68 DC |
1981 | */ |
1982 | ||
5a67e4c5 | 1983 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
1984 | /* |
1985 | * hot-plug CPU notifier support. | |
8d280b98 | 1986 | * |
5a67e4c5 CS |
1987 | * We need a notifier per filesystem as we need to be able to identify |
1988 | * the filesystem to balance the counters out. This is achieved by | |
1989 | * having a notifier block embedded in the xfs_mount_t and doing pointer | |
1990 | * magic to get the mount pointer from the notifier block address. | |
8d280b98 | 1991 | */ |
e8234a68 DC |
1992 | STATIC int |
1993 | xfs_icsb_cpu_notify( | |
1994 | struct notifier_block *nfb, | |
1995 | unsigned long action, | |
1996 | void *hcpu) | |
1997 | { | |
1998 | xfs_icsb_cnts_t *cntp; | |
1999 | xfs_mount_t *mp; | |
e8234a68 DC |
2000 | |
2001 | mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier); | |
2002 | cntp = (xfs_icsb_cnts_t *) | |
2003 | per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu); | |
2004 | switch (action) { | |
2005 | case CPU_UP_PREPARE: | |
8bb78442 | 2006 | case CPU_UP_PREPARE_FROZEN: |
e8234a68 DC |
2007 | /* Easy Case - initialize the area and locks, and |
2008 | * then rebalance when online does everything else for us. */ | |
01e1b69c | 2009 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 DC |
2010 | break; |
2011 | case CPU_ONLINE: | |
8bb78442 | 2012 | case CPU_ONLINE_FROZEN: |
03135cf7 | 2013 | xfs_icsb_lock(mp); |
45af6c6d CH |
2014 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
2015 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
2016 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
03135cf7 | 2017 | xfs_icsb_unlock(mp); |
e8234a68 DC |
2018 | break; |
2019 | case CPU_DEAD: | |
8bb78442 | 2020 | case CPU_DEAD_FROZEN: |
e8234a68 DC |
2021 | /* Disable all the counters, then fold the dead cpu's |
2022 | * count into the total on the global superblock and | |
2023 | * re-enable the counters. */ | |
03135cf7 | 2024 | xfs_icsb_lock(mp); |
3685c2a1 | 2025 | spin_lock(&mp->m_sb_lock); |
e8234a68 DC |
2026 | xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT); |
2027 | xfs_icsb_disable_counter(mp, XFS_SBS_IFREE); | |
2028 | xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS); | |
2029 | ||
2030 | mp->m_sb.sb_icount += cntp->icsb_icount; | |
2031 | mp->m_sb.sb_ifree += cntp->icsb_ifree; | |
2032 | mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks; | |
2033 | ||
01e1b69c | 2034 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 | 2035 | |
45af6c6d CH |
2036 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0); |
2037 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0); | |
2038 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0); | |
3685c2a1 | 2039 | spin_unlock(&mp->m_sb_lock); |
03135cf7 | 2040 | xfs_icsb_unlock(mp); |
e8234a68 DC |
2041 | break; |
2042 | } | |
2043 | ||
2044 | return NOTIFY_OK; | |
2045 | } | |
5a67e4c5 | 2046 | #endif /* CONFIG_HOTPLUG_CPU */ |
e8234a68 | 2047 | |
8d280b98 DC |
2048 | int |
2049 | xfs_icsb_init_counters( | |
2050 | xfs_mount_t *mp) | |
2051 | { | |
2052 | xfs_icsb_cnts_t *cntp; | |
2053 | int i; | |
2054 | ||
2055 | mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t); | |
2056 | if (mp->m_sb_cnts == NULL) | |
2057 | return -ENOMEM; | |
2058 | ||
5a67e4c5 | 2059 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
2060 | mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify; |
2061 | mp->m_icsb_notifier.priority = 0; | |
5a67e4c5 CS |
2062 | register_hotcpu_notifier(&mp->m_icsb_notifier); |
2063 | #endif /* CONFIG_HOTPLUG_CPU */ | |
e8234a68 | 2064 | |
8d280b98 DC |
2065 | for_each_online_cpu(i) { |
2066 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2067 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
8d280b98 | 2068 | } |
20b64285 DC |
2069 | |
2070 | mutex_init(&mp->m_icsb_mutex); | |
2071 | ||
8d280b98 DC |
2072 | /* |
2073 | * start with all counters disabled so that the | |
2074 | * initial balance kicks us off correctly | |
2075 | */ | |
2076 | mp->m_icsb_counters = -1; | |
2077 | return 0; | |
2078 | } | |
2079 | ||
5478eead LM |
2080 | void |
2081 | xfs_icsb_reinit_counters( | |
2082 | xfs_mount_t *mp) | |
2083 | { | |
2084 | xfs_icsb_lock(mp); | |
2085 | /* | |
2086 | * start with all counters disabled so that the | |
2087 | * initial balance kicks us off correctly | |
2088 | */ | |
2089 | mp->m_icsb_counters = -1; | |
45af6c6d CH |
2090 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
2091 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
2092 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
5478eead LM |
2093 | xfs_icsb_unlock(mp); |
2094 | } | |
2095 | ||
8d280b98 DC |
2096 | STATIC void |
2097 | xfs_icsb_destroy_counters( | |
2098 | xfs_mount_t *mp) | |
2099 | { | |
e8234a68 | 2100 | if (mp->m_sb_cnts) { |
5a67e4c5 | 2101 | unregister_hotcpu_notifier(&mp->m_icsb_notifier); |
8d280b98 | 2102 | free_percpu(mp->m_sb_cnts); |
e8234a68 | 2103 | } |
03135cf7 | 2104 | mutex_destroy(&mp->m_icsb_mutex); |
8d280b98 DC |
2105 | } |
2106 | ||
7989cb8e | 2107 | STATIC_INLINE void |
01e1b69c DC |
2108 | xfs_icsb_lock_cntr( |
2109 | xfs_icsb_cnts_t *icsbp) | |
2110 | { | |
2111 | while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) { | |
2112 | ndelay(1000); | |
2113 | } | |
2114 | } | |
2115 | ||
7989cb8e | 2116 | STATIC_INLINE void |
01e1b69c DC |
2117 | xfs_icsb_unlock_cntr( |
2118 | xfs_icsb_cnts_t *icsbp) | |
2119 | { | |
2120 | clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags); | |
2121 | } | |
2122 | ||
8d280b98 | 2123 | |
7989cb8e | 2124 | STATIC_INLINE void |
8d280b98 DC |
2125 | xfs_icsb_lock_all_counters( |
2126 | xfs_mount_t *mp) | |
2127 | { | |
2128 | xfs_icsb_cnts_t *cntp; | |
2129 | int i; | |
2130 | ||
2131 | for_each_online_cpu(i) { | |
2132 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2133 | xfs_icsb_lock_cntr(cntp); |
8d280b98 DC |
2134 | } |
2135 | } | |
2136 | ||
7989cb8e | 2137 | STATIC_INLINE void |
8d280b98 DC |
2138 | xfs_icsb_unlock_all_counters( |
2139 | xfs_mount_t *mp) | |
2140 | { | |
2141 | xfs_icsb_cnts_t *cntp; | |
2142 | int i; | |
2143 | ||
2144 | for_each_online_cpu(i) { | |
2145 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2146 | xfs_icsb_unlock_cntr(cntp); |
8d280b98 DC |
2147 | } |
2148 | } | |
2149 | ||
2150 | STATIC void | |
2151 | xfs_icsb_count( | |
2152 | xfs_mount_t *mp, | |
2153 | xfs_icsb_cnts_t *cnt, | |
2154 | int flags) | |
2155 | { | |
2156 | xfs_icsb_cnts_t *cntp; | |
2157 | int i; | |
2158 | ||
2159 | memset(cnt, 0, sizeof(xfs_icsb_cnts_t)); | |
2160 | ||
2161 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
2162 | xfs_icsb_lock_all_counters(mp); | |
2163 | ||
2164 | for_each_online_cpu(i) { | |
2165 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
2166 | cnt->icsb_icount += cntp->icsb_icount; | |
2167 | cnt->icsb_ifree += cntp->icsb_ifree; | |
2168 | cnt->icsb_fdblocks += cntp->icsb_fdblocks; | |
2169 | } | |
2170 | ||
2171 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
2172 | xfs_icsb_unlock_all_counters(mp); | |
2173 | } | |
2174 | ||
2175 | STATIC int | |
2176 | xfs_icsb_counter_disabled( | |
2177 | xfs_mount_t *mp, | |
2178 | xfs_sb_field_t field) | |
2179 | { | |
2180 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2181 | return test_bit(field, &mp->m_icsb_counters); | |
2182 | } | |
2183 | ||
36fbe6e6 | 2184 | STATIC void |
8d280b98 DC |
2185 | xfs_icsb_disable_counter( |
2186 | xfs_mount_t *mp, | |
2187 | xfs_sb_field_t field) | |
2188 | { | |
2189 | xfs_icsb_cnts_t cnt; | |
2190 | ||
2191 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2192 | ||
20b64285 DC |
2193 | /* |
2194 | * If we are already disabled, then there is nothing to do | |
2195 | * here. We check before locking all the counters to avoid | |
2196 | * the expensive lock operation when being called in the | |
2197 | * slow path and the counter is already disabled. This is | |
2198 | * safe because the only time we set or clear this state is under | |
2199 | * the m_icsb_mutex. | |
2200 | */ | |
2201 | if (xfs_icsb_counter_disabled(mp, field)) | |
36fbe6e6 | 2202 | return; |
20b64285 | 2203 | |
8d280b98 DC |
2204 | xfs_icsb_lock_all_counters(mp); |
2205 | if (!test_and_set_bit(field, &mp->m_icsb_counters)) { | |
2206 | /* drain back to superblock */ | |
2207 | ||
ce46193b | 2208 | xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT); |
8d280b98 DC |
2209 | switch(field) { |
2210 | case XFS_SBS_ICOUNT: | |
2211 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
2212 | break; | |
2213 | case XFS_SBS_IFREE: | |
2214 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
2215 | break; | |
2216 | case XFS_SBS_FDBLOCKS: | |
2217 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
2218 | break; | |
2219 | default: | |
2220 | BUG(); | |
2221 | } | |
2222 | } | |
2223 | ||
2224 | xfs_icsb_unlock_all_counters(mp); | |
8d280b98 DC |
2225 | } |
2226 | ||
2227 | STATIC void | |
2228 | xfs_icsb_enable_counter( | |
2229 | xfs_mount_t *mp, | |
2230 | xfs_sb_field_t field, | |
2231 | uint64_t count, | |
2232 | uint64_t resid) | |
2233 | { | |
2234 | xfs_icsb_cnts_t *cntp; | |
2235 | int i; | |
2236 | ||
2237 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2238 | ||
2239 | xfs_icsb_lock_all_counters(mp); | |
2240 | for_each_online_cpu(i) { | |
2241 | cntp = per_cpu_ptr(mp->m_sb_cnts, i); | |
2242 | switch (field) { | |
2243 | case XFS_SBS_ICOUNT: | |
2244 | cntp->icsb_icount = count + resid; | |
2245 | break; | |
2246 | case XFS_SBS_IFREE: | |
2247 | cntp->icsb_ifree = count + resid; | |
2248 | break; | |
2249 | case XFS_SBS_FDBLOCKS: | |
2250 | cntp->icsb_fdblocks = count + resid; | |
2251 | break; | |
2252 | default: | |
2253 | BUG(); | |
2254 | break; | |
2255 | } | |
2256 | resid = 0; | |
2257 | } | |
2258 | clear_bit(field, &mp->m_icsb_counters); | |
2259 | xfs_icsb_unlock_all_counters(mp); | |
2260 | } | |
2261 | ||
dbcabad1 | 2262 | void |
d4d90b57 | 2263 | xfs_icsb_sync_counters_locked( |
8d280b98 DC |
2264 | xfs_mount_t *mp, |
2265 | int flags) | |
2266 | { | |
2267 | xfs_icsb_cnts_t cnt; | |
8d280b98 | 2268 | |
8d280b98 DC |
2269 | xfs_icsb_count(mp, &cnt, flags); |
2270 | ||
8d280b98 DC |
2271 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT)) |
2272 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
2273 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE)) | |
2274 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
2275 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS)) | |
2276 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
8d280b98 DC |
2277 | } |
2278 | ||
2279 | /* | |
2280 | * Accurate update of per-cpu counters to incore superblock | |
2281 | */ | |
d4d90b57 | 2282 | void |
8d280b98 | 2283 | xfs_icsb_sync_counters( |
d4d90b57 CH |
2284 | xfs_mount_t *mp, |
2285 | int flags) | |
8d280b98 | 2286 | { |
d4d90b57 CH |
2287 | spin_lock(&mp->m_sb_lock); |
2288 | xfs_icsb_sync_counters_locked(mp, flags); | |
2289 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
2290 | } |
2291 | ||
2292 | /* | |
2293 | * Balance and enable/disable counters as necessary. | |
2294 | * | |
20b64285 DC |
2295 | * Thresholds for re-enabling counters are somewhat magic. inode counts are |
2296 | * chosen to be the same number as single on disk allocation chunk per CPU, and | |
2297 | * free blocks is something far enough zero that we aren't going thrash when we | |
2298 | * get near ENOSPC. We also need to supply a minimum we require per cpu to | |
2299 | * prevent looping endlessly when xfs_alloc_space asks for more than will | |
2300 | * be distributed to a single CPU but each CPU has enough blocks to be | |
2301 | * reenabled. | |
2302 | * | |
2303 | * Note that we can be called when counters are already disabled. | |
2304 | * xfs_icsb_disable_counter() optimises the counter locking in this case to | |
2305 | * prevent locking every per-cpu counter needlessly. | |
8d280b98 | 2306 | */ |
20b64285 DC |
2307 | |
2308 | #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64 | |
4be536de | 2309 | #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \ |
20b64285 | 2310 | (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp)) |
8d280b98 | 2311 | STATIC void |
45af6c6d | 2312 | xfs_icsb_balance_counter_locked( |
8d280b98 DC |
2313 | xfs_mount_t *mp, |
2314 | xfs_sb_field_t field, | |
20b64285 | 2315 | int min_per_cpu) |
8d280b98 | 2316 | { |
6fdf8ccc | 2317 | uint64_t count, resid; |
8d280b98 | 2318 | int weight = num_online_cpus(); |
20b64285 | 2319 | uint64_t min = (uint64_t)min_per_cpu; |
8d280b98 | 2320 | |
8d280b98 DC |
2321 | /* disable counter and sync counter */ |
2322 | xfs_icsb_disable_counter(mp, field); | |
2323 | ||
2324 | /* update counters - first CPU gets residual*/ | |
2325 | switch (field) { | |
2326 | case XFS_SBS_ICOUNT: | |
2327 | count = mp->m_sb.sb_icount; | |
2328 | resid = do_div(count, weight); | |
20b64285 | 2329 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 2330 | return; |
8d280b98 DC |
2331 | break; |
2332 | case XFS_SBS_IFREE: | |
2333 | count = mp->m_sb.sb_ifree; | |
2334 | resid = do_div(count, weight); | |
20b64285 | 2335 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 2336 | return; |
8d280b98 DC |
2337 | break; |
2338 | case XFS_SBS_FDBLOCKS: | |
2339 | count = mp->m_sb.sb_fdblocks; | |
2340 | resid = do_div(count, weight); | |
20b64285 | 2341 | if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp))) |
45af6c6d | 2342 | return; |
8d280b98 DC |
2343 | break; |
2344 | default: | |
2345 | BUG(); | |
6fdf8ccc | 2346 | count = resid = 0; /* quiet, gcc */ |
8d280b98 DC |
2347 | break; |
2348 | } | |
2349 | ||
2350 | xfs_icsb_enable_counter(mp, field, count, resid); | |
45af6c6d CH |
2351 | } |
2352 | ||
2353 | STATIC void | |
2354 | xfs_icsb_balance_counter( | |
2355 | xfs_mount_t *mp, | |
2356 | xfs_sb_field_t fields, | |
2357 | int min_per_cpu) | |
2358 | { | |
2359 | spin_lock(&mp->m_sb_lock); | |
2360 | xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu); | |
2361 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
2362 | } |
2363 | ||
a8272ce0 | 2364 | STATIC int |
20b64285 | 2365 | xfs_icsb_modify_counters( |
8d280b98 DC |
2366 | xfs_mount_t *mp, |
2367 | xfs_sb_field_t field, | |
20f4ebf2 | 2368 | int64_t delta, |
20b64285 | 2369 | int rsvd) |
8d280b98 DC |
2370 | { |
2371 | xfs_icsb_cnts_t *icsbp; | |
2372 | long long lcounter; /* long counter for 64 bit fields */ | |
007c61c6 | 2373 | int cpu, ret = 0; |
8d280b98 | 2374 | |
20b64285 | 2375 | might_sleep(); |
8d280b98 DC |
2376 | again: |
2377 | cpu = get_cpu(); | |
20b64285 DC |
2378 | icsbp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, cpu); |
2379 | ||
2380 | /* | |
2381 | * if the counter is disabled, go to slow path | |
2382 | */ | |
8d280b98 DC |
2383 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) |
2384 | goto slow_path; | |
20b64285 DC |
2385 | xfs_icsb_lock_cntr(icsbp); |
2386 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) { | |
2387 | xfs_icsb_unlock_cntr(icsbp); | |
2388 | goto slow_path; | |
2389 | } | |
8d280b98 DC |
2390 | |
2391 | switch (field) { | |
2392 | case XFS_SBS_ICOUNT: | |
2393 | lcounter = icsbp->icsb_icount; | |
2394 | lcounter += delta; | |
2395 | if (unlikely(lcounter < 0)) | |
20b64285 | 2396 | goto balance_counter; |
8d280b98 DC |
2397 | icsbp->icsb_icount = lcounter; |
2398 | break; | |
2399 | ||
2400 | case XFS_SBS_IFREE: | |
2401 | lcounter = icsbp->icsb_ifree; | |
2402 | lcounter += delta; | |
2403 | if (unlikely(lcounter < 0)) | |
20b64285 | 2404 | goto balance_counter; |
8d280b98 DC |
2405 | icsbp->icsb_ifree = lcounter; |
2406 | break; | |
2407 | ||
2408 | case XFS_SBS_FDBLOCKS: | |
2409 | BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0); | |
2410 | ||
4be536de | 2411 | lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
2412 | lcounter += delta; |
2413 | if (unlikely(lcounter < 0)) | |
20b64285 | 2414 | goto balance_counter; |
4be536de | 2415 | icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
2416 | break; |
2417 | default: | |
2418 | BUG(); | |
2419 | break; | |
2420 | } | |
01e1b69c | 2421 | xfs_icsb_unlock_cntr(icsbp); |
8d280b98 | 2422 | put_cpu(); |
8d280b98 DC |
2423 | return 0; |
2424 | ||
8d280b98 | 2425 | slow_path: |
8d280b98 DC |
2426 | put_cpu(); |
2427 | ||
20b64285 DC |
2428 | /* |
2429 | * serialise with a mutex so we don't burn lots of cpu on | |
2430 | * the superblock lock. We still need to hold the superblock | |
2431 | * lock, however, when we modify the global structures. | |
2432 | */ | |
03135cf7 | 2433 | xfs_icsb_lock(mp); |
20b64285 DC |
2434 | |
2435 | /* | |
2436 | * Now running atomically. | |
2437 | * | |
2438 | * If the counter is enabled, someone has beaten us to rebalancing. | |
2439 | * Drop the lock and try again in the fast path.... | |
2440 | */ | |
2441 | if (!(xfs_icsb_counter_disabled(mp, field))) { | |
03135cf7 | 2442 | xfs_icsb_unlock(mp); |
8d280b98 | 2443 | goto again; |
8d280b98 DC |
2444 | } |
2445 | ||
20b64285 DC |
2446 | /* |
2447 | * The counter is currently disabled. Because we are | |
2448 | * running atomically here, we know a rebalance cannot | |
2449 | * be in progress. Hence we can go straight to operating | |
2450 | * on the global superblock. We do not call xfs_mod_incore_sb() | |
3685c2a1 | 2451 | * here even though we need to get the m_sb_lock. Doing so |
20b64285 | 2452 | * will cause us to re-enter this function and deadlock. |
3685c2a1 | 2453 | * Hence we get the m_sb_lock ourselves and then call |
20b64285 DC |
2454 | * xfs_mod_incore_sb_unlocked() as the unlocked path operates |
2455 | * directly on the global counters. | |
2456 | */ | |
3685c2a1 | 2457 | spin_lock(&mp->m_sb_lock); |
8d280b98 | 2458 | ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); |
3685c2a1 | 2459 | spin_unlock(&mp->m_sb_lock); |
8d280b98 | 2460 | |
20b64285 DC |
2461 | /* |
2462 | * Now that we've modified the global superblock, we | |
2463 | * may be able to re-enable the distributed counters | |
2464 | * (e.g. lots of space just got freed). After that | |
2465 | * we are done. | |
2466 | */ | |
2467 | if (ret != ENOSPC) | |
45af6c6d | 2468 | xfs_icsb_balance_counter(mp, field, 0); |
03135cf7 | 2469 | xfs_icsb_unlock(mp); |
8d280b98 | 2470 | return ret; |
8d280b98 | 2471 | |
20b64285 DC |
2472 | balance_counter: |
2473 | xfs_icsb_unlock_cntr(icsbp); | |
2474 | put_cpu(); | |
8d280b98 | 2475 | |
20b64285 DC |
2476 | /* |
2477 | * We may have multiple threads here if multiple per-cpu | |
2478 | * counters run dry at the same time. This will mean we can | |
2479 | * do more balances than strictly necessary but it is not | |
2480 | * the common slowpath case. | |
2481 | */ | |
03135cf7 | 2482 | xfs_icsb_lock(mp); |
20b64285 DC |
2483 | |
2484 | /* | |
2485 | * running atomically. | |
2486 | * | |
2487 | * This will leave the counter in the correct state for future | |
2488 | * accesses. After the rebalance, we simply try again and our retry | |
2489 | * will either succeed through the fast path or slow path without | |
2490 | * another balance operation being required. | |
2491 | */ | |
45af6c6d | 2492 | xfs_icsb_balance_counter(mp, field, delta); |
03135cf7 | 2493 | xfs_icsb_unlock(mp); |
20b64285 | 2494 | goto again; |
8d280b98 | 2495 | } |
20b64285 | 2496 | |
8d280b98 | 2497 | #endif |