]>
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" | |
27 | #include "xfs_dir.h" | |
28 | #include "xfs_dir2.h" | |
29 | #include "xfs_dmapi.h" | |
30 | #include "xfs_mount.h" | |
1da177e4 | 31 | #include "xfs_bmap_btree.h" |
a844f451 | 32 | #include "xfs_alloc_btree.h" |
1da177e4 | 33 | #include "xfs_ialloc_btree.h" |
1da177e4 LT |
34 | #include "xfs_dir_sf.h" |
35 | #include "xfs_dir2_sf.h" | |
a844f451 | 36 | #include "xfs_attr_sf.h" |
1da177e4 LT |
37 | #include "xfs_dinode.h" |
38 | #include "xfs_inode.h" | |
a844f451 NS |
39 | #include "xfs_btree.h" |
40 | #include "xfs_ialloc.h" | |
1da177e4 LT |
41 | #include "xfs_alloc.h" |
42 | #include "xfs_rtalloc.h" | |
43 | #include "xfs_bmap.h" | |
44 | #include "xfs_error.h" | |
1da177e4 LT |
45 | #include "xfs_rw.h" |
46 | #include "xfs_quota.h" | |
47 | #include "xfs_fsops.h" | |
48 | ||
49 | STATIC void xfs_mount_log_sbunit(xfs_mount_t *, __int64_t); | |
50 | STATIC int xfs_uuid_mount(xfs_mount_t *); | |
51 | STATIC void xfs_uuid_unmount(xfs_mount_t *mp); | |
ba0f32d4 | 52 | STATIC void xfs_unmountfs_wait(xfs_mount_t *); |
1da177e4 | 53 | |
8d280b98 DC |
54 | |
55 | #ifdef HAVE_PERCPU_SB | |
56 | STATIC void xfs_icsb_destroy_counters(xfs_mount_t *); | |
57 | STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t, int); | |
58 | STATIC void xfs_icsb_sync_counters(xfs_mount_t *); | |
59 | STATIC int xfs_icsb_modify_counters(xfs_mount_t *, xfs_sb_field_t, | |
60 | int, int); | |
61 | STATIC int xfs_icsb_modify_counters_locked(xfs_mount_t *, xfs_sb_field_t, | |
62 | int, int); | |
e8234a68 | 63 | STATIC int xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t); |
8d280b98 DC |
64 | |
65 | #else | |
66 | ||
67 | #define xfs_icsb_destroy_counters(mp) do { } while (0) | |
68 | #define xfs_icsb_balance_counter(mp, a, b) do { } while (0) | |
69 | #define xfs_icsb_sync_counters(mp) do { } while (0) | |
70 | #define xfs_icsb_modify_counters(mp, a, b, c) do { } while (0) | |
71 | #define xfs_icsb_modify_counters_locked(mp, a, b, c) do { } while (0) | |
72 | ||
73 | #endif | |
74 | ||
1df84c93 | 75 | static const struct { |
8d280b98 DC |
76 | short offset; |
77 | short type; /* 0 = integer | |
78 | * 1 = binary / string (no translation) | |
79 | */ | |
1da177e4 LT |
80 | } xfs_sb_info[] = { |
81 | { offsetof(xfs_sb_t, sb_magicnum), 0 }, | |
82 | { offsetof(xfs_sb_t, sb_blocksize), 0 }, | |
83 | { offsetof(xfs_sb_t, sb_dblocks), 0 }, | |
84 | { offsetof(xfs_sb_t, sb_rblocks), 0 }, | |
85 | { offsetof(xfs_sb_t, sb_rextents), 0 }, | |
86 | { offsetof(xfs_sb_t, sb_uuid), 1 }, | |
87 | { offsetof(xfs_sb_t, sb_logstart), 0 }, | |
88 | { offsetof(xfs_sb_t, sb_rootino), 0 }, | |
89 | { offsetof(xfs_sb_t, sb_rbmino), 0 }, | |
90 | { offsetof(xfs_sb_t, sb_rsumino), 0 }, | |
91 | { offsetof(xfs_sb_t, sb_rextsize), 0 }, | |
92 | { offsetof(xfs_sb_t, sb_agblocks), 0 }, | |
93 | { offsetof(xfs_sb_t, sb_agcount), 0 }, | |
94 | { offsetof(xfs_sb_t, sb_rbmblocks), 0 }, | |
95 | { offsetof(xfs_sb_t, sb_logblocks), 0 }, | |
96 | { offsetof(xfs_sb_t, sb_versionnum), 0 }, | |
97 | { offsetof(xfs_sb_t, sb_sectsize), 0 }, | |
98 | { offsetof(xfs_sb_t, sb_inodesize), 0 }, | |
99 | { offsetof(xfs_sb_t, sb_inopblock), 0 }, | |
100 | { offsetof(xfs_sb_t, sb_fname[0]), 1 }, | |
101 | { offsetof(xfs_sb_t, sb_blocklog), 0 }, | |
102 | { offsetof(xfs_sb_t, sb_sectlog), 0 }, | |
103 | { offsetof(xfs_sb_t, sb_inodelog), 0 }, | |
104 | { offsetof(xfs_sb_t, sb_inopblog), 0 }, | |
105 | { offsetof(xfs_sb_t, sb_agblklog), 0 }, | |
106 | { offsetof(xfs_sb_t, sb_rextslog), 0 }, | |
107 | { offsetof(xfs_sb_t, sb_inprogress), 0 }, | |
108 | { offsetof(xfs_sb_t, sb_imax_pct), 0 }, | |
109 | { offsetof(xfs_sb_t, sb_icount), 0 }, | |
110 | { offsetof(xfs_sb_t, sb_ifree), 0 }, | |
111 | { offsetof(xfs_sb_t, sb_fdblocks), 0 }, | |
112 | { offsetof(xfs_sb_t, sb_frextents), 0 }, | |
113 | { offsetof(xfs_sb_t, sb_uquotino), 0 }, | |
114 | { offsetof(xfs_sb_t, sb_gquotino), 0 }, | |
115 | { offsetof(xfs_sb_t, sb_qflags), 0 }, | |
116 | { offsetof(xfs_sb_t, sb_flags), 0 }, | |
117 | { offsetof(xfs_sb_t, sb_shared_vn), 0 }, | |
118 | { offsetof(xfs_sb_t, sb_inoalignmt), 0 }, | |
119 | { offsetof(xfs_sb_t, sb_unit), 0 }, | |
120 | { offsetof(xfs_sb_t, sb_width), 0 }, | |
121 | { offsetof(xfs_sb_t, sb_dirblklog), 0 }, | |
122 | { offsetof(xfs_sb_t, sb_logsectlog), 0 }, | |
123 | { offsetof(xfs_sb_t, sb_logsectsize),0 }, | |
124 | { offsetof(xfs_sb_t, sb_logsunit), 0 }, | |
125 | { offsetof(xfs_sb_t, sb_features2), 0 }, | |
126 | { sizeof(xfs_sb_t), 0 } | |
127 | }; | |
128 | ||
129 | /* | |
130 | * Return a pointer to an initialized xfs_mount structure. | |
131 | */ | |
132 | xfs_mount_t * | |
133 | xfs_mount_init(void) | |
134 | { | |
135 | xfs_mount_t *mp; | |
136 | ||
8d280b98 DC |
137 | mp = kmem_zalloc(sizeof(xfs_mount_t), KM_SLEEP); |
138 | ||
139 | if (xfs_icsb_init_counters(mp)) { | |
140 | mp->m_flags |= XFS_MOUNT_NO_PERCPU_SB; | |
141 | } | |
1da177e4 LT |
142 | |
143 | AIL_LOCKINIT(&mp->m_ail_lock, "xfs_ail"); | |
144 | spinlock_init(&mp->m_sb_lock, "xfs_sb"); | |
794ee1ba | 145 | mutex_init(&mp->m_ilock); |
1da177e4 LT |
146 | initnsema(&mp->m_growlock, 1, "xfs_grow"); |
147 | /* | |
148 | * Initialize the AIL. | |
149 | */ | |
150 | xfs_trans_ail_init(mp); | |
151 | ||
152 | atomic_set(&mp->m_active_trans, 0); | |
153 | ||
154 | return mp; | |
155 | } | |
156 | ||
157 | /* | |
158 | * Free up the resources associated with a mount structure. Assume that | |
159 | * the structure was initially zeroed, so we can tell which fields got | |
160 | * initialized. | |
161 | */ | |
162 | void | |
163 | xfs_mount_free( | |
8d280b98 DC |
164 | xfs_mount_t *mp, |
165 | int remove_bhv) | |
1da177e4 LT |
166 | { |
167 | if (mp->m_ihash) | |
168 | xfs_ihash_free(mp); | |
169 | if (mp->m_chash) | |
170 | xfs_chash_free(mp); | |
171 | ||
172 | if (mp->m_perag) { | |
173 | int agno; | |
174 | ||
175 | for (agno = 0; agno < mp->m_maxagi; agno++) | |
176 | if (mp->m_perag[agno].pagb_list) | |
177 | kmem_free(mp->m_perag[agno].pagb_list, | |
178 | sizeof(xfs_perag_busy_t) * | |
179 | XFS_PAGB_NUM_SLOTS); | |
180 | kmem_free(mp->m_perag, | |
181 | sizeof(xfs_perag_t) * mp->m_sb.sb_agcount); | |
182 | } | |
183 | ||
184 | AIL_LOCK_DESTROY(&mp->m_ail_lock); | |
185 | spinlock_destroy(&mp->m_sb_lock); | |
186 | mutex_destroy(&mp->m_ilock); | |
187 | freesema(&mp->m_growlock); | |
188 | if (mp->m_quotainfo) | |
189 | XFS_QM_DONE(mp); | |
190 | ||
191 | if (mp->m_fsname != NULL) | |
192 | kmem_free(mp->m_fsname, mp->m_fsname_len); | |
fc1f8c1c NS |
193 | if (mp->m_rtname != NULL) |
194 | kmem_free(mp->m_rtname, strlen(mp->m_rtname) + 1); | |
195 | if (mp->m_logname != NULL) | |
196 | kmem_free(mp->m_logname, strlen(mp->m_logname) + 1); | |
1da177e4 LT |
197 | |
198 | if (remove_bhv) { | |
b83bd138 | 199 | struct bhv_vfs *vfsp = XFS_MTOVFS(mp); |
1da177e4 LT |
200 | |
201 | bhv_remove_all_vfsops(vfsp, 0); | |
202 | VFS_REMOVEBHV(vfsp, &mp->m_bhv); | |
203 | } | |
204 | ||
8d280b98 | 205 | xfs_icsb_destroy_counters(mp); |
1da177e4 LT |
206 | kmem_free(mp, sizeof(xfs_mount_t)); |
207 | } | |
208 | ||
209 | ||
210 | /* | |
211 | * Check the validity of the SB found. | |
212 | */ | |
213 | STATIC int | |
214 | xfs_mount_validate_sb( | |
215 | xfs_mount_t *mp, | |
764d1f89 NS |
216 | xfs_sb_t *sbp, |
217 | int flags) | |
1da177e4 LT |
218 | { |
219 | /* | |
220 | * If the log device and data device have the | |
221 | * same device number, the log is internal. | |
222 | * Consequently, the sb_logstart should be non-zero. If | |
223 | * we have a zero sb_logstart in this case, we may be trying to mount | |
224 | * a volume filesystem in a non-volume manner. | |
225 | */ | |
226 | if (sbp->sb_magicnum != XFS_SB_MAGIC) { | |
764d1f89 | 227 | xfs_fs_mount_cmn_err(flags, "bad magic number"); |
1da177e4 LT |
228 | return XFS_ERROR(EWRONGFS); |
229 | } | |
230 | ||
231 | if (!XFS_SB_GOOD_VERSION(sbp)) { | |
764d1f89 | 232 | xfs_fs_mount_cmn_err(flags, "bad version"); |
1da177e4 LT |
233 | return XFS_ERROR(EWRONGFS); |
234 | } | |
235 | ||
236 | if (unlikely( | |
237 | sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) { | |
764d1f89 NS |
238 | xfs_fs_mount_cmn_err(flags, |
239 | "filesystem is marked as having an external log; " | |
240 | "specify logdev on the\nmount command line."); | |
241 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
242 | } |
243 | ||
244 | if (unlikely( | |
245 | sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) { | |
764d1f89 NS |
246 | xfs_fs_mount_cmn_err(flags, |
247 | "filesystem is marked as having an internal log; " | |
248 | "do not specify logdev on\nthe mount command line."); | |
249 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
250 | } |
251 | ||
252 | /* | |
253 | * More sanity checking. These were stolen directly from | |
254 | * xfs_repair. | |
255 | */ | |
256 | if (unlikely( | |
257 | sbp->sb_agcount <= 0 || | |
258 | sbp->sb_sectsize < XFS_MIN_SECTORSIZE || | |
259 | sbp->sb_sectsize > XFS_MAX_SECTORSIZE || | |
260 | sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG || | |
261 | sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG || | |
262 | sbp->sb_blocksize < XFS_MIN_BLOCKSIZE || | |
263 | sbp->sb_blocksize > XFS_MAX_BLOCKSIZE || | |
264 | sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG || | |
265 | sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG || | |
266 | sbp->sb_inodesize < XFS_DINODE_MIN_SIZE || | |
267 | sbp->sb_inodesize > XFS_DINODE_MAX_SIZE || | |
9f989c94 NS |
268 | sbp->sb_inodelog < XFS_DINODE_MIN_LOG || |
269 | sbp->sb_inodelog > XFS_DINODE_MAX_LOG || | |
270 | (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) || | |
1da177e4 LT |
271 | (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) || |
272 | (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) || | |
e50bd16f | 273 | (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) { |
764d1f89 | 274 | xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed"); |
1da177e4 LT |
275 | return XFS_ERROR(EFSCORRUPTED); |
276 | } | |
277 | ||
278 | /* | |
279 | * Sanity check AG count, size fields against data size field | |
280 | */ | |
281 | if (unlikely( | |
282 | sbp->sb_dblocks == 0 || | |
283 | sbp->sb_dblocks > | |
284 | (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks || | |
285 | sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) * | |
286 | sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) { | |
764d1f89 | 287 | xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed"); |
1da177e4 LT |
288 | return XFS_ERROR(EFSCORRUPTED); |
289 | } | |
290 | ||
291 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
292 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
293 | ||
294 | #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */ | |
295 | if (unlikely( | |
296 | (sbp->sb_dblocks >> (PAGE_SHIFT - sbp->sb_blocklog)) > ULONG_MAX || | |
297 | (sbp->sb_rblocks >> (PAGE_SHIFT - sbp->sb_blocklog)) > ULONG_MAX)) { | |
298 | #else /* Limited by UINT_MAX of sectors */ | |
299 | if (unlikely( | |
300 | (sbp->sb_dblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX || | |
301 | (sbp->sb_rblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX)) { | |
302 | #endif | |
764d1f89 NS |
303 | xfs_fs_mount_cmn_err(flags, |
304 | "file system too large to be mounted on this system."); | |
1da177e4 LT |
305 | return XFS_ERROR(E2BIG); |
306 | } | |
307 | ||
308 | if (unlikely(sbp->sb_inprogress)) { | |
764d1f89 | 309 | xfs_fs_mount_cmn_err(flags, "file system busy"); |
1da177e4 LT |
310 | return XFS_ERROR(EFSCORRUPTED); |
311 | } | |
312 | ||
de20614b NS |
313 | /* |
314 | * Version 1 directory format has never worked on Linux. | |
315 | */ | |
316 | if (unlikely(!XFS_SB_VERSION_HASDIRV2(sbp))) { | |
764d1f89 NS |
317 | xfs_fs_mount_cmn_err(flags, |
318 | "file system using version 1 directory format"); | |
de20614b NS |
319 | return XFS_ERROR(ENOSYS); |
320 | } | |
321 | ||
1da177e4 LT |
322 | /* |
323 | * Until this is fixed only page-sized or smaller data blocks work. | |
324 | */ | |
325 | if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) { | |
764d1f89 NS |
326 | xfs_fs_mount_cmn_err(flags, |
327 | "file system with blocksize %d bytes", | |
1da177e4 | 328 | sbp->sb_blocksize); |
764d1f89 NS |
329 | xfs_fs_mount_cmn_err(flags, |
330 | "only pagesize (%ld) or less will currently work.", | |
1da177e4 LT |
331 | PAGE_SIZE); |
332 | return XFS_ERROR(ENOSYS); | |
333 | } | |
334 | ||
335 | return 0; | |
336 | } | |
337 | ||
338 | xfs_agnumber_t | |
c11e2c36 | 339 | xfs_initialize_perag( |
b83bd138 | 340 | bhv_vfs_t *vfs, |
c11e2c36 NS |
341 | xfs_mount_t *mp, |
342 | xfs_agnumber_t agcount) | |
1da177e4 LT |
343 | { |
344 | xfs_agnumber_t index, max_metadata; | |
345 | xfs_perag_t *pag; | |
346 | xfs_agino_t agino; | |
347 | xfs_ino_t ino; | |
348 | xfs_sb_t *sbp = &mp->m_sb; | |
349 | xfs_ino_t max_inum = XFS_MAXINUMBER_32; | |
350 | ||
351 | /* Check to see if the filesystem can overflow 32 bit inodes */ | |
352 | agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0); | |
353 | ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino); | |
354 | ||
355 | /* Clear the mount flag if no inode can overflow 32 bits | |
356 | * on this filesystem, or if specifically requested.. | |
357 | */ | |
c11e2c36 | 358 | if ((vfs->vfs_flag & VFS_32BITINODES) && ino > max_inum) { |
1da177e4 LT |
359 | mp->m_flags |= XFS_MOUNT_32BITINODES; |
360 | } else { | |
361 | mp->m_flags &= ~XFS_MOUNT_32BITINODES; | |
362 | } | |
363 | ||
364 | /* If we can overflow then setup the ag headers accordingly */ | |
365 | if (mp->m_flags & XFS_MOUNT_32BITINODES) { | |
366 | /* Calculate how much should be reserved for inodes to | |
367 | * meet the max inode percentage. | |
368 | */ | |
369 | if (mp->m_maxicount) { | |
370 | __uint64_t icount; | |
371 | ||
372 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; | |
373 | do_div(icount, 100); | |
374 | icount += sbp->sb_agblocks - 1; | |
a749ee86 | 375 | do_div(icount, sbp->sb_agblocks); |
1da177e4 LT |
376 | max_metadata = icount; |
377 | } else { | |
378 | max_metadata = agcount; | |
379 | } | |
380 | for (index = 0; index < agcount; index++) { | |
381 | ino = XFS_AGINO_TO_INO(mp, index, agino); | |
382 | if (ino > max_inum) { | |
383 | index++; | |
384 | break; | |
385 | } | |
386 | ||
c41564b5 | 387 | /* This ag is preferred for inodes */ |
1da177e4 LT |
388 | pag = &mp->m_perag[index]; |
389 | pag->pagi_inodeok = 1; | |
390 | if (index < max_metadata) | |
391 | pag->pagf_metadata = 1; | |
392 | } | |
393 | } else { | |
394 | /* Setup default behavior for smaller filesystems */ | |
395 | for (index = 0; index < agcount; index++) { | |
396 | pag = &mp->m_perag[index]; | |
397 | pag->pagi_inodeok = 1; | |
398 | } | |
399 | } | |
400 | return index; | |
401 | } | |
402 | ||
403 | /* | |
404 | * xfs_xlatesb | |
405 | * | |
406 | * data - on disk version of sb | |
407 | * sb - a superblock | |
408 | * dir - conversion direction: <0 - convert sb to buf | |
409 | * >0 - convert buf to sb | |
410 | * fields - which fields to copy (bitmask) | |
411 | */ | |
412 | void | |
413 | xfs_xlatesb( | |
414 | void *data, | |
415 | xfs_sb_t *sb, | |
416 | int dir, | |
417 | __int64_t fields) | |
418 | { | |
419 | xfs_caddr_t buf_ptr; | |
420 | xfs_caddr_t mem_ptr; | |
421 | xfs_sb_field_t f; | |
422 | int first; | |
423 | int size; | |
424 | ||
425 | ASSERT(dir); | |
426 | ASSERT(fields); | |
427 | ||
428 | if (!fields) | |
429 | return; | |
430 | ||
431 | buf_ptr = (xfs_caddr_t)data; | |
432 | mem_ptr = (xfs_caddr_t)sb; | |
433 | ||
434 | while (fields) { | |
435 | f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields); | |
436 | first = xfs_sb_info[f].offset; | |
437 | size = xfs_sb_info[f + 1].offset - first; | |
438 | ||
439 | ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1); | |
440 | ||
441 | if (size == 1 || xfs_sb_info[f].type == 1) { | |
442 | if (dir > 0) { | |
443 | memcpy(mem_ptr + first, buf_ptr + first, size); | |
444 | } else { | |
445 | memcpy(buf_ptr + first, mem_ptr + first, size); | |
446 | } | |
447 | } else { | |
448 | switch (size) { | |
449 | case 2: | |
450 | INT_XLATE(*(__uint16_t*)(buf_ptr+first), | |
451 | *(__uint16_t*)(mem_ptr+first), | |
452 | dir, ARCH_CONVERT); | |
453 | break; | |
454 | case 4: | |
455 | INT_XLATE(*(__uint32_t*)(buf_ptr+first), | |
456 | *(__uint32_t*)(mem_ptr+first), | |
457 | dir, ARCH_CONVERT); | |
458 | break; | |
459 | case 8: | |
460 | INT_XLATE(*(__uint64_t*)(buf_ptr+first), | |
461 | *(__uint64_t*)(mem_ptr+first), dir, ARCH_CONVERT); | |
462 | break; | |
463 | default: | |
464 | ASSERT(0); | |
465 | } | |
466 | } | |
467 | ||
468 | fields &= ~(1LL << f); | |
469 | } | |
470 | } | |
471 | ||
472 | /* | |
473 | * xfs_readsb | |
474 | * | |
475 | * Does the initial read of the superblock. | |
476 | */ | |
477 | int | |
764d1f89 | 478 | xfs_readsb(xfs_mount_t *mp, int flags) |
1da177e4 LT |
479 | { |
480 | unsigned int sector_size; | |
481 | unsigned int extra_flags; | |
482 | xfs_buf_t *bp; | |
483 | xfs_sb_t *sbp; | |
484 | int error; | |
485 | ||
486 | ASSERT(mp->m_sb_bp == NULL); | |
487 | ASSERT(mp->m_ddev_targp != NULL); | |
488 | ||
489 | /* | |
490 | * Allocate a (locked) buffer to hold the superblock. | |
491 | * This will be kept around at all times to optimize | |
492 | * access to the superblock. | |
493 | */ | |
494 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
495 | extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED; | |
496 | ||
497 | bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR, | |
498 | BTOBB(sector_size), extra_flags); | |
499 | if (!bp || XFS_BUF_ISERROR(bp)) { | |
764d1f89 | 500 | xfs_fs_mount_cmn_err(flags, "SB read failed"); |
1da177e4 LT |
501 | error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM; |
502 | goto fail; | |
503 | } | |
504 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
505 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); | |
506 | ||
507 | /* | |
508 | * Initialize the mount structure from the superblock. | |
509 | * But first do some basic consistency checking. | |
510 | */ | |
511 | sbp = XFS_BUF_TO_SBP(bp); | |
512 | xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), 1, XFS_SB_ALL_BITS); | |
513 | ||
764d1f89 | 514 | error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags); |
1da177e4 | 515 | if (error) { |
764d1f89 | 516 | xfs_fs_mount_cmn_err(flags, "SB validate failed"); |
1da177e4 LT |
517 | goto fail; |
518 | } | |
519 | ||
520 | /* | |
521 | * We must be able to do sector-sized and sector-aligned IO. | |
522 | */ | |
523 | if (sector_size > mp->m_sb.sb_sectsize) { | |
764d1f89 NS |
524 | xfs_fs_mount_cmn_err(flags, |
525 | "device supports only %u byte sectors (not %u)", | |
1da177e4 LT |
526 | sector_size, mp->m_sb.sb_sectsize); |
527 | error = ENOSYS; | |
528 | goto fail; | |
529 | } | |
530 | ||
531 | /* | |
532 | * If device sector size is smaller than the superblock size, | |
533 | * re-read the superblock so the buffer is correctly sized. | |
534 | */ | |
535 | if (sector_size < mp->m_sb.sb_sectsize) { | |
536 | XFS_BUF_UNMANAGE(bp); | |
537 | xfs_buf_relse(bp); | |
538 | sector_size = mp->m_sb.sb_sectsize; | |
539 | bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR, | |
540 | BTOBB(sector_size), extra_flags); | |
541 | if (!bp || XFS_BUF_ISERROR(bp)) { | |
764d1f89 | 542 | xfs_fs_mount_cmn_err(flags, "SB re-read failed"); |
1da177e4 LT |
543 | error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM; |
544 | goto fail; | |
545 | } | |
546 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
547 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); | |
548 | } | |
549 | ||
8d280b98 DC |
550 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
551 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
552 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
553 | ||
1da177e4 LT |
554 | mp->m_sb_bp = bp; |
555 | xfs_buf_relse(bp); | |
556 | ASSERT(XFS_BUF_VALUSEMA(bp) > 0); | |
557 | return 0; | |
558 | ||
559 | fail: | |
560 | if (bp) { | |
561 | XFS_BUF_UNMANAGE(bp); | |
562 | xfs_buf_relse(bp); | |
563 | } | |
564 | return error; | |
565 | } | |
566 | ||
567 | ||
568 | /* | |
569 | * xfs_mount_common | |
570 | * | |
571 | * Mount initialization code establishing various mount | |
572 | * fields from the superblock associated with the given | |
573 | * mount structure | |
574 | */ | |
ba0f32d4 | 575 | STATIC void |
1da177e4 LT |
576 | xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp) |
577 | { | |
578 | int i; | |
579 | ||
580 | mp->m_agfrotor = mp->m_agirotor = 0; | |
581 | spinlock_init(&mp->m_agirotor_lock, "m_agirotor_lock"); | |
582 | mp->m_maxagi = mp->m_sb.sb_agcount; | |
583 | mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG; | |
584 | mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT; | |
585 | mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT; | |
586 | mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1; | |
587 | mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog; | |
588 | mp->m_litino = sbp->sb_inodesize - | |
589 | ((uint)sizeof(xfs_dinode_core_t) + (uint)sizeof(xfs_agino_t)); | |
590 | mp->m_blockmask = sbp->sb_blocksize - 1; | |
591 | mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG; | |
592 | mp->m_blockwmask = mp->m_blockwsize - 1; | |
593 | INIT_LIST_HEAD(&mp->m_del_inodes); | |
594 | ||
595 | /* | |
596 | * Setup for attributes, in case they get created. | |
597 | * This value is for inodes getting attributes for the first time, | |
598 | * the per-inode value is for old attribute values. | |
599 | */ | |
600 | ASSERT(sbp->sb_inodesize >= 256 && sbp->sb_inodesize <= 2048); | |
601 | switch (sbp->sb_inodesize) { | |
602 | case 256: | |
d8cc890d NS |
603 | mp->m_attroffset = XFS_LITINO(mp) - |
604 | XFS_BMDR_SPACE_CALC(MINABTPTRS); | |
1da177e4 LT |
605 | break; |
606 | case 512: | |
607 | case 1024: | |
608 | case 2048: | |
d8cc890d | 609 | mp->m_attroffset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS); |
1da177e4 LT |
610 | break; |
611 | default: | |
612 | ASSERT(0); | |
613 | } | |
614 | ASSERT(mp->m_attroffset < XFS_LITINO(mp)); | |
615 | ||
616 | for (i = 0; i < 2; i++) { | |
617 | mp->m_alloc_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize, | |
618 | xfs_alloc, i == 0); | |
619 | mp->m_alloc_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize, | |
620 | xfs_alloc, i == 0); | |
621 | } | |
622 | for (i = 0; i < 2; i++) { | |
623 | mp->m_bmap_dmxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize, | |
624 | xfs_bmbt, i == 0); | |
625 | mp->m_bmap_dmnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize, | |
626 | xfs_bmbt, i == 0); | |
627 | } | |
628 | for (i = 0; i < 2; i++) { | |
629 | mp->m_inobt_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize, | |
630 | xfs_inobt, i == 0); | |
631 | mp->m_inobt_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize, | |
632 | xfs_inobt, i == 0); | |
633 | } | |
634 | ||
635 | mp->m_bsize = XFS_FSB_TO_BB(mp, 1); | |
636 | mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK, | |
637 | sbp->sb_inopblock); | |
638 | mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog; | |
639 | } | |
640 | /* | |
641 | * xfs_mountfs | |
642 | * | |
643 | * This function does the following on an initial mount of a file system: | |
644 | * - reads the superblock from disk and init the mount struct | |
645 | * - if we're a 32-bit kernel, do a size check on the superblock | |
646 | * so we don't mount terabyte filesystems | |
647 | * - init mount struct realtime fields | |
648 | * - allocate inode hash table for fs | |
649 | * - init directory manager | |
650 | * - perform recovery and init the log manager | |
651 | */ | |
652 | int | |
653 | xfs_mountfs( | |
b83bd138 | 654 | bhv_vfs_t *vfsp, |
1da177e4 LT |
655 | xfs_mount_t *mp, |
656 | int mfsi_flags) | |
657 | { | |
658 | xfs_buf_t *bp; | |
659 | xfs_sb_t *sbp = &(mp->m_sb); | |
660 | xfs_inode_t *rip; | |
661 | vnode_t *rvp = NULL; | |
662 | int readio_log, writeio_log; | |
663 | xfs_daddr_t d; | |
664 | __uint64_t ret64; | |
665 | __int64_t update_flags; | |
666 | uint quotamount, quotaflags; | |
667 | int agno; | |
668 | int uuid_mounted = 0; | |
669 | int error = 0; | |
670 | ||
671 | if (mp->m_sb_bp == NULL) { | |
764d1f89 | 672 | if ((error = xfs_readsb(mp, mfsi_flags))) { |
014c2544 | 673 | return error; |
1da177e4 LT |
674 | } |
675 | } | |
676 | xfs_mount_common(mp, sbp); | |
677 | ||
678 | /* | |
679 | * Check if sb_agblocks is aligned at stripe boundary | |
680 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
681 | * allocator alignment is within an ag, therefore ag has | |
682 | * to be aligned at stripe boundary. | |
683 | */ | |
684 | update_flags = 0LL; | |
685 | if (mp->m_dalign && !(mfsi_flags & XFS_MFSI_SECOND)) { | |
686 | /* | |
687 | * If stripe unit and stripe width are not multiples | |
688 | * of the fs blocksize turn off alignment. | |
689 | */ | |
690 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
691 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
692 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
693 | cmn_err(CE_WARN, | |
694 | "XFS: alignment check 1 failed"); | |
695 | error = XFS_ERROR(EINVAL); | |
696 | goto error1; | |
697 | } | |
698 | mp->m_dalign = mp->m_swidth = 0; | |
699 | } else { | |
700 | /* | |
701 | * Convert the stripe unit and width to FSBs. | |
702 | */ | |
703 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
704 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
705 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
706 | error = XFS_ERROR(EINVAL); | |
707 | goto error1; | |
708 | } | |
709 | xfs_fs_cmn_err(CE_WARN, mp, | |
710 | "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)", | |
711 | mp->m_dalign, mp->m_swidth, | |
712 | sbp->sb_agblocks); | |
713 | ||
714 | mp->m_dalign = 0; | |
715 | mp->m_swidth = 0; | |
716 | } else if (mp->m_dalign) { | |
717 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
718 | } else { | |
719 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
720 | xfs_fs_cmn_err(CE_WARN, mp, | |
721 | "stripe alignment turned off: sunit(%d) less than bsize(%d)", | |
722 | mp->m_dalign, | |
723 | mp->m_blockmask +1); | |
724 | error = XFS_ERROR(EINVAL); | |
725 | goto error1; | |
726 | } | |
727 | mp->m_swidth = 0; | |
728 | } | |
729 | } | |
730 | ||
731 | /* | |
732 | * Update superblock with new values | |
733 | * and log changes | |
734 | */ | |
735 | if (XFS_SB_VERSION_HASDALIGN(sbp)) { | |
736 | if (sbp->sb_unit != mp->m_dalign) { | |
737 | sbp->sb_unit = mp->m_dalign; | |
738 | update_flags |= XFS_SB_UNIT; | |
739 | } | |
740 | if (sbp->sb_width != mp->m_swidth) { | |
741 | sbp->sb_width = mp->m_swidth; | |
742 | update_flags |= XFS_SB_WIDTH; | |
743 | } | |
744 | } | |
745 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
746 | XFS_SB_VERSION_HASDALIGN(&mp->m_sb)) { | |
747 | mp->m_dalign = sbp->sb_unit; | |
748 | mp->m_swidth = sbp->sb_width; | |
749 | } | |
750 | ||
751 | xfs_alloc_compute_maxlevels(mp); | |
752 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
753 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
754 | xfs_ialloc_compute_maxlevels(mp); | |
755 | ||
756 | if (sbp->sb_imax_pct) { | |
757 | __uint64_t icount; | |
758 | ||
759 | /* Make sure the maximum inode count is a multiple of the | |
760 | * units we allocate inodes in. | |
761 | */ | |
762 | ||
763 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; | |
764 | do_div(icount, 100); | |
765 | do_div(icount, mp->m_ialloc_blks); | |
766 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << | |
767 | sbp->sb_inopblog; | |
768 | } else | |
769 | mp->m_maxicount = 0; | |
770 | ||
771 | mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog); | |
772 | ||
773 | /* | |
774 | * XFS uses the uuid from the superblock as the unique | |
775 | * identifier for fsid. We can not use the uuid from the volume | |
776 | * since a single partition filesystem is identical to a single | |
777 | * partition volume/filesystem. | |
778 | */ | |
779 | if ((mfsi_flags & XFS_MFSI_SECOND) == 0 && | |
780 | (mp->m_flags & XFS_MOUNT_NOUUID) == 0) { | |
781 | if (xfs_uuid_mount(mp)) { | |
782 | error = XFS_ERROR(EINVAL); | |
783 | goto error1; | |
784 | } | |
785 | uuid_mounted=1; | |
786 | ret64 = uuid_hash64(&sbp->sb_uuid); | |
787 | memcpy(&vfsp->vfs_fsid, &ret64, sizeof(ret64)); | |
788 | } | |
789 | ||
790 | /* | |
791 | * Set the default minimum read and write sizes unless | |
792 | * already specified in a mount option. | |
793 | * We use smaller I/O sizes when the file system | |
794 | * is being used for NFS service (wsync mount option). | |
795 | */ | |
796 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { | |
797 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
798 | readio_log = XFS_WSYNC_READIO_LOG; | |
799 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
800 | } else { | |
801 | readio_log = XFS_READIO_LOG_LARGE; | |
802 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
803 | } | |
804 | } else { | |
805 | readio_log = mp->m_readio_log; | |
806 | writeio_log = mp->m_writeio_log; | |
807 | } | |
808 | ||
809 | /* | |
810 | * Set the number of readahead buffers to use based on | |
811 | * physical memory size. | |
812 | */ | |
813 | if (xfs_physmem <= 4096) /* <= 16MB */ | |
814 | mp->m_nreadaheads = XFS_RW_NREADAHEAD_16MB; | |
815 | else if (xfs_physmem <= 8192) /* <= 32MB */ | |
816 | mp->m_nreadaheads = XFS_RW_NREADAHEAD_32MB; | |
817 | else | |
818 | mp->m_nreadaheads = XFS_RW_NREADAHEAD_K32; | |
819 | if (sbp->sb_blocklog > readio_log) { | |
820 | mp->m_readio_log = sbp->sb_blocklog; | |
821 | } else { | |
822 | mp->m_readio_log = readio_log; | |
823 | } | |
824 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
825 | if (sbp->sb_blocklog > writeio_log) { | |
826 | mp->m_writeio_log = sbp->sb_blocklog; | |
827 | } else { | |
828 | mp->m_writeio_log = writeio_log; | |
829 | } | |
830 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
831 | ||
832 | /* | |
833 | * Set the inode cluster size based on the physical memory | |
834 | * size. This may still be overridden by the file system | |
835 | * block size if it is larger than the chosen cluster size. | |
836 | */ | |
837 | if (xfs_physmem <= btoc(32 * 1024 * 1024)) { /* <= 32 MB */ | |
838 | mp->m_inode_cluster_size = XFS_INODE_SMALL_CLUSTER_SIZE; | |
839 | } else { | |
840 | mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; | |
841 | } | |
842 | /* | |
843 | * Set whether we're using inode alignment. | |
844 | */ | |
845 | if (XFS_SB_VERSION_HASALIGN(&mp->m_sb) && | |
846 | mp->m_sb.sb_inoalignmt >= | |
847 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
848 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; | |
849 | else | |
850 | mp->m_inoalign_mask = 0; | |
851 | /* | |
852 | * If we are using stripe alignment, check whether | |
853 | * the stripe unit is a multiple of the inode alignment | |
854 | */ | |
855 | if (mp->m_dalign && mp->m_inoalign_mask && | |
856 | !(mp->m_dalign & mp->m_inoalign_mask)) | |
857 | mp->m_sinoalign = mp->m_dalign; | |
858 | else | |
859 | mp->m_sinoalign = 0; | |
860 | /* | |
861 | * Check that the data (and log if separate) are an ok size. | |
862 | */ | |
863 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); | |
864 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
865 | cmn_err(CE_WARN, "XFS: size check 1 failed"); | |
866 | error = XFS_ERROR(E2BIG); | |
867 | goto error1; | |
868 | } | |
869 | error = xfs_read_buf(mp, mp->m_ddev_targp, | |
870 | d - XFS_FSS_TO_BB(mp, 1), | |
871 | XFS_FSS_TO_BB(mp, 1), 0, &bp); | |
872 | if (!error) { | |
873 | xfs_buf_relse(bp); | |
874 | } else { | |
875 | cmn_err(CE_WARN, "XFS: size check 2 failed"); | |
876 | if (error == ENOSPC) { | |
877 | error = XFS_ERROR(E2BIG); | |
878 | } | |
879 | goto error1; | |
880 | } | |
881 | ||
882 | if (((mfsi_flags & XFS_MFSI_CLIENT) == 0) && | |
883 | mp->m_logdev_targp != mp->m_ddev_targp) { | |
884 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); | |
885 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
886 | cmn_err(CE_WARN, "XFS: size check 3 failed"); | |
887 | error = XFS_ERROR(E2BIG); | |
888 | goto error1; | |
889 | } | |
890 | error = xfs_read_buf(mp, mp->m_logdev_targp, | |
891 | d - XFS_FSB_TO_BB(mp, 1), | |
892 | XFS_FSB_TO_BB(mp, 1), 0, &bp); | |
893 | if (!error) { | |
894 | xfs_buf_relse(bp); | |
895 | } else { | |
896 | cmn_err(CE_WARN, "XFS: size check 3 failed"); | |
897 | if (error == ENOSPC) { | |
898 | error = XFS_ERROR(E2BIG); | |
899 | } | |
900 | goto error1; | |
901 | } | |
902 | } | |
903 | ||
904 | /* | |
905 | * Initialize realtime fields in the mount structure | |
906 | */ | |
907 | if ((error = xfs_rtmount_init(mp))) { | |
908 | cmn_err(CE_WARN, "XFS: RT mount failed"); | |
909 | goto error1; | |
910 | } | |
911 | ||
912 | /* | |
913 | * For client case we are done now | |
914 | */ | |
915 | if (mfsi_flags & XFS_MFSI_CLIENT) { | |
014c2544 | 916 | return 0; |
1da177e4 LT |
917 | } |
918 | ||
919 | /* | |
920 | * Copies the low order bits of the timestamp and the randomly | |
921 | * set "sequence" number out of a UUID. | |
922 | */ | |
923 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); | |
924 | ||
925 | /* | |
926 | * The vfs structure needs to have a file system independent | |
927 | * way of checking for the invariant file system ID. Since it | |
928 | * can't look at mount structures it has a pointer to the data | |
929 | * in the mount structure. | |
930 | * | |
931 | * File systems that don't support user level file handles (i.e. | |
932 | * all of them except for XFS) will leave vfs_altfsid as NULL. | |
933 | */ | |
934 | vfsp->vfs_altfsid = (xfs_fsid_t *)mp->m_fixedfsid; | |
935 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ | |
936 | ||
937 | /* | |
938 | * Select the right directory manager. | |
939 | */ | |
940 | mp->m_dirops = | |
941 | XFS_SB_VERSION_HASDIRV2(&mp->m_sb) ? | |
942 | xfsv2_dirops : | |
943 | xfsv1_dirops; | |
944 | ||
945 | /* | |
946 | * Initialize directory manager's entries. | |
947 | */ | |
948 | XFS_DIR_MOUNT(mp); | |
949 | ||
950 | /* | |
951 | * Initialize the attribute manager's entries. | |
952 | */ | |
953 | mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100; | |
954 | ||
955 | /* | |
956 | * Initialize the precomputed transaction reservations values. | |
957 | */ | |
958 | xfs_trans_init(mp); | |
959 | ||
960 | /* | |
961 | * Allocate and initialize the inode hash table for this | |
962 | * file system. | |
963 | */ | |
964 | xfs_ihash_init(mp); | |
965 | xfs_chash_init(mp); | |
966 | ||
967 | /* | |
968 | * Allocate and initialize the per-ag data. | |
969 | */ | |
970 | init_rwsem(&mp->m_peraglock); | |
971 | mp->m_perag = | |
972 | kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t), KM_SLEEP); | |
973 | ||
c11e2c36 | 974 | mp->m_maxagi = xfs_initialize_perag(vfsp, mp, sbp->sb_agcount); |
1da177e4 LT |
975 | |
976 | /* | |
977 | * log's mount-time initialization. Perform 1st part recovery if needed | |
978 | */ | |
979 | if (likely(sbp->sb_logblocks > 0)) { /* check for volume case */ | |
980 | error = xfs_log_mount(mp, mp->m_logdev_targp, | |
981 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
982 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
983 | if (error) { | |
984 | cmn_err(CE_WARN, "XFS: log mount failed"); | |
985 | goto error2; | |
986 | } | |
987 | } else { /* No log has been defined */ | |
988 | cmn_err(CE_WARN, "XFS: no log defined"); | |
989 | XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW, mp); | |
990 | error = XFS_ERROR(EFSCORRUPTED); | |
991 | goto error2; | |
992 | } | |
993 | ||
994 | /* | |
995 | * Get and sanity-check the root inode. | |
996 | * Save the pointer to it in the mount structure. | |
997 | */ | |
998 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0); | |
999 | if (error) { | |
1000 | cmn_err(CE_WARN, "XFS: failed to read root inode"); | |
1001 | goto error3; | |
1002 | } | |
1003 | ||
1004 | ASSERT(rip != NULL); | |
1005 | rvp = XFS_ITOV(rip); | |
1006 | ||
1007 | if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) { | |
1008 | cmn_err(CE_WARN, "XFS: corrupted root inode"); | |
b6574520 NS |
1009 | cmn_err(CE_WARN, "Device %s - root %llu is not a directory", |
1010 | XFS_BUFTARG_NAME(mp->m_ddev_targp), | |
1011 | (unsigned long long)rip->i_ino); | |
1da177e4 LT |
1012 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
1013 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
1014 | mp); | |
1015 | error = XFS_ERROR(EFSCORRUPTED); | |
1016 | goto error4; | |
1017 | } | |
1018 | mp->m_rootip = rip; /* save it */ | |
1019 | ||
1020 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
1021 | ||
1022 | /* | |
1023 | * Initialize realtime inode pointers in the mount structure | |
1024 | */ | |
1025 | if ((error = xfs_rtmount_inodes(mp))) { | |
1026 | /* | |
1027 | * Free up the root inode. | |
1028 | */ | |
1029 | cmn_err(CE_WARN, "XFS: failed to read RT inodes"); | |
1030 | goto error4; | |
1031 | } | |
1032 | ||
1033 | /* | |
1034 | * If fs is not mounted readonly, then update the superblock | |
1035 | * unit and width changes. | |
1036 | */ | |
1037 | if (update_flags && !(vfsp->vfs_flag & VFS_RDONLY)) | |
1038 | xfs_mount_log_sbunit(mp, update_flags); | |
1039 | ||
1040 | /* | |
1041 | * Initialise the XFS quota management subsystem for this mount | |
1042 | */ | |
1043 | if ((error = XFS_QM_INIT(mp, "amount, "aflags))) | |
1044 | goto error4; | |
1045 | ||
1046 | /* | |
1047 | * Finish recovering the file system. This part needed to be | |
1048 | * delayed until after the root and real-time bitmap inodes | |
1049 | * were consistently read in. | |
1050 | */ | |
1051 | error = xfs_log_mount_finish(mp, mfsi_flags); | |
1052 | if (error) { | |
1053 | cmn_err(CE_WARN, "XFS: log mount finish failed"); | |
1054 | goto error4; | |
1055 | } | |
1056 | ||
1057 | /* | |
1058 | * Complete the quota initialisation, post-log-replay component. | |
1059 | */ | |
1060 | if ((error = XFS_QM_MOUNT(mp, quotamount, quotaflags, mfsi_flags))) | |
1061 | goto error4; | |
1062 | ||
1063 | return 0; | |
1064 | ||
1065 | error4: | |
1066 | /* | |
1067 | * Free up the root inode. | |
1068 | */ | |
1069 | VN_RELE(rvp); | |
1070 | error3: | |
1071 | xfs_log_unmount_dealloc(mp); | |
1072 | error2: | |
1073 | xfs_ihash_free(mp); | |
1074 | xfs_chash_free(mp); | |
1075 | for (agno = 0; agno < sbp->sb_agcount; agno++) | |
1076 | if (mp->m_perag[agno].pagb_list) | |
1077 | kmem_free(mp->m_perag[agno].pagb_list, | |
1078 | sizeof(xfs_perag_busy_t) * XFS_PAGB_NUM_SLOTS); | |
1079 | kmem_free(mp->m_perag, sbp->sb_agcount * sizeof(xfs_perag_t)); | |
1080 | mp->m_perag = NULL; | |
1081 | /* FALLTHROUGH */ | |
1082 | error1: | |
1083 | if (uuid_mounted) | |
1084 | xfs_uuid_unmount(mp); | |
1085 | xfs_freesb(mp); | |
1086 | return error; | |
1087 | } | |
1088 | ||
1089 | /* | |
1090 | * xfs_unmountfs | |
1091 | * | |
1092 | * This flushes out the inodes,dquots and the superblock, unmounts the | |
1093 | * log and makes sure that incore structures are freed. | |
1094 | */ | |
1095 | int | |
1096 | xfs_unmountfs(xfs_mount_t *mp, struct cred *cr) | |
1097 | { | |
b83bd138 | 1098 | struct bhv_vfs *vfsp = XFS_MTOVFS(mp); |
1da177e4 LT |
1099 | #if defined(DEBUG) || defined(INDUCE_IO_ERROR) |
1100 | int64_t fsid; | |
1101 | #endif | |
1102 | ||
efa80278 | 1103 | xfs_iflush_all(mp); |
1da177e4 | 1104 | |
ee2a4f7c | 1105 | XFS_QM_DQPURGEALL(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_UMOUNTING); |
1da177e4 LT |
1106 | |
1107 | /* | |
1108 | * Flush out the log synchronously so that we know for sure | |
1109 | * that nothing is pinned. This is important because bflush() | |
1110 | * will skip pinned buffers. | |
1111 | */ | |
1112 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC); | |
1113 | ||
1114 | xfs_binval(mp->m_ddev_targp); | |
1115 | if (mp->m_rtdev_targp) { | |
1116 | xfs_binval(mp->m_rtdev_targp); | |
1117 | } | |
1118 | ||
1119 | xfs_unmountfs_writesb(mp); | |
1120 | ||
1121 | xfs_unmountfs_wait(mp); /* wait for async bufs */ | |
1122 | ||
1123 | xfs_log_unmount(mp); /* Done! No more fs ops. */ | |
1124 | ||
1125 | xfs_freesb(mp); | |
1126 | ||
1127 | /* | |
1128 | * All inodes from this mount point should be freed. | |
1129 | */ | |
1130 | ASSERT(mp->m_inodes == NULL); | |
1131 | ||
1da177e4 LT |
1132 | xfs_unmountfs_close(mp, cr); |
1133 | if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0) | |
1134 | xfs_uuid_unmount(mp); | |
1135 | ||
1136 | #if defined(DEBUG) || defined(INDUCE_IO_ERROR) | |
1137 | /* | |
1138 | * clear all error tags on this filesystem | |
1139 | */ | |
1140 | memcpy(&fsid, &vfsp->vfs_fsid, sizeof(int64_t)); | |
1141 | xfs_errortag_clearall_umount(fsid, mp->m_fsname, 0); | |
1142 | #endif | |
1143 | XFS_IODONE(vfsp); | |
1144 | xfs_mount_free(mp, 1); | |
1145 | return 0; | |
1146 | } | |
1147 | ||
1148 | void | |
1149 | xfs_unmountfs_close(xfs_mount_t *mp, struct cred *cr) | |
1150 | { | |
1151 | if (mp->m_logdev_targp != mp->m_ddev_targp) | |
1152 | xfs_free_buftarg(mp->m_logdev_targp, 1); | |
1153 | if (mp->m_rtdev_targp) | |
1154 | xfs_free_buftarg(mp->m_rtdev_targp, 1); | |
1155 | xfs_free_buftarg(mp->m_ddev_targp, 0); | |
1156 | } | |
1157 | ||
ba0f32d4 | 1158 | STATIC void |
1da177e4 LT |
1159 | xfs_unmountfs_wait(xfs_mount_t *mp) |
1160 | { | |
1161 | if (mp->m_logdev_targp != mp->m_ddev_targp) | |
1162 | xfs_wait_buftarg(mp->m_logdev_targp); | |
1163 | if (mp->m_rtdev_targp) | |
1164 | xfs_wait_buftarg(mp->m_rtdev_targp); | |
1165 | xfs_wait_buftarg(mp->m_ddev_targp); | |
1166 | } | |
1167 | ||
1168 | int | |
1169 | xfs_unmountfs_writesb(xfs_mount_t *mp) | |
1170 | { | |
1171 | xfs_buf_t *sbp; | |
1172 | xfs_sb_t *sb; | |
1173 | int error = 0; | |
1174 | ||
1175 | /* | |
1176 | * skip superblock write if fs is read-only, or | |
1177 | * if we are doing a forced umount. | |
1178 | */ | |
1179 | sbp = xfs_getsb(mp, 0); | |
1180 | if (!(XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY || | |
1181 | XFS_FORCED_SHUTDOWN(mp))) { | |
8d280b98 DC |
1182 | |
1183 | xfs_icsb_sync_counters(mp); | |
1184 | ||
1da177e4 LT |
1185 | /* |
1186 | * mark shared-readonly if desired | |
1187 | */ | |
1188 | sb = XFS_BUF_TO_SBP(sbp); | |
1189 | if (mp->m_mk_sharedro) { | |
1190 | if (!(sb->sb_flags & XFS_SBF_READONLY)) | |
1191 | sb->sb_flags |= XFS_SBF_READONLY; | |
1192 | if (!XFS_SB_VERSION_HASSHARED(sb)) | |
1193 | XFS_SB_VERSION_ADDSHARED(sb); | |
1194 | xfs_fs_cmn_err(CE_NOTE, mp, | |
1195 | "Unmounting, marking shared read-only"); | |
1196 | } | |
1197 | XFS_BUF_UNDONE(sbp); | |
1198 | XFS_BUF_UNREAD(sbp); | |
1199 | XFS_BUF_UNDELAYWRITE(sbp); | |
1200 | XFS_BUF_WRITE(sbp); | |
1201 | XFS_BUF_UNASYNC(sbp); | |
1202 | ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp); | |
1203 | xfsbdstrat(mp, sbp); | |
1204 | /* Nevermind errors we might get here. */ | |
1205 | error = xfs_iowait(sbp); | |
1206 | if (error) | |
1207 | xfs_ioerror_alert("xfs_unmountfs_writesb", | |
1208 | mp, sbp, XFS_BUF_ADDR(sbp)); | |
1209 | if (error && mp->m_mk_sharedro) | |
1210 | xfs_fs_cmn_err(CE_ALERT, mp, "Superblock write error detected while unmounting. Filesystem may not be marked shared readonly"); | |
1211 | } | |
1212 | xfs_buf_relse(sbp); | |
014c2544 | 1213 | return error; |
1da177e4 LT |
1214 | } |
1215 | ||
1216 | /* | |
1217 | * xfs_mod_sb() can be used to copy arbitrary changes to the | |
1218 | * in-core superblock into the superblock buffer to be logged. | |
1219 | * It does not provide the higher level of locking that is | |
1220 | * needed to protect the in-core superblock from concurrent | |
1221 | * access. | |
1222 | */ | |
1223 | void | |
1224 | xfs_mod_sb(xfs_trans_t *tp, __int64_t fields) | |
1225 | { | |
1226 | xfs_buf_t *bp; | |
1227 | int first; | |
1228 | int last; | |
1229 | xfs_mount_t *mp; | |
1230 | xfs_sb_t *sbp; | |
1231 | xfs_sb_field_t f; | |
1232 | ||
1233 | ASSERT(fields); | |
1234 | if (!fields) | |
1235 | return; | |
1236 | mp = tp->t_mountp; | |
1237 | bp = xfs_trans_getsb(tp, mp, 0); | |
1238 | sbp = XFS_BUF_TO_SBP(bp); | |
1239 | first = sizeof(xfs_sb_t); | |
1240 | last = 0; | |
1241 | ||
1242 | /* translate/copy */ | |
1243 | ||
1244 | xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), -1, fields); | |
1245 | ||
1246 | /* find modified range */ | |
1247 | ||
1248 | f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields); | |
1249 | ASSERT((1LL << f) & XFS_SB_MOD_BITS); | |
1250 | first = xfs_sb_info[f].offset; | |
1251 | ||
1252 | f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields); | |
1253 | ASSERT((1LL << f) & XFS_SB_MOD_BITS); | |
1254 | last = xfs_sb_info[f + 1].offset - 1; | |
1255 | ||
1256 | xfs_trans_log_buf(tp, bp, first, last); | |
1257 | } | |
d210a28c YL |
1258 | |
1259 | /* | |
1260 | * In order to avoid ENOSPC-related deadlock caused by | |
1261 | * out-of-order locking of AGF buffer (PV 947395), we place | |
1262 | * constraints on the relationship among actual allocations for | |
1263 | * data blocks, freelist blocks, and potential file data bmap | |
1264 | * btree blocks. However, these restrictions may result in no | |
1265 | * actual space allocated for a delayed extent, for example, a data | |
1266 | * block in a certain AG is allocated but there is no additional | |
1267 | * block for the additional bmap btree block due to a split of the | |
1268 | * bmap btree of the file. The result of this may lead to an | |
1269 | * infinite loop in xfssyncd when the file gets flushed to disk and | |
1270 | * all delayed extents need to be actually allocated. To get around | |
1271 | * this, we explicitly set aside a few blocks which will not be | |
1272 | * reserved in delayed allocation. Considering the minimum number of | |
1273 | * needed freelist blocks is 4 fsbs, a potential split of file's bmap | |
1274 | * btree requires 1 fsb, so we set the number of set-aside blocks to 8. | |
1275 | */ | |
1276 | #define SET_ASIDE_BLOCKS 8 | |
1277 | ||
1da177e4 LT |
1278 | /* |
1279 | * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply | |
1280 | * a delta to a specified field in the in-core superblock. Simply | |
1281 | * switch on the field indicated and apply the delta to that field. | |
1282 | * Fields are not allowed to dip below zero, so if the delta would | |
1283 | * do this do not apply it and return EINVAL. | |
1284 | * | |
1285 | * The SB_LOCK must be held when this routine is called. | |
1286 | */ | |
8d280b98 | 1287 | int |
1da177e4 LT |
1288 | xfs_mod_incore_sb_unlocked(xfs_mount_t *mp, xfs_sb_field_t field, |
1289 | int delta, int rsvd) | |
1290 | { | |
1291 | int scounter; /* short counter for 32 bit fields */ | |
1292 | long long lcounter; /* long counter for 64 bit fields */ | |
1293 | long long res_used, rem; | |
1294 | ||
1295 | /* | |
1296 | * With the in-core superblock spin lock held, switch | |
1297 | * on the indicated field. Apply the delta to the | |
1298 | * proper field. If the fields value would dip below | |
1299 | * 0, then do not apply the delta and return EINVAL. | |
1300 | */ | |
1301 | switch (field) { | |
1302 | case XFS_SBS_ICOUNT: | |
1303 | lcounter = (long long)mp->m_sb.sb_icount; | |
1304 | lcounter += delta; | |
1305 | if (lcounter < 0) { | |
1306 | ASSERT(0); | |
014c2544 | 1307 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1308 | } |
1309 | mp->m_sb.sb_icount = lcounter; | |
014c2544 | 1310 | return 0; |
1da177e4 LT |
1311 | case XFS_SBS_IFREE: |
1312 | lcounter = (long long)mp->m_sb.sb_ifree; | |
1313 | lcounter += delta; | |
1314 | if (lcounter < 0) { | |
1315 | ASSERT(0); | |
014c2544 | 1316 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1317 | } |
1318 | mp->m_sb.sb_ifree = lcounter; | |
014c2544 | 1319 | return 0; |
1da177e4 LT |
1320 | case XFS_SBS_FDBLOCKS: |
1321 | ||
d210a28c | 1322 | lcounter = (long long)mp->m_sb.sb_fdblocks - SET_ASIDE_BLOCKS; |
1da177e4 LT |
1323 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); |
1324 | ||
1325 | if (delta > 0) { /* Putting blocks back */ | |
1326 | if (res_used > delta) { | |
1327 | mp->m_resblks_avail += delta; | |
1328 | } else { | |
1329 | rem = delta - res_used; | |
1330 | mp->m_resblks_avail = mp->m_resblks; | |
1331 | lcounter += rem; | |
1332 | } | |
1333 | } else { /* Taking blocks away */ | |
1334 | ||
1335 | lcounter += delta; | |
1336 | ||
1337 | /* | |
1338 | * If were out of blocks, use any available reserved blocks if | |
1339 | * were allowed to. | |
1340 | */ | |
1341 | ||
1342 | if (lcounter < 0) { | |
1343 | if (rsvd) { | |
1344 | lcounter = (long long)mp->m_resblks_avail + delta; | |
1345 | if (lcounter < 0) { | |
014c2544 | 1346 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1347 | } |
1348 | mp->m_resblks_avail = lcounter; | |
014c2544 | 1349 | return 0; |
1da177e4 | 1350 | } else { /* not reserved */ |
014c2544 | 1351 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1352 | } |
1353 | } | |
1354 | } | |
1355 | ||
d210a28c | 1356 | mp->m_sb.sb_fdblocks = lcounter + SET_ASIDE_BLOCKS; |
014c2544 | 1357 | return 0; |
1da177e4 LT |
1358 | case XFS_SBS_FREXTENTS: |
1359 | lcounter = (long long)mp->m_sb.sb_frextents; | |
1360 | lcounter += delta; | |
1361 | if (lcounter < 0) { | |
014c2544 | 1362 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1363 | } |
1364 | mp->m_sb.sb_frextents = lcounter; | |
014c2544 | 1365 | return 0; |
1da177e4 LT |
1366 | case XFS_SBS_DBLOCKS: |
1367 | lcounter = (long long)mp->m_sb.sb_dblocks; | |
1368 | lcounter += delta; | |
1369 | if (lcounter < 0) { | |
1370 | ASSERT(0); | |
014c2544 | 1371 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1372 | } |
1373 | mp->m_sb.sb_dblocks = lcounter; | |
014c2544 | 1374 | return 0; |
1da177e4 LT |
1375 | case XFS_SBS_AGCOUNT: |
1376 | scounter = mp->m_sb.sb_agcount; | |
1377 | scounter += delta; | |
1378 | if (scounter < 0) { | |
1379 | ASSERT(0); | |
014c2544 | 1380 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1381 | } |
1382 | mp->m_sb.sb_agcount = scounter; | |
014c2544 | 1383 | return 0; |
1da177e4 LT |
1384 | case XFS_SBS_IMAX_PCT: |
1385 | scounter = mp->m_sb.sb_imax_pct; | |
1386 | scounter += delta; | |
1387 | if (scounter < 0) { | |
1388 | ASSERT(0); | |
014c2544 | 1389 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1390 | } |
1391 | mp->m_sb.sb_imax_pct = scounter; | |
014c2544 | 1392 | return 0; |
1da177e4 LT |
1393 | case XFS_SBS_REXTSIZE: |
1394 | scounter = mp->m_sb.sb_rextsize; | |
1395 | scounter += delta; | |
1396 | if (scounter < 0) { | |
1397 | ASSERT(0); | |
014c2544 | 1398 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1399 | } |
1400 | mp->m_sb.sb_rextsize = scounter; | |
014c2544 | 1401 | return 0; |
1da177e4 LT |
1402 | case XFS_SBS_RBMBLOCKS: |
1403 | scounter = mp->m_sb.sb_rbmblocks; | |
1404 | scounter += delta; | |
1405 | if (scounter < 0) { | |
1406 | ASSERT(0); | |
014c2544 | 1407 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1408 | } |
1409 | mp->m_sb.sb_rbmblocks = scounter; | |
014c2544 | 1410 | return 0; |
1da177e4 LT |
1411 | case XFS_SBS_RBLOCKS: |
1412 | lcounter = (long long)mp->m_sb.sb_rblocks; | |
1413 | lcounter += delta; | |
1414 | if (lcounter < 0) { | |
1415 | ASSERT(0); | |
014c2544 | 1416 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1417 | } |
1418 | mp->m_sb.sb_rblocks = lcounter; | |
014c2544 | 1419 | return 0; |
1da177e4 LT |
1420 | case XFS_SBS_REXTENTS: |
1421 | lcounter = (long long)mp->m_sb.sb_rextents; | |
1422 | lcounter += delta; | |
1423 | if (lcounter < 0) { | |
1424 | ASSERT(0); | |
014c2544 | 1425 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1426 | } |
1427 | mp->m_sb.sb_rextents = lcounter; | |
014c2544 | 1428 | return 0; |
1da177e4 LT |
1429 | case XFS_SBS_REXTSLOG: |
1430 | scounter = mp->m_sb.sb_rextslog; | |
1431 | scounter += delta; | |
1432 | if (scounter < 0) { | |
1433 | ASSERT(0); | |
014c2544 | 1434 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1435 | } |
1436 | mp->m_sb.sb_rextslog = scounter; | |
014c2544 | 1437 | return 0; |
1da177e4 LT |
1438 | default: |
1439 | ASSERT(0); | |
014c2544 | 1440 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1441 | } |
1442 | } | |
1443 | ||
1444 | /* | |
1445 | * xfs_mod_incore_sb() is used to change a field in the in-core | |
1446 | * superblock structure by the specified delta. This modification | |
1447 | * is protected by the SB_LOCK. Just use the xfs_mod_incore_sb_unlocked() | |
1448 | * routine to do the work. | |
1449 | */ | |
1450 | int | |
1451 | xfs_mod_incore_sb(xfs_mount_t *mp, xfs_sb_field_t field, int delta, int rsvd) | |
1452 | { | |
1453 | unsigned long s; | |
1454 | int status; | |
1455 | ||
8d280b98 DC |
1456 | /* check for per-cpu counters */ |
1457 | switch (field) { | |
1458 | #ifdef HAVE_PERCPU_SB | |
1459 | case XFS_SBS_ICOUNT: | |
1460 | case XFS_SBS_IFREE: | |
1461 | case XFS_SBS_FDBLOCKS: | |
1462 | if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) { | |
1463 | status = xfs_icsb_modify_counters(mp, field, | |
1464 | delta, rsvd); | |
1465 | break; | |
1466 | } | |
1467 | /* FALLTHROUGH */ | |
1468 | #endif | |
1469 | default: | |
1470 | s = XFS_SB_LOCK(mp); | |
1471 | status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); | |
1472 | XFS_SB_UNLOCK(mp, s); | |
1473 | break; | |
1474 | } | |
1475 | ||
014c2544 | 1476 | return status; |
1da177e4 LT |
1477 | } |
1478 | ||
1479 | /* | |
1480 | * xfs_mod_incore_sb_batch() is used to change more than one field | |
1481 | * in the in-core superblock structure at a time. This modification | |
1482 | * is protected by a lock internal to this module. The fields and | |
1483 | * changes to those fields are specified in the array of xfs_mod_sb | |
1484 | * structures passed in. | |
1485 | * | |
1486 | * Either all of the specified deltas will be applied or none of | |
1487 | * them will. If any modified field dips below 0, then all modifications | |
1488 | * will be backed out and EINVAL will be returned. | |
1489 | */ | |
1490 | int | |
1491 | xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd) | |
1492 | { | |
1493 | unsigned long s; | |
1494 | int status=0; | |
1495 | xfs_mod_sb_t *msbp; | |
1496 | ||
1497 | /* | |
1498 | * Loop through the array of mod structures and apply each | |
1499 | * individually. If any fail, then back out all those | |
1500 | * which have already been applied. Do all of this within | |
1501 | * the scope of the SB_LOCK so that all of the changes will | |
1502 | * be atomic. | |
1503 | */ | |
1504 | s = XFS_SB_LOCK(mp); | |
1505 | msbp = &msb[0]; | |
1506 | for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) { | |
1507 | /* | |
1508 | * Apply the delta at index n. If it fails, break | |
1509 | * from the loop so we'll fall into the undo loop | |
1510 | * below. | |
1511 | */ | |
8d280b98 DC |
1512 | switch (msbp->msb_field) { |
1513 | #ifdef HAVE_PERCPU_SB | |
1514 | case XFS_SBS_ICOUNT: | |
1515 | case XFS_SBS_IFREE: | |
1516 | case XFS_SBS_FDBLOCKS: | |
1517 | if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) { | |
1518 | status = xfs_icsb_modify_counters_locked(mp, | |
1519 | msbp->msb_field, | |
1520 | msbp->msb_delta, rsvd); | |
1521 | break; | |
1522 | } | |
1523 | /* FALLTHROUGH */ | |
1524 | #endif | |
1525 | default: | |
1526 | status = xfs_mod_incore_sb_unlocked(mp, | |
1527 | msbp->msb_field, | |
1528 | msbp->msb_delta, rsvd); | |
1529 | break; | |
1530 | } | |
1531 | ||
1da177e4 LT |
1532 | if (status != 0) { |
1533 | break; | |
1534 | } | |
1535 | } | |
1536 | ||
1537 | /* | |
1538 | * If we didn't complete the loop above, then back out | |
1539 | * any changes made to the superblock. If you add code | |
1540 | * between the loop above and here, make sure that you | |
1541 | * preserve the value of status. Loop back until | |
1542 | * we step below the beginning of the array. Make sure | |
1543 | * we don't touch anything back there. | |
1544 | */ | |
1545 | if (status != 0) { | |
1546 | msbp--; | |
1547 | while (msbp >= msb) { | |
8d280b98 DC |
1548 | switch (msbp->msb_field) { |
1549 | #ifdef HAVE_PERCPU_SB | |
1550 | case XFS_SBS_ICOUNT: | |
1551 | case XFS_SBS_IFREE: | |
1552 | case XFS_SBS_FDBLOCKS: | |
1553 | if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) { | |
1554 | status = | |
1555 | xfs_icsb_modify_counters_locked(mp, | |
1556 | msbp->msb_field, | |
1557 | -(msbp->msb_delta), | |
1558 | rsvd); | |
1559 | break; | |
1560 | } | |
1561 | /* FALLTHROUGH */ | |
1562 | #endif | |
1563 | default: | |
1564 | status = xfs_mod_incore_sb_unlocked(mp, | |
1565 | msbp->msb_field, | |
1566 | -(msbp->msb_delta), | |
1567 | rsvd); | |
1568 | break; | |
1569 | } | |
1da177e4 LT |
1570 | ASSERT(status == 0); |
1571 | msbp--; | |
1572 | } | |
1573 | } | |
1574 | XFS_SB_UNLOCK(mp, s); | |
014c2544 | 1575 | return status; |
1da177e4 LT |
1576 | } |
1577 | ||
1578 | /* | |
1579 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
1580 | * The buffer is returned locked and read in from disk. | |
1581 | * The buffer should be released with a call to xfs_brelse(). | |
1582 | * | |
1583 | * If the flags parameter is BUF_TRYLOCK, then we'll only return | |
1584 | * the superblock buffer if it can be locked without sleeping. | |
1585 | * If it can't then we'll return NULL. | |
1586 | */ | |
1587 | xfs_buf_t * | |
1588 | xfs_getsb( | |
1589 | xfs_mount_t *mp, | |
1590 | int flags) | |
1591 | { | |
1592 | xfs_buf_t *bp; | |
1593 | ||
1594 | ASSERT(mp->m_sb_bp != NULL); | |
1595 | bp = mp->m_sb_bp; | |
1596 | if (flags & XFS_BUF_TRYLOCK) { | |
1597 | if (!XFS_BUF_CPSEMA(bp)) { | |
1598 | return NULL; | |
1599 | } | |
1600 | } else { | |
1601 | XFS_BUF_PSEMA(bp, PRIBIO); | |
1602 | } | |
1603 | XFS_BUF_HOLD(bp); | |
1604 | ASSERT(XFS_BUF_ISDONE(bp)); | |
014c2544 | 1605 | return bp; |
1da177e4 LT |
1606 | } |
1607 | ||
1608 | /* | |
1609 | * Used to free the superblock along various error paths. | |
1610 | */ | |
1611 | void | |
1612 | xfs_freesb( | |
1613 | xfs_mount_t *mp) | |
1614 | { | |
1615 | xfs_buf_t *bp; | |
1616 | ||
1617 | /* | |
1618 | * Use xfs_getsb() so that the buffer will be locked | |
1619 | * when we call xfs_buf_relse(). | |
1620 | */ | |
1621 | bp = xfs_getsb(mp, 0); | |
1622 | XFS_BUF_UNMANAGE(bp); | |
1623 | xfs_buf_relse(bp); | |
1624 | mp->m_sb_bp = NULL; | |
1625 | } | |
1626 | ||
1627 | /* | |
1628 | * See if the UUID is unique among mounted XFS filesystems. | |
1629 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. | |
1630 | */ | |
1631 | STATIC int | |
1632 | xfs_uuid_mount( | |
1633 | xfs_mount_t *mp) | |
1634 | { | |
1635 | if (uuid_is_nil(&mp->m_sb.sb_uuid)) { | |
1636 | cmn_err(CE_WARN, | |
1637 | "XFS: Filesystem %s has nil UUID - can't mount", | |
1638 | mp->m_fsname); | |
1639 | return -1; | |
1640 | } | |
1641 | if (!uuid_table_insert(&mp->m_sb.sb_uuid)) { | |
1642 | cmn_err(CE_WARN, | |
1643 | "XFS: Filesystem %s has duplicate UUID - can't mount", | |
1644 | mp->m_fsname); | |
1645 | return -1; | |
1646 | } | |
1647 | return 0; | |
1648 | } | |
1649 | ||
1650 | /* | |
1651 | * Remove filesystem from the UUID table. | |
1652 | */ | |
1653 | STATIC void | |
1654 | xfs_uuid_unmount( | |
1655 | xfs_mount_t *mp) | |
1656 | { | |
1657 | uuid_table_remove(&mp->m_sb.sb_uuid); | |
1658 | } | |
1659 | ||
1660 | /* | |
1661 | * Used to log changes to the superblock unit and width fields which could | |
1662 | * be altered by the mount options. Only the first superblock is updated. | |
1663 | */ | |
1664 | STATIC void | |
1665 | xfs_mount_log_sbunit( | |
1666 | xfs_mount_t *mp, | |
1667 | __int64_t fields) | |
1668 | { | |
1669 | xfs_trans_t *tp; | |
1670 | ||
1671 | ASSERT(fields & (XFS_SB_UNIT|XFS_SB_WIDTH|XFS_SB_UUID)); | |
1672 | ||
1673 | tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT); | |
1674 | if (xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0, | |
1675 | XFS_DEFAULT_LOG_COUNT)) { | |
1676 | xfs_trans_cancel(tp, 0); | |
1677 | return; | |
1678 | } | |
1679 | xfs_mod_sb(tp, fields); | |
1680 | xfs_trans_commit(tp, 0, NULL); | |
1681 | } | |
8d280b98 DC |
1682 | |
1683 | ||
1684 | #ifdef HAVE_PERCPU_SB | |
1685 | /* | |
1686 | * Per-cpu incore superblock counters | |
1687 | * | |
1688 | * Simple concept, difficult implementation | |
1689 | * | |
1690 | * Basically, replace the incore superblock counters with a distributed per cpu | |
1691 | * counter for contended fields (e.g. free block count). | |
1692 | * | |
1693 | * Difficulties arise in that the incore sb is used for ENOSPC checking, and | |
1694 | * hence needs to be accurately read when we are running low on space. Hence | |
1695 | * there is a method to enable and disable the per-cpu counters based on how | |
1696 | * much "stuff" is available in them. | |
1697 | * | |
1698 | * Basically, a counter is enabled if there is enough free resource to justify | |
1699 | * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local | |
1700 | * ENOSPC), then we disable the counters to synchronise all callers and | |
1701 | * re-distribute the available resources. | |
1702 | * | |
1703 | * If, once we redistributed the available resources, we still get a failure, | |
1704 | * we disable the per-cpu counter and go through the slow path. | |
1705 | * | |
1706 | * The slow path is the current xfs_mod_incore_sb() function. This means that | |
1707 | * when we disable a per-cpu counter, we need to drain it's resources back to | |
1708 | * the global superblock. We do this after disabling the counter to prevent | |
1709 | * more threads from queueing up on the counter. | |
1710 | * | |
1711 | * Essentially, this means that we still need a lock in the fast path to enable | |
1712 | * synchronisation between the global counters and the per-cpu counters. This | |
1713 | * is not a problem because the lock will be local to a CPU almost all the time | |
1714 | * and have little contention except when we get to ENOSPC conditions. | |
1715 | * | |
1716 | * Basically, this lock becomes a barrier that enables us to lock out the fast | |
1717 | * path while we do things like enabling and disabling counters and | |
1718 | * synchronising the counters. | |
1719 | * | |
1720 | * Locking rules: | |
1721 | * | |
1722 | * 1. XFS_SB_LOCK() before picking up per-cpu locks | |
1723 | * 2. per-cpu locks always picked up via for_each_online_cpu() order | |
1724 | * 3. accurate counter sync requires XFS_SB_LOCK + per cpu locks | |
1725 | * 4. modifying per-cpu counters requires holding per-cpu lock | |
1726 | * 5. modifying global counters requires holding XFS_SB_LOCK | |
1727 | * 6. enabling or disabling a counter requires holding the XFS_SB_LOCK | |
1728 | * and _none_ of the per-cpu locks. | |
1729 | * | |
1730 | * Disabled counters are only ever re-enabled by a balance operation | |
1731 | * that results in more free resources per CPU than a given threshold. | |
1732 | * To ensure counters don't remain disabled, they are rebalanced when | |
1733 | * the global resource goes above a higher threshold (i.e. some hysteresis | |
1734 | * is present to prevent thrashing). | |
e8234a68 DC |
1735 | */ |
1736 | ||
1737 | /* | |
1738 | * hot-plug CPU notifier support. | |
8d280b98 | 1739 | * |
e8234a68 DC |
1740 | * We cannot use the hotcpu_register() function because it does |
1741 | * not allow notifier instances. We need a notifier per filesystem | |
1742 | * as we need to be able to identify the filesystem to balance | |
c41564b5 | 1743 | * the counters out. This is achieved by having a notifier block |
e8234a68 DC |
1744 | * embedded in the xfs_mount_t and doing pointer magic to get the |
1745 | * mount pointer from the notifier block address. | |
8d280b98 | 1746 | */ |
e8234a68 DC |
1747 | STATIC int |
1748 | xfs_icsb_cpu_notify( | |
1749 | struct notifier_block *nfb, | |
1750 | unsigned long action, | |
1751 | void *hcpu) | |
1752 | { | |
1753 | xfs_icsb_cnts_t *cntp; | |
1754 | xfs_mount_t *mp; | |
1755 | int s; | |
1756 | ||
1757 | mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier); | |
1758 | cntp = (xfs_icsb_cnts_t *) | |
1759 | per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu); | |
1760 | switch (action) { | |
1761 | case CPU_UP_PREPARE: | |
1762 | /* Easy Case - initialize the area and locks, and | |
1763 | * then rebalance when online does everything else for us. */ | |
01e1b69c | 1764 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 DC |
1765 | break; |
1766 | case CPU_ONLINE: | |
1767 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); | |
1768 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
1769 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
1770 | break; | |
1771 | case CPU_DEAD: | |
1772 | /* Disable all the counters, then fold the dead cpu's | |
1773 | * count into the total on the global superblock and | |
1774 | * re-enable the counters. */ | |
1775 | s = XFS_SB_LOCK(mp); | |
1776 | xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT); | |
1777 | xfs_icsb_disable_counter(mp, XFS_SBS_IFREE); | |
1778 | xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS); | |
1779 | ||
1780 | mp->m_sb.sb_icount += cntp->icsb_icount; | |
1781 | mp->m_sb.sb_ifree += cntp->icsb_ifree; | |
1782 | mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks; | |
1783 | ||
01e1b69c | 1784 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 DC |
1785 | |
1786 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, XFS_ICSB_SB_LOCKED); | |
1787 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, XFS_ICSB_SB_LOCKED); | |
1788 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, XFS_ICSB_SB_LOCKED); | |
1789 | XFS_SB_UNLOCK(mp, s); | |
1790 | break; | |
1791 | } | |
1792 | ||
1793 | return NOTIFY_OK; | |
1794 | } | |
1795 | ||
8d280b98 DC |
1796 | int |
1797 | xfs_icsb_init_counters( | |
1798 | xfs_mount_t *mp) | |
1799 | { | |
1800 | xfs_icsb_cnts_t *cntp; | |
1801 | int i; | |
1802 | ||
1803 | mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t); | |
1804 | if (mp->m_sb_cnts == NULL) | |
1805 | return -ENOMEM; | |
1806 | ||
e8234a68 DC |
1807 | mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify; |
1808 | mp->m_icsb_notifier.priority = 0; | |
1809 | register_cpu_notifier(&mp->m_icsb_notifier); | |
1810 | ||
8d280b98 DC |
1811 | for_each_online_cpu(i) { |
1812 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 1813 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
8d280b98 DC |
1814 | } |
1815 | /* | |
1816 | * start with all counters disabled so that the | |
1817 | * initial balance kicks us off correctly | |
1818 | */ | |
1819 | mp->m_icsb_counters = -1; | |
1820 | return 0; | |
1821 | } | |
1822 | ||
1823 | STATIC void | |
1824 | xfs_icsb_destroy_counters( | |
1825 | xfs_mount_t *mp) | |
1826 | { | |
e8234a68 DC |
1827 | if (mp->m_sb_cnts) { |
1828 | unregister_cpu_notifier(&mp->m_icsb_notifier); | |
8d280b98 | 1829 | free_percpu(mp->m_sb_cnts); |
e8234a68 | 1830 | } |
8d280b98 DC |
1831 | } |
1832 | ||
01e1b69c DC |
1833 | STATIC inline void |
1834 | xfs_icsb_lock_cntr( | |
1835 | xfs_icsb_cnts_t *icsbp) | |
1836 | { | |
1837 | while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) { | |
1838 | ndelay(1000); | |
1839 | } | |
1840 | } | |
1841 | ||
1842 | STATIC inline void | |
1843 | xfs_icsb_unlock_cntr( | |
1844 | xfs_icsb_cnts_t *icsbp) | |
1845 | { | |
1846 | clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags); | |
1847 | } | |
1848 | ||
8d280b98 DC |
1849 | |
1850 | STATIC inline void | |
1851 | xfs_icsb_lock_all_counters( | |
1852 | xfs_mount_t *mp) | |
1853 | { | |
1854 | xfs_icsb_cnts_t *cntp; | |
1855 | int i; | |
1856 | ||
1857 | for_each_online_cpu(i) { | |
1858 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 1859 | xfs_icsb_lock_cntr(cntp); |
8d280b98 DC |
1860 | } |
1861 | } | |
1862 | ||
1863 | STATIC inline void | |
1864 | xfs_icsb_unlock_all_counters( | |
1865 | xfs_mount_t *mp) | |
1866 | { | |
1867 | xfs_icsb_cnts_t *cntp; | |
1868 | int i; | |
1869 | ||
1870 | for_each_online_cpu(i) { | |
1871 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 1872 | xfs_icsb_unlock_cntr(cntp); |
8d280b98 DC |
1873 | } |
1874 | } | |
1875 | ||
1876 | STATIC void | |
1877 | xfs_icsb_count( | |
1878 | xfs_mount_t *mp, | |
1879 | xfs_icsb_cnts_t *cnt, | |
1880 | int flags) | |
1881 | { | |
1882 | xfs_icsb_cnts_t *cntp; | |
1883 | int i; | |
1884 | ||
1885 | memset(cnt, 0, sizeof(xfs_icsb_cnts_t)); | |
1886 | ||
1887 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
1888 | xfs_icsb_lock_all_counters(mp); | |
1889 | ||
1890 | for_each_online_cpu(i) { | |
1891 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
1892 | cnt->icsb_icount += cntp->icsb_icount; | |
1893 | cnt->icsb_ifree += cntp->icsb_ifree; | |
1894 | cnt->icsb_fdblocks += cntp->icsb_fdblocks; | |
1895 | } | |
1896 | ||
1897 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
1898 | xfs_icsb_unlock_all_counters(mp); | |
1899 | } | |
1900 | ||
1901 | STATIC int | |
1902 | xfs_icsb_counter_disabled( | |
1903 | xfs_mount_t *mp, | |
1904 | xfs_sb_field_t field) | |
1905 | { | |
1906 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
1907 | return test_bit(field, &mp->m_icsb_counters); | |
1908 | } | |
1909 | ||
1910 | STATIC int | |
1911 | xfs_icsb_disable_counter( | |
1912 | xfs_mount_t *mp, | |
1913 | xfs_sb_field_t field) | |
1914 | { | |
1915 | xfs_icsb_cnts_t cnt; | |
1916 | ||
1917 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
1918 | ||
1919 | xfs_icsb_lock_all_counters(mp); | |
1920 | if (!test_and_set_bit(field, &mp->m_icsb_counters)) { | |
1921 | /* drain back to superblock */ | |
1922 | ||
1923 | xfs_icsb_count(mp, &cnt, XFS_ICSB_SB_LOCKED|XFS_ICSB_LAZY_COUNT); | |
1924 | switch(field) { | |
1925 | case XFS_SBS_ICOUNT: | |
1926 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
1927 | break; | |
1928 | case XFS_SBS_IFREE: | |
1929 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
1930 | break; | |
1931 | case XFS_SBS_FDBLOCKS: | |
1932 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
1933 | break; | |
1934 | default: | |
1935 | BUG(); | |
1936 | } | |
1937 | } | |
1938 | ||
1939 | xfs_icsb_unlock_all_counters(mp); | |
1940 | ||
1941 | return 0; | |
1942 | } | |
1943 | ||
1944 | STATIC void | |
1945 | xfs_icsb_enable_counter( | |
1946 | xfs_mount_t *mp, | |
1947 | xfs_sb_field_t field, | |
1948 | uint64_t count, | |
1949 | uint64_t resid) | |
1950 | { | |
1951 | xfs_icsb_cnts_t *cntp; | |
1952 | int i; | |
1953 | ||
1954 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
1955 | ||
1956 | xfs_icsb_lock_all_counters(mp); | |
1957 | for_each_online_cpu(i) { | |
1958 | cntp = per_cpu_ptr(mp->m_sb_cnts, i); | |
1959 | switch (field) { | |
1960 | case XFS_SBS_ICOUNT: | |
1961 | cntp->icsb_icount = count + resid; | |
1962 | break; | |
1963 | case XFS_SBS_IFREE: | |
1964 | cntp->icsb_ifree = count + resid; | |
1965 | break; | |
1966 | case XFS_SBS_FDBLOCKS: | |
1967 | cntp->icsb_fdblocks = count + resid; | |
1968 | break; | |
1969 | default: | |
1970 | BUG(); | |
1971 | break; | |
1972 | } | |
1973 | resid = 0; | |
1974 | } | |
1975 | clear_bit(field, &mp->m_icsb_counters); | |
1976 | xfs_icsb_unlock_all_counters(mp); | |
1977 | } | |
1978 | ||
1979 | STATIC void | |
1980 | xfs_icsb_sync_counters_int( | |
1981 | xfs_mount_t *mp, | |
1982 | int flags) | |
1983 | { | |
1984 | xfs_icsb_cnts_t cnt; | |
1985 | int s; | |
1986 | ||
1987 | /* Pass 1: lock all counters */ | |
1988 | if ((flags & XFS_ICSB_SB_LOCKED) == 0) | |
1989 | s = XFS_SB_LOCK(mp); | |
1990 | ||
1991 | xfs_icsb_count(mp, &cnt, flags); | |
1992 | ||
1993 | /* Step 3: update mp->m_sb fields */ | |
1994 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT)) | |
1995 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
1996 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE)) | |
1997 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
1998 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS)) | |
1999 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
2000 | ||
2001 | if ((flags & XFS_ICSB_SB_LOCKED) == 0) | |
2002 | XFS_SB_UNLOCK(mp, s); | |
2003 | } | |
2004 | ||
2005 | /* | |
2006 | * Accurate update of per-cpu counters to incore superblock | |
2007 | */ | |
2008 | STATIC void | |
2009 | xfs_icsb_sync_counters( | |
2010 | xfs_mount_t *mp) | |
2011 | { | |
2012 | xfs_icsb_sync_counters_int(mp, 0); | |
2013 | } | |
2014 | ||
2015 | /* | |
2016 | * lazy addition used for things like df, background sb syncs, etc | |
2017 | */ | |
2018 | void | |
2019 | xfs_icsb_sync_counters_lazy( | |
2020 | xfs_mount_t *mp) | |
2021 | { | |
2022 | xfs_icsb_sync_counters_int(mp, XFS_ICSB_LAZY_COUNT); | |
2023 | } | |
2024 | ||
2025 | /* | |
2026 | * Balance and enable/disable counters as necessary. | |
2027 | * | |
2028 | * Thresholds for re-enabling counters are somewhat magic. | |
2029 | * inode counts are chosen to be the same number as single | |
2030 | * on disk allocation chunk per CPU, and free blocks is | |
2031 | * something far enough zero that we aren't going thrash | |
2032 | * when we get near ENOSPC. | |
2033 | */ | |
2034 | #define XFS_ICSB_INO_CNTR_REENABLE 64 | |
2035 | #define XFS_ICSB_FDBLK_CNTR_REENABLE 512 | |
2036 | STATIC void | |
2037 | xfs_icsb_balance_counter( | |
2038 | xfs_mount_t *mp, | |
2039 | xfs_sb_field_t field, | |
2040 | int flags) | |
2041 | { | |
2042 | uint64_t count, resid = 0; | |
2043 | int weight = num_online_cpus(); | |
2044 | int s; | |
2045 | ||
2046 | if (!(flags & XFS_ICSB_SB_LOCKED)) | |
2047 | s = XFS_SB_LOCK(mp); | |
2048 | ||
2049 | /* disable counter and sync counter */ | |
2050 | xfs_icsb_disable_counter(mp, field); | |
2051 | ||
2052 | /* update counters - first CPU gets residual*/ | |
2053 | switch (field) { | |
2054 | case XFS_SBS_ICOUNT: | |
2055 | count = mp->m_sb.sb_icount; | |
2056 | resid = do_div(count, weight); | |
2057 | if (count < XFS_ICSB_INO_CNTR_REENABLE) | |
2058 | goto out; | |
2059 | break; | |
2060 | case XFS_SBS_IFREE: | |
2061 | count = mp->m_sb.sb_ifree; | |
2062 | resid = do_div(count, weight); | |
2063 | if (count < XFS_ICSB_INO_CNTR_REENABLE) | |
2064 | goto out; | |
2065 | break; | |
2066 | case XFS_SBS_FDBLOCKS: | |
2067 | count = mp->m_sb.sb_fdblocks; | |
2068 | resid = do_div(count, weight); | |
2069 | if (count < XFS_ICSB_FDBLK_CNTR_REENABLE) | |
2070 | goto out; | |
2071 | break; | |
2072 | default: | |
2073 | BUG(); | |
2074 | break; | |
2075 | } | |
2076 | ||
2077 | xfs_icsb_enable_counter(mp, field, count, resid); | |
2078 | out: | |
2079 | if (!(flags & XFS_ICSB_SB_LOCKED)) | |
2080 | XFS_SB_UNLOCK(mp, s); | |
2081 | } | |
2082 | ||
2083 | STATIC int | |
2084 | xfs_icsb_modify_counters_int( | |
2085 | xfs_mount_t *mp, | |
2086 | xfs_sb_field_t field, | |
2087 | int delta, | |
2088 | int rsvd, | |
2089 | int flags) | |
2090 | { | |
2091 | xfs_icsb_cnts_t *icsbp; | |
2092 | long long lcounter; /* long counter for 64 bit fields */ | |
2093 | int cpu, s, locked = 0; | |
2094 | int ret = 0, balance_done = 0; | |
2095 | ||
2096 | again: | |
2097 | cpu = get_cpu(); | |
2098 | icsbp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, cpu), | |
01e1b69c | 2099 | xfs_icsb_lock_cntr(icsbp); |
8d280b98 DC |
2100 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) |
2101 | goto slow_path; | |
2102 | ||
2103 | switch (field) { | |
2104 | case XFS_SBS_ICOUNT: | |
2105 | lcounter = icsbp->icsb_icount; | |
2106 | lcounter += delta; | |
2107 | if (unlikely(lcounter < 0)) | |
2108 | goto slow_path; | |
2109 | icsbp->icsb_icount = lcounter; | |
2110 | break; | |
2111 | ||
2112 | case XFS_SBS_IFREE: | |
2113 | lcounter = icsbp->icsb_ifree; | |
2114 | lcounter += delta; | |
2115 | if (unlikely(lcounter < 0)) | |
2116 | goto slow_path; | |
2117 | icsbp->icsb_ifree = lcounter; | |
2118 | break; | |
2119 | ||
2120 | case XFS_SBS_FDBLOCKS: | |
2121 | BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0); | |
2122 | ||
2123 | lcounter = icsbp->icsb_fdblocks; | |
2124 | lcounter += delta; | |
2125 | if (unlikely(lcounter < 0)) | |
2126 | goto slow_path; | |
2127 | icsbp->icsb_fdblocks = lcounter; | |
2128 | break; | |
2129 | default: | |
2130 | BUG(); | |
2131 | break; | |
2132 | } | |
01e1b69c | 2133 | xfs_icsb_unlock_cntr(icsbp); |
8d280b98 DC |
2134 | put_cpu(); |
2135 | if (locked) | |
2136 | XFS_SB_UNLOCK(mp, s); | |
2137 | return 0; | |
2138 | ||
2139 | /* | |
2140 | * The slow path needs to be run with the SBLOCK | |
2141 | * held so that we prevent other threads from | |
2142 | * attempting to run this path at the same time. | |
2143 | * this provides exclusion for the balancing code, | |
2144 | * and exclusive fallback if the balance does not | |
2145 | * provide enough resources to continue in an unlocked | |
2146 | * manner. | |
2147 | */ | |
2148 | slow_path: | |
01e1b69c | 2149 | xfs_icsb_unlock_cntr(icsbp); |
8d280b98 DC |
2150 | put_cpu(); |
2151 | ||
2152 | /* need to hold superblock incase we need | |
2153 | * to disable a counter */ | |
2154 | if (!(flags & XFS_ICSB_SB_LOCKED)) { | |
2155 | s = XFS_SB_LOCK(mp); | |
2156 | locked = 1; | |
2157 | flags |= XFS_ICSB_SB_LOCKED; | |
2158 | } | |
2159 | if (!balance_done) { | |
2160 | xfs_icsb_balance_counter(mp, field, flags); | |
2161 | balance_done = 1; | |
2162 | goto again; | |
2163 | } else { | |
2164 | /* | |
2165 | * we might not have enough on this local | |
2166 | * cpu to allocate for a bulk request. | |
2167 | * We need to drain this field from all CPUs | |
2168 | * and disable the counter fastpath | |
2169 | */ | |
2170 | xfs_icsb_disable_counter(mp, field); | |
2171 | } | |
2172 | ||
2173 | ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); | |
2174 | ||
2175 | if (locked) | |
2176 | XFS_SB_UNLOCK(mp, s); | |
2177 | return ret; | |
2178 | } | |
2179 | ||
2180 | STATIC int | |
2181 | xfs_icsb_modify_counters( | |
2182 | xfs_mount_t *mp, | |
2183 | xfs_sb_field_t field, | |
2184 | int delta, | |
2185 | int rsvd) | |
2186 | { | |
2187 | return xfs_icsb_modify_counters_int(mp, field, delta, rsvd, 0); | |
2188 | } | |
2189 | ||
2190 | /* | |
2191 | * Called when superblock is already locked | |
2192 | */ | |
2193 | STATIC int | |
2194 | xfs_icsb_modify_counters_locked( | |
2195 | xfs_mount_t *mp, | |
2196 | xfs_sb_field_t field, | |
2197 | int delta, | |
2198 | int rsvd) | |
2199 | { | |
2200 | return xfs_icsb_modify_counters_int(mp, field, delta, | |
2201 | rsvd, XFS_ICSB_SB_LOCKED); | |
2202 | } | |
2203 | #endif |