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 | |
1da177e4 | 48 | STATIC int xfs_uuid_mount(xfs_mount_t *); |
ba0f32d4 | 49 | STATIC void xfs_unmountfs_wait(xfs_mount_t *); |
1da177e4 | 50 | |
8d280b98 DC |
51 | |
52 | #ifdef HAVE_PERCPU_SB | |
20f4ebf2 | 53 | STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t, |
45af6c6d CH |
54 | int); |
55 | STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t, | |
56 | int); | |
8d280b98 | 57 | STATIC int xfs_icsb_modify_counters(xfs_mount_t *, xfs_sb_field_t, |
20f4ebf2 | 58 | int64_t, int); |
36fbe6e6 | 59 | STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t); |
8d280b98 DC |
60 | |
61 | #else | |
62 | ||
45af6c6d CH |
63 | #define xfs_icsb_balance_counter(mp, a, b) do { } while (0) |
64 | #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0) | |
8d280b98 | 65 | #define xfs_icsb_modify_counters(mp, a, b, c) do { } while (0) |
8d280b98 DC |
66 | |
67 | #endif | |
68 | ||
1df84c93 | 69 | static const struct { |
8d280b98 DC |
70 | short offset; |
71 | short type; /* 0 = integer | |
72 | * 1 = binary / string (no translation) | |
73 | */ | |
1da177e4 LT |
74 | } xfs_sb_info[] = { |
75 | { offsetof(xfs_sb_t, sb_magicnum), 0 }, | |
76 | { offsetof(xfs_sb_t, sb_blocksize), 0 }, | |
77 | { offsetof(xfs_sb_t, sb_dblocks), 0 }, | |
78 | { offsetof(xfs_sb_t, sb_rblocks), 0 }, | |
79 | { offsetof(xfs_sb_t, sb_rextents), 0 }, | |
80 | { offsetof(xfs_sb_t, sb_uuid), 1 }, | |
81 | { offsetof(xfs_sb_t, sb_logstart), 0 }, | |
82 | { offsetof(xfs_sb_t, sb_rootino), 0 }, | |
83 | { offsetof(xfs_sb_t, sb_rbmino), 0 }, | |
84 | { offsetof(xfs_sb_t, sb_rsumino), 0 }, | |
85 | { offsetof(xfs_sb_t, sb_rextsize), 0 }, | |
86 | { offsetof(xfs_sb_t, sb_agblocks), 0 }, | |
87 | { offsetof(xfs_sb_t, sb_agcount), 0 }, | |
88 | { offsetof(xfs_sb_t, sb_rbmblocks), 0 }, | |
89 | { offsetof(xfs_sb_t, sb_logblocks), 0 }, | |
90 | { offsetof(xfs_sb_t, sb_versionnum), 0 }, | |
91 | { offsetof(xfs_sb_t, sb_sectsize), 0 }, | |
92 | { offsetof(xfs_sb_t, sb_inodesize), 0 }, | |
93 | { offsetof(xfs_sb_t, sb_inopblock), 0 }, | |
94 | { offsetof(xfs_sb_t, sb_fname[0]), 1 }, | |
95 | { offsetof(xfs_sb_t, sb_blocklog), 0 }, | |
96 | { offsetof(xfs_sb_t, sb_sectlog), 0 }, | |
97 | { offsetof(xfs_sb_t, sb_inodelog), 0 }, | |
98 | { offsetof(xfs_sb_t, sb_inopblog), 0 }, | |
99 | { offsetof(xfs_sb_t, sb_agblklog), 0 }, | |
100 | { offsetof(xfs_sb_t, sb_rextslog), 0 }, | |
101 | { offsetof(xfs_sb_t, sb_inprogress), 0 }, | |
102 | { offsetof(xfs_sb_t, sb_imax_pct), 0 }, | |
103 | { offsetof(xfs_sb_t, sb_icount), 0 }, | |
104 | { offsetof(xfs_sb_t, sb_ifree), 0 }, | |
105 | { offsetof(xfs_sb_t, sb_fdblocks), 0 }, | |
106 | { offsetof(xfs_sb_t, sb_frextents), 0 }, | |
107 | { offsetof(xfs_sb_t, sb_uquotino), 0 }, | |
108 | { offsetof(xfs_sb_t, sb_gquotino), 0 }, | |
109 | { offsetof(xfs_sb_t, sb_qflags), 0 }, | |
110 | { offsetof(xfs_sb_t, sb_flags), 0 }, | |
111 | { offsetof(xfs_sb_t, sb_shared_vn), 0 }, | |
112 | { offsetof(xfs_sb_t, sb_inoalignmt), 0 }, | |
113 | { offsetof(xfs_sb_t, sb_unit), 0 }, | |
114 | { offsetof(xfs_sb_t, sb_width), 0 }, | |
115 | { offsetof(xfs_sb_t, sb_dirblklog), 0 }, | |
116 | { offsetof(xfs_sb_t, sb_logsectlog), 0 }, | |
117 | { offsetof(xfs_sb_t, sb_logsectsize),0 }, | |
118 | { offsetof(xfs_sb_t, sb_logsunit), 0 }, | |
119 | { offsetof(xfs_sb_t, sb_features2), 0 }, | |
ee1c0908 | 120 | { offsetof(xfs_sb_t, sb_bad_features2), 0 }, |
1da177e4 LT |
121 | { sizeof(xfs_sb_t), 0 } |
122 | }; | |
123 | ||
1da177e4 LT |
124 | /* |
125 | * Free up the resources associated with a mount structure. Assume that | |
126 | * the structure was initially zeroed, so we can tell which fields got | |
127 | * initialized. | |
128 | */ | |
c962fb79 | 129 | STATIC void |
ff4f038c | 130 | xfs_free_perag( |
745f6919 | 131 | xfs_mount_t *mp) |
1da177e4 | 132 | { |
1da177e4 LT |
133 | if (mp->m_perag) { |
134 | int agno; | |
135 | ||
136 | for (agno = 0; agno < mp->m_maxagi; agno++) | |
137 | if (mp->m_perag[agno].pagb_list) | |
f0e2d93c DV |
138 | kmem_free(mp->m_perag[agno].pagb_list); |
139 | kmem_free(mp->m_perag); | |
1da177e4 | 140 | } |
1da177e4 LT |
141 | } |
142 | ||
4cc929ee NS |
143 | /* |
144 | * Check size of device based on the (data/realtime) block count. | |
145 | * Note: this check is used by the growfs code as well as mount. | |
146 | */ | |
147 | int | |
148 | xfs_sb_validate_fsb_count( | |
149 | xfs_sb_t *sbp, | |
150 | __uint64_t nblocks) | |
151 | { | |
152 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
153 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
154 | ||
155 | #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */ | |
156 | if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) | |
157 | return E2BIG; | |
158 | #else /* Limited by UINT_MAX of sectors */ | |
159 | if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX) | |
160 | return E2BIG; | |
161 | #endif | |
162 | return 0; | |
163 | } | |
1da177e4 LT |
164 | |
165 | /* | |
166 | * Check the validity of the SB found. | |
167 | */ | |
168 | STATIC int | |
169 | xfs_mount_validate_sb( | |
170 | xfs_mount_t *mp, | |
764d1f89 NS |
171 | xfs_sb_t *sbp, |
172 | int flags) | |
1da177e4 LT |
173 | { |
174 | /* | |
175 | * If the log device and data device have the | |
176 | * same device number, the log is internal. | |
177 | * Consequently, the sb_logstart should be non-zero. If | |
178 | * we have a zero sb_logstart in this case, we may be trying to mount | |
179 | * a volume filesystem in a non-volume manner. | |
180 | */ | |
181 | if (sbp->sb_magicnum != XFS_SB_MAGIC) { | |
764d1f89 | 182 | xfs_fs_mount_cmn_err(flags, "bad magic number"); |
1da177e4 LT |
183 | return XFS_ERROR(EWRONGFS); |
184 | } | |
185 | ||
62118709 | 186 | if (!xfs_sb_good_version(sbp)) { |
764d1f89 | 187 | xfs_fs_mount_cmn_err(flags, "bad version"); |
1da177e4 LT |
188 | return XFS_ERROR(EWRONGFS); |
189 | } | |
190 | ||
191 | if (unlikely( | |
192 | sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) { | |
764d1f89 NS |
193 | xfs_fs_mount_cmn_err(flags, |
194 | "filesystem is marked as having an external log; " | |
195 | "specify logdev on the\nmount command line."); | |
196 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
197 | } |
198 | ||
199 | if (unlikely( | |
200 | sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) { | |
764d1f89 NS |
201 | xfs_fs_mount_cmn_err(flags, |
202 | "filesystem is marked as having an internal log; " | |
203 | "do not specify logdev on\nthe mount command line."); | |
204 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
205 | } |
206 | ||
207 | /* | |
208 | * More sanity checking. These were stolen directly from | |
209 | * xfs_repair. | |
210 | */ | |
211 | if (unlikely( | |
212 | sbp->sb_agcount <= 0 || | |
213 | sbp->sb_sectsize < XFS_MIN_SECTORSIZE || | |
214 | sbp->sb_sectsize > XFS_MAX_SECTORSIZE || | |
215 | sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG || | |
216 | sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG || | |
217 | sbp->sb_blocksize < XFS_MIN_BLOCKSIZE || | |
218 | sbp->sb_blocksize > XFS_MAX_BLOCKSIZE || | |
219 | sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG || | |
220 | sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG || | |
221 | sbp->sb_inodesize < XFS_DINODE_MIN_SIZE || | |
222 | sbp->sb_inodesize > XFS_DINODE_MAX_SIZE || | |
9f989c94 NS |
223 | sbp->sb_inodelog < XFS_DINODE_MIN_LOG || |
224 | sbp->sb_inodelog > XFS_DINODE_MAX_LOG || | |
225 | (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) || | |
1da177e4 LT |
226 | (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) || |
227 | (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) || | |
e50bd16f | 228 | (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) { |
764d1f89 | 229 | xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed"); |
1da177e4 LT |
230 | return XFS_ERROR(EFSCORRUPTED); |
231 | } | |
232 | ||
233 | /* | |
234 | * Sanity check AG count, size fields against data size field | |
235 | */ | |
236 | if (unlikely( | |
237 | sbp->sb_dblocks == 0 || | |
238 | sbp->sb_dblocks > | |
239 | (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks || | |
240 | sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) * | |
241 | sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) { | |
764d1f89 | 242 | xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed"); |
1da177e4 LT |
243 | return XFS_ERROR(EFSCORRUPTED); |
244 | } | |
245 | ||
2edbddd5 LM |
246 | /* |
247 | * Until this is fixed only page-sized or smaller data blocks work. | |
248 | */ | |
249 | if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) { | |
250 | xfs_fs_mount_cmn_err(flags, | |
251 | "file system with blocksize %d bytes", | |
252 | sbp->sb_blocksize); | |
253 | xfs_fs_mount_cmn_err(flags, | |
254 | "only pagesize (%ld) or less will currently work.", | |
255 | PAGE_SIZE); | |
256 | return XFS_ERROR(ENOSYS); | |
257 | } | |
258 | ||
4cc929ee NS |
259 | if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) || |
260 | xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) { | |
764d1f89 NS |
261 | xfs_fs_mount_cmn_err(flags, |
262 | "file system too large to be mounted on this system."); | |
1da177e4 LT |
263 | return XFS_ERROR(E2BIG); |
264 | } | |
265 | ||
266 | if (unlikely(sbp->sb_inprogress)) { | |
764d1f89 | 267 | xfs_fs_mount_cmn_err(flags, "file system busy"); |
1da177e4 LT |
268 | return XFS_ERROR(EFSCORRUPTED); |
269 | } | |
270 | ||
de20614b NS |
271 | /* |
272 | * Version 1 directory format has never worked on Linux. | |
273 | */ | |
62118709 | 274 | if (unlikely(!xfs_sb_version_hasdirv2(sbp))) { |
764d1f89 NS |
275 | xfs_fs_mount_cmn_err(flags, |
276 | "file system using version 1 directory format"); | |
de20614b NS |
277 | return XFS_ERROR(ENOSYS); |
278 | } | |
279 | ||
1da177e4 LT |
280 | return 0; |
281 | } | |
282 | ||
da353b0d DC |
283 | STATIC void |
284 | xfs_initialize_perag_icache( | |
285 | xfs_perag_t *pag) | |
286 | { | |
287 | if (!pag->pag_ici_init) { | |
288 | rwlock_init(&pag->pag_ici_lock); | |
289 | INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); | |
290 | pag->pag_ici_init = 1; | |
291 | } | |
292 | } | |
293 | ||
1da177e4 | 294 | xfs_agnumber_t |
c11e2c36 | 295 | xfs_initialize_perag( |
c11e2c36 NS |
296 | xfs_mount_t *mp, |
297 | xfs_agnumber_t agcount) | |
1da177e4 LT |
298 | { |
299 | xfs_agnumber_t index, max_metadata; | |
300 | xfs_perag_t *pag; | |
301 | xfs_agino_t agino; | |
302 | xfs_ino_t ino; | |
303 | xfs_sb_t *sbp = &mp->m_sb; | |
304 | xfs_ino_t max_inum = XFS_MAXINUMBER_32; | |
305 | ||
306 | /* Check to see if the filesystem can overflow 32 bit inodes */ | |
307 | agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0); | |
308 | ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino); | |
309 | ||
310 | /* Clear the mount flag if no inode can overflow 32 bits | |
311 | * on this filesystem, or if specifically requested.. | |
312 | */ | |
bd186aa9 | 313 | if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > max_inum) { |
1da177e4 LT |
314 | mp->m_flags |= XFS_MOUNT_32BITINODES; |
315 | } else { | |
316 | mp->m_flags &= ~XFS_MOUNT_32BITINODES; | |
317 | } | |
318 | ||
319 | /* If we can overflow then setup the ag headers accordingly */ | |
320 | if (mp->m_flags & XFS_MOUNT_32BITINODES) { | |
321 | /* Calculate how much should be reserved for inodes to | |
322 | * meet the max inode percentage. | |
323 | */ | |
324 | if (mp->m_maxicount) { | |
325 | __uint64_t icount; | |
326 | ||
327 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; | |
328 | do_div(icount, 100); | |
329 | icount += sbp->sb_agblocks - 1; | |
a749ee86 | 330 | do_div(icount, sbp->sb_agblocks); |
1da177e4 LT |
331 | max_metadata = icount; |
332 | } else { | |
333 | max_metadata = agcount; | |
334 | } | |
335 | for (index = 0; index < agcount; index++) { | |
336 | ino = XFS_AGINO_TO_INO(mp, index, agino); | |
337 | if (ino > max_inum) { | |
338 | index++; | |
339 | break; | |
340 | } | |
341 | ||
c41564b5 | 342 | /* This ag is preferred for inodes */ |
1da177e4 LT |
343 | pag = &mp->m_perag[index]; |
344 | pag->pagi_inodeok = 1; | |
345 | if (index < max_metadata) | |
346 | pag->pagf_metadata = 1; | |
da353b0d | 347 | xfs_initialize_perag_icache(pag); |
1da177e4 LT |
348 | } |
349 | } else { | |
350 | /* Setup default behavior for smaller filesystems */ | |
351 | for (index = 0; index < agcount; index++) { | |
352 | pag = &mp->m_perag[index]; | |
353 | pag->pagi_inodeok = 1; | |
da353b0d | 354 | xfs_initialize_perag_icache(pag); |
1da177e4 LT |
355 | } |
356 | } | |
357 | return index; | |
358 | } | |
359 | ||
2bdf7cd0 CH |
360 | void |
361 | xfs_sb_from_disk( | |
362 | xfs_sb_t *to, | |
363 | xfs_dsb_t *from) | |
364 | { | |
365 | to->sb_magicnum = be32_to_cpu(from->sb_magicnum); | |
366 | to->sb_blocksize = be32_to_cpu(from->sb_blocksize); | |
367 | to->sb_dblocks = be64_to_cpu(from->sb_dblocks); | |
368 | to->sb_rblocks = be64_to_cpu(from->sb_rblocks); | |
369 | to->sb_rextents = be64_to_cpu(from->sb_rextents); | |
370 | memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid)); | |
371 | to->sb_logstart = be64_to_cpu(from->sb_logstart); | |
372 | to->sb_rootino = be64_to_cpu(from->sb_rootino); | |
373 | to->sb_rbmino = be64_to_cpu(from->sb_rbmino); | |
374 | to->sb_rsumino = be64_to_cpu(from->sb_rsumino); | |
375 | to->sb_rextsize = be32_to_cpu(from->sb_rextsize); | |
376 | to->sb_agblocks = be32_to_cpu(from->sb_agblocks); | |
377 | to->sb_agcount = be32_to_cpu(from->sb_agcount); | |
378 | to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks); | |
379 | to->sb_logblocks = be32_to_cpu(from->sb_logblocks); | |
380 | to->sb_versionnum = be16_to_cpu(from->sb_versionnum); | |
381 | to->sb_sectsize = be16_to_cpu(from->sb_sectsize); | |
382 | to->sb_inodesize = be16_to_cpu(from->sb_inodesize); | |
383 | to->sb_inopblock = be16_to_cpu(from->sb_inopblock); | |
384 | memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname)); | |
385 | to->sb_blocklog = from->sb_blocklog; | |
386 | to->sb_sectlog = from->sb_sectlog; | |
387 | to->sb_inodelog = from->sb_inodelog; | |
388 | to->sb_inopblog = from->sb_inopblog; | |
389 | to->sb_agblklog = from->sb_agblklog; | |
390 | to->sb_rextslog = from->sb_rextslog; | |
391 | to->sb_inprogress = from->sb_inprogress; | |
392 | to->sb_imax_pct = from->sb_imax_pct; | |
393 | to->sb_icount = be64_to_cpu(from->sb_icount); | |
394 | to->sb_ifree = be64_to_cpu(from->sb_ifree); | |
395 | to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks); | |
396 | to->sb_frextents = be64_to_cpu(from->sb_frextents); | |
397 | to->sb_uquotino = be64_to_cpu(from->sb_uquotino); | |
398 | to->sb_gquotino = be64_to_cpu(from->sb_gquotino); | |
399 | to->sb_qflags = be16_to_cpu(from->sb_qflags); | |
400 | to->sb_flags = from->sb_flags; | |
401 | to->sb_shared_vn = from->sb_shared_vn; | |
402 | to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt); | |
403 | to->sb_unit = be32_to_cpu(from->sb_unit); | |
404 | to->sb_width = be32_to_cpu(from->sb_width); | |
405 | to->sb_dirblklog = from->sb_dirblklog; | |
406 | to->sb_logsectlog = from->sb_logsectlog; | |
407 | to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize); | |
408 | to->sb_logsunit = be32_to_cpu(from->sb_logsunit); | |
409 | to->sb_features2 = be32_to_cpu(from->sb_features2); | |
ee1c0908 | 410 | to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2); |
2bdf7cd0 CH |
411 | } |
412 | ||
1da177e4 | 413 | /* |
2bdf7cd0 | 414 | * Copy in core superblock to ondisk one. |
1da177e4 | 415 | * |
2bdf7cd0 | 416 | * The fields argument is mask of superblock fields to copy. |
1da177e4 LT |
417 | */ |
418 | void | |
2bdf7cd0 CH |
419 | xfs_sb_to_disk( |
420 | xfs_dsb_t *to, | |
421 | xfs_sb_t *from, | |
1da177e4 LT |
422 | __int64_t fields) |
423 | { | |
2bdf7cd0 CH |
424 | xfs_caddr_t to_ptr = (xfs_caddr_t)to; |
425 | xfs_caddr_t from_ptr = (xfs_caddr_t)from; | |
1da177e4 LT |
426 | xfs_sb_field_t f; |
427 | int first; | |
428 | int size; | |
429 | ||
1da177e4 | 430 | ASSERT(fields); |
1da177e4 LT |
431 | if (!fields) |
432 | return; | |
433 | ||
1da177e4 LT |
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) { | |
2bdf7cd0 | 442 | memcpy(to_ptr + first, from_ptr + first, size); |
1da177e4 LT |
443 | } else { |
444 | switch (size) { | |
445 | case 2: | |
2bdf7cd0 CH |
446 | *(__be16 *)(to_ptr + first) = |
447 | cpu_to_be16(*(__u16 *)(from_ptr + first)); | |
1da177e4 LT |
448 | break; |
449 | case 4: | |
2bdf7cd0 CH |
450 | *(__be32 *)(to_ptr + first) = |
451 | cpu_to_be32(*(__u32 *)(from_ptr + first)); | |
1da177e4 LT |
452 | break; |
453 | case 8: | |
2bdf7cd0 CH |
454 | *(__be64 *)(to_ptr + first) = |
455 | cpu_to_be64(*(__u64 *)(from_ptr + first)); | |
1da177e4 LT |
456 | break; |
457 | default: | |
458 | ASSERT(0); | |
459 | } | |
460 | } | |
461 | ||
462 | fields &= ~(1LL << f); | |
463 | } | |
464 | } | |
465 | ||
466 | /* | |
467 | * xfs_readsb | |
468 | * | |
469 | * Does the initial read of the superblock. | |
470 | */ | |
471 | int | |
764d1f89 | 472 | xfs_readsb(xfs_mount_t *mp, int flags) |
1da177e4 LT |
473 | { |
474 | unsigned int sector_size; | |
475 | unsigned int extra_flags; | |
476 | xfs_buf_t *bp; | |
1da177e4 LT |
477 | int error; |
478 | ||
479 | ASSERT(mp->m_sb_bp == NULL); | |
480 | ASSERT(mp->m_ddev_targp != NULL); | |
481 | ||
482 | /* | |
483 | * Allocate a (locked) buffer to hold the superblock. | |
484 | * This will be kept around at all times to optimize | |
485 | * access to the superblock. | |
486 | */ | |
487 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
488 | extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED; | |
489 | ||
490 | bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR, | |
491 | BTOBB(sector_size), extra_flags); | |
492 | if (!bp || XFS_BUF_ISERROR(bp)) { | |
764d1f89 | 493 | xfs_fs_mount_cmn_err(flags, "SB read failed"); |
1da177e4 LT |
494 | error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM; |
495 | goto fail; | |
496 | } | |
497 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
498 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); | |
499 | ||
500 | /* | |
501 | * Initialize the mount structure from the superblock. | |
502 | * But first do some basic consistency checking. | |
503 | */ | |
2bdf7cd0 | 504 | xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp)); |
1da177e4 | 505 | |
764d1f89 | 506 | error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags); |
1da177e4 | 507 | if (error) { |
764d1f89 | 508 | xfs_fs_mount_cmn_err(flags, "SB validate failed"); |
1da177e4 LT |
509 | goto fail; |
510 | } | |
511 | ||
512 | /* | |
513 | * We must be able to do sector-sized and sector-aligned IO. | |
514 | */ | |
515 | if (sector_size > mp->m_sb.sb_sectsize) { | |
764d1f89 NS |
516 | xfs_fs_mount_cmn_err(flags, |
517 | "device supports only %u byte sectors (not %u)", | |
1da177e4 LT |
518 | sector_size, mp->m_sb.sb_sectsize); |
519 | error = ENOSYS; | |
520 | goto fail; | |
521 | } | |
522 | ||
523 | /* | |
524 | * If device sector size is smaller than the superblock size, | |
525 | * re-read the superblock so the buffer is correctly sized. | |
526 | */ | |
527 | if (sector_size < mp->m_sb.sb_sectsize) { | |
528 | XFS_BUF_UNMANAGE(bp); | |
529 | xfs_buf_relse(bp); | |
530 | sector_size = mp->m_sb.sb_sectsize; | |
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 re-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 | ||
5478eead LM |
542 | /* Initialize per-cpu counters */ |
543 | xfs_icsb_reinit_counters(mp); | |
8d280b98 | 544 | |
1da177e4 LT |
545 | mp->m_sb_bp = bp; |
546 | xfs_buf_relse(bp); | |
547 | ASSERT(XFS_BUF_VALUSEMA(bp) > 0); | |
548 | return 0; | |
549 | ||
550 | fail: | |
551 | if (bp) { | |
552 | XFS_BUF_UNMANAGE(bp); | |
553 | xfs_buf_relse(bp); | |
554 | } | |
555 | return error; | |
556 | } | |
557 | ||
558 | ||
559 | /* | |
560 | * xfs_mount_common | |
561 | * | |
562 | * Mount initialization code establishing various mount | |
563 | * fields from the superblock associated with the given | |
564 | * mount structure | |
565 | */ | |
ba0f32d4 | 566 | STATIC void |
1da177e4 LT |
567 | xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp) |
568 | { | |
1da177e4 | 569 | mp->m_agfrotor = mp->m_agirotor = 0; |
007c61c6 | 570 | spin_lock_init(&mp->m_agirotor_lock); |
1da177e4 LT |
571 | mp->m_maxagi = mp->m_sb.sb_agcount; |
572 | mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG; | |
573 | mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT; | |
574 | mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT; | |
575 | mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1; | |
576 | mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog; | |
1da177e4 LT |
577 | mp->m_blockmask = sbp->sb_blocksize - 1; |
578 | mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG; | |
579 | mp->m_blockwmask = mp->m_blockwsize - 1; | |
1da177e4 LT |
580 | |
581 | /* | |
582 | * Setup for attributes, in case they get created. | |
583 | * This value is for inodes getting attributes for the first time, | |
584 | * the per-inode value is for old attribute values. | |
585 | */ | |
586 | ASSERT(sbp->sb_inodesize >= 256 && sbp->sb_inodesize <= 2048); | |
587 | switch (sbp->sb_inodesize) { | |
588 | case 256: | |
d8cc890d NS |
589 | mp->m_attroffset = XFS_LITINO(mp) - |
590 | XFS_BMDR_SPACE_CALC(MINABTPTRS); | |
1da177e4 LT |
591 | break; |
592 | case 512: | |
593 | case 1024: | |
594 | case 2048: | |
d8cc890d | 595 | mp->m_attroffset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS); |
1da177e4 LT |
596 | break; |
597 | default: | |
598 | ASSERT(0); | |
599 | } | |
600 | ASSERT(mp->m_attroffset < XFS_LITINO(mp)); | |
601 | ||
60197e8d CH |
602 | mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1); |
603 | mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0); | |
604 | mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2; | |
605 | mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2; | |
606 | ||
607 | mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1); | |
608 | mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0); | |
609 | mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2; | |
610 | mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2; | |
611 | ||
612 | mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1); | |
613 | mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0); | |
614 | mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2; | |
615 | mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2; | |
1da177e4 LT |
616 | |
617 | mp->m_bsize = XFS_FSB_TO_BB(mp, 1); | |
618 | mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK, | |
619 | sbp->sb_inopblock); | |
620 | mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog; | |
621 | } | |
92821e2b DC |
622 | |
623 | /* | |
624 | * xfs_initialize_perag_data | |
625 | * | |
626 | * Read in each per-ag structure so we can count up the number of | |
627 | * allocated inodes, free inodes and used filesystem blocks as this | |
628 | * information is no longer persistent in the superblock. Once we have | |
629 | * this information, write it into the in-core superblock structure. | |
630 | */ | |
631 | STATIC int | |
632 | xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount) | |
633 | { | |
634 | xfs_agnumber_t index; | |
635 | xfs_perag_t *pag; | |
636 | xfs_sb_t *sbp = &mp->m_sb; | |
637 | uint64_t ifree = 0; | |
638 | uint64_t ialloc = 0; | |
639 | uint64_t bfree = 0; | |
640 | uint64_t bfreelst = 0; | |
641 | uint64_t btree = 0; | |
642 | int error; | |
92821e2b DC |
643 | |
644 | for (index = 0; index < agcount; index++) { | |
645 | /* | |
646 | * read the agf, then the agi. This gets us | |
647 | * all the inforamtion we need and populates the | |
648 | * per-ag structures for us. | |
649 | */ | |
650 | error = xfs_alloc_pagf_init(mp, NULL, index, 0); | |
651 | if (error) | |
652 | return error; | |
653 | ||
654 | error = xfs_ialloc_pagi_init(mp, NULL, index); | |
655 | if (error) | |
656 | return error; | |
657 | pag = &mp->m_perag[index]; | |
658 | ifree += pag->pagi_freecount; | |
659 | ialloc += pag->pagi_count; | |
660 | bfree += pag->pagf_freeblks; | |
661 | bfreelst += pag->pagf_flcount; | |
662 | btree += pag->pagf_btreeblks; | |
663 | } | |
664 | /* | |
665 | * Overwrite incore superblock counters with just-read data | |
666 | */ | |
3685c2a1 | 667 | spin_lock(&mp->m_sb_lock); |
92821e2b DC |
668 | sbp->sb_ifree = ifree; |
669 | sbp->sb_icount = ialloc; | |
670 | sbp->sb_fdblocks = bfree + bfreelst + btree; | |
3685c2a1 | 671 | spin_unlock(&mp->m_sb_lock); |
92821e2b DC |
672 | |
673 | /* Fixup the per-cpu counters as well. */ | |
674 | xfs_icsb_reinit_counters(mp); | |
675 | ||
676 | return 0; | |
677 | } | |
678 | ||
1da177e4 | 679 | /* |
0771fb45 | 680 | * Update alignment values based on mount options and sb values |
1da177e4 | 681 | */ |
0771fb45 | 682 | STATIC int |
7884bc86 | 683 | xfs_update_alignment(xfs_mount_t *mp) |
1da177e4 | 684 | { |
1da177e4 | 685 | xfs_sb_t *sbp = &(mp->m_sb); |
1da177e4 | 686 | |
4249023a | 687 | if (mp->m_dalign) { |
1da177e4 LT |
688 | /* |
689 | * If stripe unit and stripe width are not multiples | |
690 | * of the fs blocksize turn off alignment. | |
691 | */ | |
692 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
693 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
694 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
695 | cmn_err(CE_WARN, | |
696 | "XFS: alignment check 1 failed"); | |
0771fb45 | 697 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
698 | } |
699 | mp->m_dalign = mp->m_swidth = 0; | |
700 | } else { | |
701 | /* | |
702 | * Convert the stripe unit and width to FSBs. | |
703 | */ | |
704 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
705 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
706 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
0771fb45 | 707 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
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); | |
0771fb45 | 724 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
725 | } |
726 | mp->m_swidth = 0; | |
727 | } | |
728 | } | |
729 | ||
730 | /* | |
731 | * Update superblock with new values | |
732 | * and log changes | |
733 | */ | |
62118709 | 734 | if (xfs_sb_version_hasdalign(sbp)) { |
1da177e4 LT |
735 | if (sbp->sb_unit != mp->m_dalign) { |
736 | sbp->sb_unit = mp->m_dalign; | |
7884bc86 | 737 | mp->m_update_flags |= XFS_SB_UNIT; |
1da177e4 LT |
738 | } |
739 | if (sbp->sb_width != mp->m_swidth) { | |
740 | sbp->sb_width = mp->m_swidth; | |
7884bc86 | 741 | mp->m_update_flags |= XFS_SB_WIDTH; |
1da177e4 LT |
742 | } |
743 | } | |
744 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
62118709 | 745 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
1da177e4 LT |
746 | mp->m_dalign = sbp->sb_unit; |
747 | mp->m_swidth = sbp->sb_width; | |
748 | } | |
749 | ||
0771fb45 ES |
750 | return 0; |
751 | } | |
1da177e4 | 752 | |
0771fb45 ES |
753 | /* |
754 | * Set the maximum inode count for this filesystem | |
755 | */ | |
756 | STATIC void | |
757 | xfs_set_maxicount(xfs_mount_t *mp) | |
758 | { | |
759 | xfs_sb_t *sbp = &(mp->m_sb); | |
760 | __uint64_t icount; | |
1da177e4 | 761 | |
0771fb45 ES |
762 | if (sbp->sb_imax_pct) { |
763 | /* | |
764 | * Make sure the maximum inode count is a multiple | |
765 | * of the units we allocate inodes in. | |
1da177e4 | 766 | */ |
1da177e4 LT |
767 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; |
768 | do_div(icount, 100); | |
769 | do_div(icount, mp->m_ialloc_blks); | |
770 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << | |
771 | sbp->sb_inopblog; | |
0771fb45 | 772 | } else { |
1da177e4 | 773 | mp->m_maxicount = 0; |
1da177e4 | 774 | } |
0771fb45 ES |
775 | } |
776 | ||
777 | /* | |
778 | * Set the default minimum read and write sizes unless | |
779 | * already specified in a mount option. | |
780 | * We use smaller I/O sizes when the file system | |
781 | * is being used for NFS service (wsync mount option). | |
782 | */ | |
783 | STATIC void | |
784 | xfs_set_rw_sizes(xfs_mount_t *mp) | |
785 | { | |
786 | xfs_sb_t *sbp = &(mp->m_sb); | |
787 | int readio_log, writeio_log; | |
1da177e4 | 788 | |
1da177e4 LT |
789 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
790 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
791 | readio_log = XFS_WSYNC_READIO_LOG; | |
792 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
793 | } else { | |
794 | readio_log = XFS_READIO_LOG_LARGE; | |
795 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
796 | } | |
797 | } else { | |
798 | readio_log = mp->m_readio_log; | |
799 | writeio_log = mp->m_writeio_log; | |
800 | } | |
801 | ||
1da177e4 LT |
802 | if (sbp->sb_blocklog > readio_log) { |
803 | mp->m_readio_log = sbp->sb_blocklog; | |
804 | } else { | |
805 | mp->m_readio_log = readio_log; | |
806 | } | |
807 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
808 | if (sbp->sb_blocklog > writeio_log) { | |
809 | mp->m_writeio_log = sbp->sb_blocklog; | |
810 | } else { | |
811 | mp->m_writeio_log = writeio_log; | |
812 | } | |
813 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
0771fb45 | 814 | } |
1da177e4 | 815 | |
0771fb45 ES |
816 | /* |
817 | * Set whether we're using inode alignment. | |
818 | */ | |
819 | STATIC void | |
820 | xfs_set_inoalignment(xfs_mount_t *mp) | |
821 | { | |
62118709 | 822 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
1da177e4 LT |
823 | mp->m_sb.sb_inoalignmt >= |
824 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
825 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; | |
826 | else | |
827 | mp->m_inoalign_mask = 0; | |
828 | /* | |
829 | * If we are using stripe alignment, check whether | |
830 | * the stripe unit is a multiple of the inode alignment | |
831 | */ | |
832 | if (mp->m_dalign && mp->m_inoalign_mask && | |
833 | !(mp->m_dalign & mp->m_inoalign_mask)) | |
834 | mp->m_sinoalign = mp->m_dalign; | |
835 | else | |
836 | mp->m_sinoalign = 0; | |
0771fb45 ES |
837 | } |
838 | ||
839 | /* | |
840 | * Check that the data (and log if separate) are an ok size. | |
841 | */ | |
842 | STATIC int | |
4249023a | 843 | xfs_check_sizes(xfs_mount_t *mp) |
0771fb45 ES |
844 | { |
845 | xfs_buf_t *bp; | |
846 | xfs_daddr_t d; | |
847 | int error; | |
848 | ||
1da177e4 LT |
849 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
850 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
851 | cmn_err(CE_WARN, "XFS: size check 1 failed"); | |
0771fb45 | 852 | return XFS_ERROR(E2BIG); |
1da177e4 LT |
853 | } |
854 | error = xfs_read_buf(mp, mp->m_ddev_targp, | |
855 | d - XFS_FSS_TO_BB(mp, 1), | |
856 | XFS_FSS_TO_BB(mp, 1), 0, &bp); | |
857 | if (!error) { | |
858 | xfs_buf_relse(bp); | |
859 | } else { | |
860 | cmn_err(CE_WARN, "XFS: size check 2 failed"); | |
0771fb45 | 861 | if (error == ENOSPC) |
1da177e4 | 862 | error = XFS_ERROR(E2BIG); |
0771fb45 | 863 | return error; |
1da177e4 LT |
864 | } |
865 | ||
4249023a | 866 | if (mp->m_logdev_targp != mp->m_ddev_targp) { |
1da177e4 LT |
867 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); |
868 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
869 | cmn_err(CE_WARN, "XFS: size check 3 failed"); | |
0771fb45 | 870 | return XFS_ERROR(E2BIG); |
1da177e4 LT |
871 | } |
872 | error = xfs_read_buf(mp, mp->m_logdev_targp, | |
873 | d - XFS_FSB_TO_BB(mp, 1), | |
874 | XFS_FSB_TO_BB(mp, 1), 0, &bp); | |
875 | if (!error) { | |
876 | xfs_buf_relse(bp); | |
877 | } else { | |
878 | cmn_err(CE_WARN, "XFS: size check 3 failed"); | |
0771fb45 | 879 | if (error == ENOSPC) |
1da177e4 | 880 | error = XFS_ERROR(E2BIG); |
0771fb45 ES |
881 | return error; |
882 | } | |
883 | } | |
884 | return 0; | |
885 | } | |
886 | ||
887 | /* | |
0771fb45 ES |
888 | * This function does the following on an initial mount of a file system: |
889 | * - reads the superblock from disk and init the mount struct | |
890 | * - if we're a 32-bit kernel, do a size check on the superblock | |
891 | * so we don't mount terabyte filesystems | |
892 | * - init mount struct realtime fields | |
893 | * - allocate inode hash table for fs | |
894 | * - init directory manager | |
895 | * - perform recovery and init the log manager | |
896 | */ | |
897 | int | |
898 | xfs_mountfs( | |
4249023a | 899 | xfs_mount_t *mp) |
0771fb45 ES |
900 | { |
901 | xfs_sb_t *sbp = &(mp->m_sb); | |
902 | xfs_inode_t *rip; | |
0771fb45 | 903 | __uint64_t resblks; |
0771fb45 | 904 | uint quotamount, quotaflags; |
0771fb45 ES |
905 | int error = 0; |
906 | ||
0771fb45 ES |
907 | xfs_mount_common(mp, sbp); |
908 | ||
ee1c0908 | 909 | /* |
e6957ea4 ES |
910 | * Check for a mismatched features2 values. Older kernels |
911 | * read & wrote into the wrong sb offset for sb_features2 | |
912 | * on some platforms due to xfs_sb_t not being 64bit size aligned | |
913 | * when sb_features2 was added, which made older superblock | |
914 | * reading/writing routines swap it as a 64-bit value. | |
ee1c0908 | 915 | * |
e6957ea4 ES |
916 | * For backwards compatibility, we make both slots equal. |
917 | * | |
918 | * If we detect a mismatched field, we OR the set bits into the | |
919 | * existing features2 field in case it has already been modified; we | |
920 | * don't want to lose any features. We then update the bad location | |
921 | * with the ORed value so that older kernels will see any features2 | |
922 | * flags, and mark the two fields as needing updates once the | |
923 | * transaction subsystem is online. | |
ee1c0908 | 924 | */ |
e6957ea4 | 925 | if (xfs_sb_has_mismatched_features2(sbp)) { |
ee1c0908 DC |
926 | cmn_err(CE_WARN, |
927 | "XFS: correcting sb_features alignment problem"); | |
928 | sbp->sb_features2 |= sbp->sb_bad_features2; | |
e6957ea4 | 929 | sbp->sb_bad_features2 = sbp->sb_features2; |
7884bc86 | 930 | mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2; |
e6957ea4 ES |
931 | |
932 | /* | |
933 | * Re-check for ATTR2 in case it was found in bad_features2 | |
934 | * slot. | |
935 | */ | |
7c12f296 TS |
936 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
937 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 938 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
939 | } |
940 | ||
941 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
942 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
943 | xfs_sb_version_removeattr2(&mp->m_sb); | |
7884bc86 | 944 | mp->m_update_flags |= XFS_SB_FEATURES2; |
e6957ea4 | 945 | |
7c12f296 TS |
946 | /* update sb_versionnum for the clearing of the morebits */ |
947 | if (!sbp->sb_features2) | |
7884bc86 | 948 | mp->m_update_flags |= XFS_SB_VERSIONNUM; |
ee1c0908 DC |
949 | } |
950 | ||
0771fb45 ES |
951 | /* |
952 | * Check if sb_agblocks is aligned at stripe boundary | |
953 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
954 | * allocator alignment is within an ag, therefore ag has | |
955 | * to be aligned at stripe boundary. | |
956 | */ | |
7884bc86 | 957 | error = xfs_update_alignment(mp); |
0771fb45 | 958 | if (error) |
f9057e3d | 959 | goto out; |
0771fb45 ES |
960 | |
961 | xfs_alloc_compute_maxlevels(mp); | |
962 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
963 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
964 | xfs_ialloc_compute_maxlevels(mp); | |
965 | ||
966 | xfs_set_maxicount(mp); | |
967 | ||
968 | mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog); | |
969 | ||
970 | /* | |
971 | * XFS uses the uuid from the superblock as the unique | |
972 | * identifier for fsid. We can not use the uuid from the volume | |
973 | * since a single partition filesystem is identical to a single | |
974 | * partition volume/filesystem. | |
975 | */ | |
f9057e3d | 976 | if (!(mp->m_flags & XFS_MOUNT_NOUUID)) { |
0771fb45 ES |
977 | if (xfs_uuid_mount(mp)) { |
978 | error = XFS_ERROR(EINVAL); | |
f9057e3d | 979 | goto out; |
1da177e4 LT |
980 | } |
981 | } | |
982 | ||
0771fb45 ES |
983 | /* |
984 | * Set the minimum read and write sizes | |
985 | */ | |
986 | xfs_set_rw_sizes(mp); | |
987 | ||
988 | /* | |
989 | * Set the inode cluster size. | |
990 | * This may still be overridden by the file system | |
991 | * block size if it is larger than the chosen cluster size. | |
992 | */ | |
993 | mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; | |
994 | ||
995 | /* | |
996 | * Set inode alignment fields | |
997 | */ | |
998 | xfs_set_inoalignment(mp); | |
999 | ||
1000 | /* | |
1001 | * Check that the data (and log if separate) are an ok size. | |
1002 | */ | |
4249023a | 1003 | error = xfs_check_sizes(mp); |
0771fb45 | 1004 | if (error) |
f9057e3d | 1005 | goto out_remove_uuid; |
0771fb45 | 1006 | |
1da177e4 LT |
1007 | /* |
1008 | * Initialize realtime fields in the mount structure | |
1009 | */ | |
0771fb45 ES |
1010 | error = xfs_rtmount_init(mp); |
1011 | if (error) { | |
1da177e4 | 1012 | cmn_err(CE_WARN, "XFS: RT mount failed"); |
f9057e3d | 1013 | goto out_remove_uuid; |
1da177e4 LT |
1014 | } |
1015 | ||
1da177e4 LT |
1016 | /* |
1017 | * Copies the low order bits of the timestamp and the randomly | |
1018 | * set "sequence" number out of a UUID. | |
1019 | */ | |
1020 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); | |
1021 | ||
1da177e4 LT |
1022 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ |
1023 | ||
f6c2d1fa | 1024 | xfs_dir_mount(mp); |
1da177e4 LT |
1025 | |
1026 | /* | |
1027 | * Initialize the attribute manager's entries. | |
1028 | */ | |
1029 | mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100; | |
1030 | ||
1031 | /* | |
1032 | * Initialize the precomputed transaction reservations values. | |
1033 | */ | |
1034 | xfs_trans_init(mp); | |
1035 | ||
1da177e4 LT |
1036 | /* |
1037 | * Allocate and initialize the per-ag data. | |
1038 | */ | |
1039 | init_rwsem(&mp->m_peraglock); | |
d62c251f CH |
1040 | mp->m_perag = kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t), |
1041 | KM_MAYFAIL); | |
1042 | if (!mp->m_perag) | |
f9057e3d | 1043 | goto out_remove_uuid; |
1da177e4 | 1044 | |
b267ce99 | 1045 | mp->m_maxagi = xfs_initialize_perag(mp, sbp->sb_agcount); |
1da177e4 | 1046 | |
f9057e3d CH |
1047 | if (!sbp->sb_logblocks) { |
1048 | cmn_err(CE_WARN, "XFS: no log defined"); | |
1049 | XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); | |
1050 | error = XFS_ERROR(EFSCORRUPTED); | |
1051 | goto out_free_perag; | |
1052 | } | |
1053 | ||
1da177e4 LT |
1054 | /* |
1055 | * log's mount-time initialization. Perform 1st part recovery if needed | |
1056 | */ | |
f9057e3d CH |
1057 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
1058 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
1059 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
1060 | if (error) { | |
1061 | cmn_err(CE_WARN, "XFS: log mount failed"); | |
1062 | goto out_free_perag; | |
1da177e4 LT |
1063 | } |
1064 | ||
92821e2b DC |
1065 | /* |
1066 | * Now the log is mounted, we know if it was an unclean shutdown or | |
1067 | * not. If it was, with the first phase of recovery has completed, we | |
1068 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
1069 | * but they are recovered transactionally in the second recovery phase | |
1070 | * later. | |
1071 | * | |
1072 | * Hence we can safely re-initialise incore superblock counters from | |
1073 | * the per-ag data. These may not be correct if the filesystem was not | |
1074 | * cleanly unmounted, so we need to wait for recovery to finish before | |
1075 | * doing this. | |
1076 | * | |
1077 | * If the filesystem was cleanly unmounted, then we can trust the | |
1078 | * values in the superblock to be correct and we don't need to do | |
1079 | * anything here. | |
1080 | * | |
1081 | * If we are currently making the filesystem, the initialisation will | |
1082 | * fail as the perag data is in an undefined state. | |
1083 | */ | |
92821e2b DC |
1084 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
1085 | !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
1086 | !mp->m_sb.sb_inprogress) { | |
1087 | error = xfs_initialize_perag_data(mp, sbp->sb_agcount); | |
f9057e3d CH |
1088 | if (error) |
1089 | goto out_free_perag; | |
92821e2b | 1090 | } |
f9057e3d | 1091 | |
1da177e4 LT |
1092 | /* |
1093 | * Get and sanity-check the root inode. | |
1094 | * Save the pointer to it in the mount structure. | |
1095 | */ | |
1096 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0); | |
1097 | if (error) { | |
1098 | cmn_err(CE_WARN, "XFS: failed to read root inode"); | |
f9057e3d | 1099 | goto out_log_dealloc; |
1da177e4 LT |
1100 | } |
1101 | ||
1102 | ASSERT(rip != NULL); | |
1da177e4 LT |
1103 | |
1104 | if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) { | |
1105 | cmn_err(CE_WARN, "XFS: corrupted root inode"); | |
b6574520 NS |
1106 | cmn_err(CE_WARN, "Device %s - root %llu is not a directory", |
1107 | XFS_BUFTARG_NAME(mp->m_ddev_targp), | |
1108 | (unsigned long long)rip->i_ino); | |
1da177e4 LT |
1109 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
1110 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
1111 | mp); | |
1112 | error = XFS_ERROR(EFSCORRUPTED); | |
f9057e3d | 1113 | goto out_rele_rip; |
1da177e4 LT |
1114 | } |
1115 | mp->m_rootip = rip; /* save it */ | |
1116 | ||
1117 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
1118 | ||
1119 | /* | |
1120 | * Initialize realtime inode pointers in the mount structure | |
1121 | */ | |
0771fb45 ES |
1122 | error = xfs_rtmount_inodes(mp); |
1123 | if (error) { | |
1da177e4 LT |
1124 | /* |
1125 | * Free up the root inode. | |
1126 | */ | |
1127 | cmn_err(CE_WARN, "XFS: failed to read RT inodes"); | |
f9057e3d | 1128 | goto out_rele_rip; |
1da177e4 LT |
1129 | } |
1130 | ||
1131 | /* | |
7884bc86 CH |
1132 | * If this is a read-only mount defer the superblock updates until |
1133 | * the next remount into writeable mode. Otherwise we would never | |
1134 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 1135 | */ |
7884bc86 CH |
1136 | if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
1137 | error = xfs_mount_log_sb(mp, mp->m_update_flags); | |
e5720eec DC |
1138 | if (error) { |
1139 | cmn_err(CE_WARN, "XFS: failed to write sb changes"); | |
b93b6e43 | 1140 | goto out_rtunmount; |
e5720eec DC |
1141 | } |
1142 | } | |
1da177e4 LT |
1143 | |
1144 | /* | |
1145 | * Initialise the XFS quota management subsystem for this mount | |
1146 | */ | |
0771fb45 ES |
1147 | error = XFS_QM_INIT(mp, "amount, "aflags); |
1148 | if (error) | |
b93b6e43 | 1149 | goto out_rtunmount; |
1da177e4 LT |
1150 | |
1151 | /* | |
1152 | * Finish recovering the file system. This part needed to be | |
1153 | * delayed until after the root and real-time bitmap inodes | |
1154 | * were consistently read in. | |
1155 | */ | |
4249023a | 1156 | error = xfs_log_mount_finish(mp); |
1da177e4 LT |
1157 | if (error) { |
1158 | cmn_err(CE_WARN, "XFS: log mount finish failed"); | |
b93b6e43 | 1159 | goto out_rtunmount; |
1da177e4 LT |
1160 | } |
1161 | ||
1162 | /* | |
1163 | * Complete the quota initialisation, post-log-replay component. | |
1164 | */ | |
4249023a | 1165 | error = XFS_QM_MOUNT(mp, quotamount, quotaflags); |
0771fb45 | 1166 | if (error) |
b93b6e43 | 1167 | goto out_rtunmount; |
1da177e4 | 1168 | |
84e1e99f DC |
1169 | /* |
1170 | * Now we are mounted, reserve a small amount of unused space for | |
1171 | * privileged transactions. This is needed so that transaction | |
1172 | * space required for critical operations can dip into this pool | |
1173 | * when at ENOSPC. This is needed for operations like create with | |
1174 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
1175 | * are not allowed to use this reserved space. | |
1176 | * | |
1177 | * We default to 5% or 1024 fsbs of space reserved, whichever is smaller. | |
1178 | * This may drive us straight to ENOSPC on mount, but that implies | |
714082bc | 1179 | * we were already there on the last unmount. Warn if this occurs. |
84e1e99f | 1180 | */ |
39726be2 CH |
1181 | resblks = mp->m_sb.sb_dblocks; |
1182 | do_div(resblks, 20); | |
1183 | resblks = min_t(__uint64_t, resblks, 1024); | |
714082bc DC |
1184 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1185 | if (error) | |
1186 | cmn_err(CE_WARN, "XFS: Unable to allocate reserve blocks. " | |
1187 | "Continuing without a reserve pool."); | |
84e1e99f | 1188 | |
1da177e4 LT |
1189 | return 0; |
1190 | ||
b93b6e43 CH |
1191 | out_rtunmount: |
1192 | xfs_rtunmount_inodes(mp); | |
f9057e3d | 1193 | out_rele_rip: |
43355099 | 1194 | IRELE(rip); |
f9057e3d | 1195 | out_log_dealloc: |
21b699c8 | 1196 | xfs_log_unmount(mp); |
f9057e3d | 1197 | out_free_perag: |
ff4f038c | 1198 | xfs_free_perag(mp); |
f9057e3d CH |
1199 | out_remove_uuid: |
1200 | if (!(mp->m_flags & XFS_MOUNT_NOUUID)) | |
fa6adbe0 | 1201 | uuid_table_remove(&mp->m_sb.sb_uuid); |
f9057e3d | 1202 | out: |
1da177e4 LT |
1203 | return error; |
1204 | } | |
1205 | ||
1206 | /* | |
1da177e4 LT |
1207 | * This flushes out the inodes,dquots and the superblock, unmounts the |
1208 | * log and makes sure that incore structures are freed. | |
1209 | */ | |
41b5c2e7 CH |
1210 | void |
1211 | xfs_unmountfs( | |
1212 | struct xfs_mount *mp) | |
1da177e4 | 1213 | { |
41b5c2e7 CH |
1214 | __uint64_t resblks; |
1215 | int error; | |
1da177e4 | 1216 | |
f95099ba CH |
1217 | /* |
1218 | * Release dquot that rootinode, rbmino and rsumino might be holding, | |
1219 | * and release the quota inodes. | |
1220 | */ | |
1221 | XFS_QM_UNMOUNT(mp); | |
1222 | ||
b93b6e43 | 1223 | xfs_rtunmount_inodes(mp); |
77508ec8 CH |
1224 | IRELE(mp->m_rootip); |
1225 | ||
641c56fb DC |
1226 | /* |
1227 | * We can potentially deadlock here if we have an inode cluster | |
1228 | * that has been freed has it's buffer still pinned in memory because | |
1229 | * the transaction is still sitting in a iclog. The stale inodes | |
1230 | * on that buffer will have their flush locks held until the | |
1231 | * transaction hits the disk and the callbacks run. the inode | |
1232 | * flush takes the flush lock unconditionally and with nothing to | |
1233 | * push out the iclog we will never get that unlocked. hence we | |
1234 | * need to force the log first. | |
1235 | */ | |
1236 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC); | |
1dc3318a | 1237 | xfs_reclaim_inodes(mp, 0, XFS_IFLUSH_ASYNC); |
1da177e4 | 1238 | |
ee2a4f7c | 1239 | XFS_QM_DQPURGEALL(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_UMOUNTING); |
1da177e4 | 1240 | |
a357a121 LM |
1241 | if (mp->m_quotainfo) |
1242 | XFS_QM_DONE(mp); | |
1243 | ||
1da177e4 LT |
1244 | /* |
1245 | * Flush out the log synchronously so that we know for sure | |
1246 | * that nothing is pinned. This is important because bflush() | |
1247 | * will skip pinned buffers. | |
1248 | */ | |
1249 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC); | |
1250 | ||
1251 | xfs_binval(mp->m_ddev_targp); | |
1252 | if (mp->m_rtdev_targp) { | |
1253 | xfs_binval(mp->m_rtdev_targp); | |
1254 | } | |
1255 | ||
84e1e99f DC |
1256 | /* |
1257 | * Unreserve any blocks we have so that when we unmount we don't account | |
1258 | * the reserved free space as used. This is really only necessary for | |
1259 | * lazy superblock counting because it trusts the incore superblock | |
1260 | * counters to be aboslutely correct on clean unmount. | |
1261 | * | |
1262 | * We don't bother correcting this elsewhere for lazy superblock | |
1263 | * counting because on mount of an unclean filesystem we reconstruct the | |
1264 | * correct counter value and this is irrelevant. | |
1265 | * | |
1266 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1267 | * we only every apply deltas to the superblock and hence the incore | |
1268 | * value does not matter.... | |
1269 | */ | |
1270 | resblks = 0; | |
714082bc DC |
1271 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1272 | if (error) | |
1273 | cmn_err(CE_WARN, "XFS: Unable to free reserved block pool. " | |
1274 | "Freespace may not be correct on next mount."); | |
1275 | ||
e5720eec DC |
1276 | error = xfs_log_sbcount(mp, 1); |
1277 | if (error) | |
1278 | cmn_err(CE_WARN, "XFS: Unable to update superblock counters. " | |
1279 | "Freespace may not be correct on next mount."); | |
1da177e4 | 1280 | xfs_unmountfs_writesb(mp); |
1da177e4 | 1281 | xfs_unmountfs_wait(mp); /* wait for async bufs */ |
21b699c8 CH |
1282 | xfs_log_unmount_write(mp); |
1283 | xfs_log_unmount(mp); | |
1da177e4 | 1284 | |
1da177e4 | 1285 | if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0) |
fa6adbe0 | 1286 | uuid_table_remove(&mp->m_sb.sb_uuid); |
1da177e4 | 1287 | |
1550d0b0 | 1288 | #if defined(DEBUG) |
0ce4cfd4 | 1289 | xfs_errortag_clearall(mp, 0); |
1da177e4 | 1290 | #endif |
ff4f038c | 1291 | xfs_free_perag(mp); |
1da177e4 LT |
1292 | } |
1293 | ||
ba0f32d4 | 1294 | STATIC void |
1da177e4 LT |
1295 | xfs_unmountfs_wait(xfs_mount_t *mp) |
1296 | { | |
1297 | if (mp->m_logdev_targp != mp->m_ddev_targp) | |
1298 | xfs_wait_buftarg(mp->m_logdev_targp); | |
1299 | if (mp->m_rtdev_targp) | |
1300 | xfs_wait_buftarg(mp->m_rtdev_targp); | |
1301 | xfs_wait_buftarg(mp->m_ddev_targp); | |
1302 | } | |
1303 | ||
92821e2b DC |
1304 | int |
1305 | xfs_fs_writable(xfs_mount_t *mp) | |
1306 | { | |
b267ce99 | 1307 | return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) || |
bd186aa9 | 1308 | (mp->m_flags & XFS_MOUNT_RDONLY)); |
92821e2b DC |
1309 | } |
1310 | ||
1311 | /* | |
1312 | * xfs_log_sbcount | |
1313 | * | |
1314 | * Called either periodically to keep the on disk superblock values | |
1315 | * roughly up to date or from unmount to make sure the values are | |
1316 | * correct on a clean unmount. | |
1317 | * | |
1318 | * Note this code can be called during the process of freezing, so | |
1319 | * we may need to use the transaction allocator which does not not | |
1320 | * block when the transaction subsystem is in its frozen state. | |
1321 | */ | |
1322 | int | |
1323 | xfs_log_sbcount( | |
1324 | xfs_mount_t *mp, | |
1325 | uint sync) | |
1326 | { | |
1327 | xfs_trans_t *tp; | |
1328 | int error; | |
1329 | ||
1330 | if (!xfs_fs_writable(mp)) | |
1331 | return 0; | |
1332 | ||
d4d90b57 | 1333 | xfs_icsb_sync_counters(mp, 0); |
92821e2b DC |
1334 | |
1335 | /* | |
1336 | * we don't need to do this if we are updating the superblock | |
1337 | * counters on every modification. | |
1338 | */ | |
1339 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1340 | return 0; | |
1341 | ||
1342 | tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT); | |
1343 | error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0, | |
1344 | XFS_DEFAULT_LOG_COUNT); | |
1345 | if (error) { | |
1346 | xfs_trans_cancel(tp, 0); | |
1347 | return error; | |
1348 | } | |
1349 | ||
1350 | xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS); | |
1351 | if (sync) | |
1352 | xfs_trans_set_sync(tp); | |
e5720eec DC |
1353 | error = xfs_trans_commit(tp, 0); |
1354 | return error; | |
92821e2b DC |
1355 | } |
1356 | ||
1da177e4 LT |
1357 | int |
1358 | xfs_unmountfs_writesb(xfs_mount_t *mp) | |
1359 | { | |
1360 | xfs_buf_t *sbp; | |
1da177e4 LT |
1361 | int error = 0; |
1362 | ||
1363 | /* | |
1364 | * skip superblock write if fs is read-only, or | |
1365 | * if we are doing a forced umount. | |
1366 | */ | |
bd186aa9 | 1367 | if (!((mp->m_flags & XFS_MOUNT_RDONLY) || |
1da177e4 | 1368 | XFS_FORCED_SHUTDOWN(mp))) { |
8d280b98 | 1369 | |
92821e2b | 1370 | sbp = xfs_getsb(mp, 0); |
8d280b98 | 1371 | |
1da177e4 LT |
1372 | XFS_BUF_UNDONE(sbp); |
1373 | XFS_BUF_UNREAD(sbp); | |
1374 | XFS_BUF_UNDELAYWRITE(sbp); | |
1375 | XFS_BUF_WRITE(sbp); | |
1376 | XFS_BUF_UNASYNC(sbp); | |
1377 | ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp); | |
1378 | xfsbdstrat(mp, sbp); | |
1da177e4 LT |
1379 | error = xfs_iowait(sbp); |
1380 | if (error) | |
1381 | xfs_ioerror_alert("xfs_unmountfs_writesb", | |
1382 | mp, sbp, XFS_BUF_ADDR(sbp)); | |
92821e2b | 1383 | xfs_buf_relse(sbp); |
1da177e4 | 1384 | } |
014c2544 | 1385 | return error; |
1da177e4 LT |
1386 | } |
1387 | ||
1388 | /* | |
1389 | * xfs_mod_sb() can be used to copy arbitrary changes to the | |
1390 | * in-core superblock into the superblock buffer to be logged. | |
1391 | * It does not provide the higher level of locking that is | |
1392 | * needed to protect the in-core superblock from concurrent | |
1393 | * access. | |
1394 | */ | |
1395 | void | |
1396 | xfs_mod_sb(xfs_trans_t *tp, __int64_t fields) | |
1397 | { | |
1398 | xfs_buf_t *bp; | |
1399 | int first; | |
1400 | int last; | |
1401 | xfs_mount_t *mp; | |
1da177e4 LT |
1402 | xfs_sb_field_t f; |
1403 | ||
1404 | ASSERT(fields); | |
1405 | if (!fields) | |
1406 | return; | |
1407 | mp = tp->t_mountp; | |
1408 | bp = xfs_trans_getsb(tp, mp, 0); | |
1da177e4 LT |
1409 | first = sizeof(xfs_sb_t); |
1410 | last = 0; | |
1411 | ||
1412 | /* translate/copy */ | |
1413 | ||
2bdf7cd0 | 1414 | xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields); |
1da177e4 LT |
1415 | |
1416 | /* find modified range */ | |
1417 | ||
1418 | f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields); | |
1419 | ASSERT((1LL << f) & XFS_SB_MOD_BITS); | |
1420 | first = xfs_sb_info[f].offset; | |
1421 | ||
1422 | f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields); | |
1423 | ASSERT((1LL << f) & XFS_SB_MOD_BITS); | |
1424 | last = xfs_sb_info[f + 1].offset - 1; | |
1425 | ||
1426 | xfs_trans_log_buf(tp, bp, first, last); | |
1427 | } | |
d210a28c | 1428 | |
d210a28c | 1429 | |
1da177e4 LT |
1430 | /* |
1431 | * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply | |
1432 | * a delta to a specified field in the in-core superblock. Simply | |
1433 | * switch on the field indicated and apply the delta to that field. | |
1434 | * Fields are not allowed to dip below zero, so if the delta would | |
1435 | * do this do not apply it and return EINVAL. | |
1436 | * | |
3685c2a1 | 1437 | * The m_sb_lock must be held when this routine is called. |
1da177e4 | 1438 | */ |
8d280b98 | 1439 | int |
20f4ebf2 DC |
1440 | xfs_mod_incore_sb_unlocked( |
1441 | xfs_mount_t *mp, | |
1442 | xfs_sb_field_t field, | |
1443 | int64_t delta, | |
1444 | int rsvd) | |
1da177e4 LT |
1445 | { |
1446 | int scounter; /* short counter for 32 bit fields */ | |
1447 | long long lcounter; /* long counter for 64 bit fields */ | |
1448 | long long res_used, rem; | |
1449 | ||
1450 | /* | |
1451 | * With the in-core superblock spin lock held, switch | |
1452 | * on the indicated field. Apply the delta to the | |
1453 | * proper field. If the fields value would dip below | |
1454 | * 0, then do not apply the delta and return EINVAL. | |
1455 | */ | |
1456 | switch (field) { | |
1457 | case XFS_SBS_ICOUNT: | |
1458 | lcounter = (long long)mp->m_sb.sb_icount; | |
1459 | lcounter += delta; | |
1460 | if (lcounter < 0) { | |
1461 | ASSERT(0); | |
014c2544 | 1462 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1463 | } |
1464 | mp->m_sb.sb_icount = lcounter; | |
014c2544 | 1465 | return 0; |
1da177e4 LT |
1466 | case XFS_SBS_IFREE: |
1467 | lcounter = (long long)mp->m_sb.sb_ifree; | |
1468 | lcounter += delta; | |
1469 | if (lcounter < 0) { | |
1470 | ASSERT(0); | |
014c2544 | 1471 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1472 | } |
1473 | mp->m_sb.sb_ifree = lcounter; | |
014c2544 | 1474 | return 0; |
1da177e4 | 1475 | case XFS_SBS_FDBLOCKS: |
4be536de DC |
1476 | lcounter = (long long) |
1477 | mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); | |
1da177e4 LT |
1478 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); |
1479 | ||
1480 | if (delta > 0) { /* Putting blocks back */ | |
1481 | if (res_used > delta) { | |
1482 | mp->m_resblks_avail += delta; | |
1483 | } else { | |
1484 | rem = delta - res_used; | |
1485 | mp->m_resblks_avail = mp->m_resblks; | |
1486 | lcounter += rem; | |
1487 | } | |
1488 | } else { /* Taking blocks away */ | |
1489 | ||
1490 | lcounter += delta; | |
1491 | ||
1492 | /* | |
1493 | * If were out of blocks, use any available reserved blocks if | |
1494 | * were allowed to. | |
1495 | */ | |
1496 | ||
1497 | if (lcounter < 0) { | |
1498 | if (rsvd) { | |
1499 | lcounter = (long long)mp->m_resblks_avail + delta; | |
1500 | if (lcounter < 0) { | |
014c2544 | 1501 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1502 | } |
1503 | mp->m_resblks_avail = lcounter; | |
014c2544 | 1504 | return 0; |
1da177e4 | 1505 | } else { /* not reserved */ |
014c2544 | 1506 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1507 | } |
1508 | } | |
1509 | } | |
1510 | ||
4be536de | 1511 | mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
014c2544 | 1512 | return 0; |
1da177e4 LT |
1513 | case XFS_SBS_FREXTENTS: |
1514 | lcounter = (long long)mp->m_sb.sb_frextents; | |
1515 | lcounter += delta; | |
1516 | if (lcounter < 0) { | |
014c2544 | 1517 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1518 | } |
1519 | mp->m_sb.sb_frextents = lcounter; | |
014c2544 | 1520 | return 0; |
1da177e4 LT |
1521 | case XFS_SBS_DBLOCKS: |
1522 | lcounter = (long long)mp->m_sb.sb_dblocks; | |
1523 | lcounter += delta; | |
1524 | if (lcounter < 0) { | |
1525 | ASSERT(0); | |
014c2544 | 1526 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1527 | } |
1528 | mp->m_sb.sb_dblocks = lcounter; | |
014c2544 | 1529 | return 0; |
1da177e4 LT |
1530 | case XFS_SBS_AGCOUNT: |
1531 | scounter = mp->m_sb.sb_agcount; | |
1532 | scounter += delta; | |
1533 | if (scounter < 0) { | |
1534 | ASSERT(0); | |
014c2544 | 1535 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1536 | } |
1537 | mp->m_sb.sb_agcount = scounter; | |
014c2544 | 1538 | return 0; |
1da177e4 LT |
1539 | case XFS_SBS_IMAX_PCT: |
1540 | scounter = mp->m_sb.sb_imax_pct; | |
1541 | scounter += delta; | |
1542 | if (scounter < 0) { | |
1543 | ASSERT(0); | |
014c2544 | 1544 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1545 | } |
1546 | mp->m_sb.sb_imax_pct = scounter; | |
014c2544 | 1547 | return 0; |
1da177e4 LT |
1548 | case XFS_SBS_REXTSIZE: |
1549 | scounter = mp->m_sb.sb_rextsize; | |
1550 | scounter += delta; | |
1551 | if (scounter < 0) { | |
1552 | ASSERT(0); | |
014c2544 | 1553 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1554 | } |
1555 | mp->m_sb.sb_rextsize = scounter; | |
014c2544 | 1556 | return 0; |
1da177e4 LT |
1557 | case XFS_SBS_RBMBLOCKS: |
1558 | scounter = mp->m_sb.sb_rbmblocks; | |
1559 | scounter += delta; | |
1560 | if (scounter < 0) { | |
1561 | ASSERT(0); | |
014c2544 | 1562 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1563 | } |
1564 | mp->m_sb.sb_rbmblocks = scounter; | |
014c2544 | 1565 | return 0; |
1da177e4 LT |
1566 | case XFS_SBS_RBLOCKS: |
1567 | lcounter = (long long)mp->m_sb.sb_rblocks; | |
1568 | lcounter += delta; | |
1569 | if (lcounter < 0) { | |
1570 | ASSERT(0); | |
014c2544 | 1571 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1572 | } |
1573 | mp->m_sb.sb_rblocks = lcounter; | |
014c2544 | 1574 | return 0; |
1da177e4 LT |
1575 | case XFS_SBS_REXTENTS: |
1576 | lcounter = (long long)mp->m_sb.sb_rextents; | |
1577 | lcounter += delta; | |
1578 | if (lcounter < 0) { | |
1579 | ASSERT(0); | |
014c2544 | 1580 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1581 | } |
1582 | mp->m_sb.sb_rextents = lcounter; | |
014c2544 | 1583 | return 0; |
1da177e4 LT |
1584 | case XFS_SBS_REXTSLOG: |
1585 | scounter = mp->m_sb.sb_rextslog; | |
1586 | scounter += delta; | |
1587 | if (scounter < 0) { | |
1588 | ASSERT(0); | |
014c2544 | 1589 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1590 | } |
1591 | mp->m_sb.sb_rextslog = scounter; | |
014c2544 | 1592 | return 0; |
1da177e4 LT |
1593 | default: |
1594 | ASSERT(0); | |
014c2544 | 1595 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1596 | } |
1597 | } | |
1598 | ||
1599 | /* | |
1600 | * xfs_mod_incore_sb() is used to change a field in the in-core | |
1601 | * superblock structure by the specified delta. This modification | |
3685c2a1 | 1602 | * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked() |
1da177e4 LT |
1603 | * routine to do the work. |
1604 | */ | |
1605 | int | |
20f4ebf2 DC |
1606 | xfs_mod_incore_sb( |
1607 | xfs_mount_t *mp, | |
1608 | xfs_sb_field_t field, | |
1609 | int64_t delta, | |
1610 | int rsvd) | |
1da177e4 | 1611 | { |
1da177e4 LT |
1612 | int status; |
1613 | ||
8d280b98 DC |
1614 | /* check for per-cpu counters */ |
1615 | switch (field) { | |
1616 | #ifdef HAVE_PERCPU_SB | |
1617 | case XFS_SBS_ICOUNT: | |
1618 | case XFS_SBS_IFREE: | |
1619 | case XFS_SBS_FDBLOCKS: | |
1620 | if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) { | |
1621 | status = xfs_icsb_modify_counters(mp, field, | |
1622 | delta, rsvd); | |
1623 | break; | |
1624 | } | |
1625 | /* FALLTHROUGH */ | |
1626 | #endif | |
1627 | default: | |
3685c2a1 | 1628 | spin_lock(&mp->m_sb_lock); |
8d280b98 | 1629 | status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); |
3685c2a1 | 1630 | spin_unlock(&mp->m_sb_lock); |
8d280b98 DC |
1631 | break; |
1632 | } | |
1633 | ||
014c2544 | 1634 | return status; |
1da177e4 LT |
1635 | } |
1636 | ||
1637 | /* | |
1638 | * xfs_mod_incore_sb_batch() is used to change more than one field | |
1639 | * in the in-core superblock structure at a time. This modification | |
1640 | * is protected by a lock internal to this module. The fields and | |
1641 | * changes to those fields are specified in the array of xfs_mod_sb | |
1642 | * structures passed in. | |
1643 | * | |
1644 | * Either all of the specified deltas will be applied or none of | |
1645 | * them will. If any modified field dips below 0, then all modifications | |
1646 | * will be backed out and EINVAL will be returned. | |
1647 | */ | |
1648 | int | |
1649 | xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd) | |
1650 | { | |
1da177e4 LT |
1651 | int status=0; |
1652 | xfs_mod_sb_t *msbp; | |
1653 | ||
1654 | /* | |
1655 | * Loop through the array of mod structures and apply each | |
1656 | * individually. If any fail, then back out all those | |
1657 | * which have already been applied. Do all of this within | |
3685c2a1 | 1658 | * the scope of the m_sb_lock so that all of the changes will |
1da177e4 LT |
1659 | * be atomic. |
1660 | */ | |
3685c2a1 | 1661 | spin_lock(&mp->m_sb_lock); |
1da177e4 LT |
1662 | msbp = &msb[0]; |
1663 | for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) { | |
1664 | /* | |
1665 | * Apply the delta at index n. If it fails, break | |
1666 | * from the loop so we'll fall into the undo loop | |
1667 | * below. | |
1668 | */ | |
8d280b98 DC |
1669 | switch (msbp->msb_field) { |
1670 | #ifdef HAVE_PERCPU_SB | |
1671 | case XFS_SBS_ICOUNT: | |
1672 | case XFS_SBS_IFREE: | |
1673 | case XFS_SBS_FDBLOCKS: | |
1674 | if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) { | |
3685c2a1 | 1675 | spin_unlock(&mp->m_sb_lock); |
20b64285 | 1676 | status = xfs_icsb_modify_counters(mp, |
8d280b98 DC |
1677 | msbp->msb_field, |
1678 | msbp->msb_delta, rsvd); | |
3685c2a1 | 1679 | spin_lock(&mp->m_sb_lock); |
8d280b98 DC |
1680 | break; |
1681 | } | |
1682 | /* FALLTHROUGH */ | |
1683 | #endif | |
1684 | default: | |
1685 | status = xfs_mod_incore_sb_unlocked(mp, | |
1686 | msbp->msb_field, | |
1687 | msbp->msb_delta, rsvd); | |
1688 | break; | |
1689 | } | |
1690 | ||
1da177e4 LT |
1691 | if (status != 0) { |
1692 | break; | |
1693 | } | |
1694 | } | |
1695 | ||
1696 | /* | |
1697 | * If we didn't complete the loop above, then back out | |
1698 | * any changes made to the superblock. If you add code | |
1699 | * between the loop above and here, make sure that you | |
1700 | * preserve the value of status. Loop back until | |
1701 | * we step below the beginning of the array. Make sure | |
1702 | * we don't touch anything back there. | |
1703 | */ | |
1704 | if (status != 0) { | |
1705 | msbp--; | |
1706 | while (msbp >= msb) { | |
8d280b98 DC |
1707 | switch (msbp->msb_field) { |
1708 | #ifdef HAVE_PERCPU_SB | |
1709 | case XFS_SBS_ICOUNT: | |
1710 | case XFS_SBS_IFREE: | |
1711 | case XFS_SBS_FDBLOCKS: | |
1712 | if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) { | |
3685c2a1 | 1713 | spin_unlock(&mp->m_sb_lock); |
20b64285 | 1714 | status = xfs_icsb_modify_counters(mp, |
8d280b98 DC |
1715 | msbp->msb_field, |
1716 | -(msbp->msb_delta), | |
1717 | rsvd); | |
3685c2a1 | 1718 | spin_lock(&mp->m_sb_lock); |
8d280b98 DC |
1719 | break; |
1720 | } | |
1721 | /* FALLTHROUGH */ | |
1722 | #endif | |
1723 | default: | |
1724 | status = xfs_mod_incore_sb_unlocked(mp, | |
1725 | msbp->msb_field, | |
1726 | -(msbp->msb_delta), | |
1727 | rsvd); | |
1728 | break; | |
1729 | } | |
1da177e4 LT |
1730 | ASSERT(status == 0); |
1731 | msbp--; | |
1732 | } | |
1733 | } | |
3685c2a1 | 1734 | spin_unlock(&mp->m_sb_lock); |
014c2544 | 1735 | return status; |
1da177e4 LT |
1736 | } |
1737 | ||
1738 | /* | |
1739 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
1740 | * The buffer is returned locked and read in from disk. | |
1741 | * The buffer should be released with a call to xfs_brelse(). | |
1742 | * | |
1743 | * If the flags parameter is BUF_TRYLOCK, then we'll only return | |
1744 | * the superblock buffer if it can be locked without sleeping. | |
1745 | * If it can't then we'll return NULL. | |
1746 | */ | |
1747 | xfs_buf_t * | |
1748 | xfs_getsb( | |
1749 | xfs_mount_t *mp, | |
1750 | int flags) | |
1751 | { | |
1752 | xfs_buf_t *bp; | |
1753 | ||
1754 | ASSERT(mp->m_sb_bp != NULL); | |
1755 | bp = mp->m_sb_bp; | |
1756 | if (flags & XFS_BUF_TRYLOCK) { | |
1757 | if (!XFS_BUF_CPSEMA(bp)) { | |
1758 | return NULL; | |
1759 | } | |
1760 | } else { | |
1761 | XFS_BUF_PSEMA(bp, PRIBIO); | |
1762 | } | |
1763 | XFS_BUF_HOLD(bp); | |
1764 | ASSERT(XFS_BUF_ISDONE(bp)); | |
014c2544 | 1765 | return bp; |
1da177e4 LT |
1766 | } |
1767 | ||
1768 | /* | |
1769 | * Used to free the superblock along various error paths. | |
1770 | */ | |
1771 | void | |
1772 | xfs_freesb( | |
1773 | xfs_mount_t *mp) | |
1774 | { | |
1775 | xfs_buf_t *bp; | |
1776 | ||
1777 | /* | |
1778 | * Use xfs_getsb() so that the buffer will be locked | |
1779 | * when we call xfs_buf_relse(). | |
1780 | */ | |
1781 | bp = xfs_getsb(mp, 0); | |
1782 | XFS_BUF_UNMANAGE(bp); | |
1783 | xfs_buf_relse(bp); | |
1784 | mp->m_sb_bp = NULL; | |
1785 | } | |
1786 | ||
1787 | /* | |
1788 | * See if the UUID is unique among mounted XFS filesystems. | |
1789 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. | |
1790 | */ | |
1791 | STATIC int | |
1792 | xfs_uuid_mount( | |
1793 | xfs_mount_t *mp) | |
1794 | { | |
1795 | if (uuid_is_nil(&mp->m_sb.sb_uuid)) { | |
1796 | cmn_err(CE_WARN, | |
1797 | "XFS: Filesystem %s has nil UUID - can't mount", | |
1798 | mp->m_fsname); | |
1799 | return -1; | |
1800 | } | |
1801 | if (!uuid_table_insert(&mp->m_sb.sb_uuid)) { | |
1802 | cmn_err(CE_WARN, | |
1803 | "XFS: Filesystem %s has duplicate UUID - can't mount", | |
1804 | mp->m_fsname); | |
1805 | return -1; | |
1806 | } | |
1807 | return 0; | |
1808 | } | |
1809 | ||
1da177e4 LT |
1810 | /* |
1811 | * Used to log changes to the superblock unit and width fields which could | |
e6957ea4 ES |
1812 | * be altered by the mount options, as well as any potential sb_features2 |
1813 | * fixup. Only the first superblock is updated. | |
1da177e4 | 1814 | */ |
7884bc86 | 1815 | int |
ee1c0908 | 1816 | xfs_mount_log_sb( |
1da177e4 LT |
1817 | xfs_mount_t *mp, |
1818 | __int64_t fields) | |
1819 | { | |
1820 | xfs_trans_t *tp; | |
e5720eec | 1821 | int error; |
1da177e4 | 1822 | |
ee1c0908 | 1823 | ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID | |
4b166de0 DC |
1824 | XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 | |
1825 | XFS_SB_VERSIONNUM)); | |
1da177e4 LT |
1826 | |
1827 | tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT); | |
e5720eec DC |
1828 | error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0, |
1829 | XFS_DEFAULT_LOG_COUNT); | |
1830 | if (error) { | |
1da177e4 | 1831 | xfs_trans_cancel(tp, 0); |
e5720eec | 1832 | return error; |
1da177e4 LT |
1833 | } |
1834 | xfs_mod_sb(tp, fields); | |
e5720eec DC |
1835 | error = xfs_trans_commit(tp, 0); |
1836 | return error; | |
1da177e4 | 1837 | } |
8d280b98 DC |
1838 | |
1839 | ||
1840 | #ifdef HAVE_PERCPU_SB | |
1841 | /* | |
1842 | * Per-cpu incore superblock counters | |
1843 | * | |
1844 | * Simple concept, difficult implementation | |
1845 | * | |
1846 | * Basically, replace the incore superblock counters with a distributed per cpu | |
1847 | * counter for contended fields (e.g. free block count). | |
1848 | * | |
1849 | * Difficulties arise in that the incore sb is used for ENOSPC checking, and | |
1850 | * hence needs to be accurately read when we are running low on space. Hence | |
1851 | * there is a method to enable and disable the per-cpu counters based on how | |
1852 | * much "stuff" is available in them. | |
1853 | * | |
1854 | * Basically, a counter is enabled if there is enough free resource to justify | |
1855 | * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local | |
1856 | * ENOSPC), then we disable the counters to synchronise all callers and | |
1857 | * re-distribute the available resources. | |
1858 | * | |
1859 | * If, once we redistributed the available resources, we still get a failure, | |
1860 | * we disable the per-cpu counter and go through the slow path. | |
1861 | * | |
1862 | * The slow path is the current xfs_mod_incore_sb() function. This means that | |
1863 | * when we disable a per-cpu counter, we need to drain it's resources back to | |
1864 | * the global superblock. We do this after disabling the counter to prevent | |
1865 | * more threads from queueing up on the counter. | |
1866 | * | |
1867 | * Essentially, this means that we still need a lock in the fast path to enable | |
1868 | * synchronisation between the global counters and the per-cpu counters. This | |
1869 | * is not a problem because the lock will be local to a CPU almost all the time | |
1870 | * and have little contention except when we get to ENOSPC conditions. | |
1871 | * | |
1872 | * Basically, this lock becomes a barrier that enables us to lock out the fast | |
1873 | * path while we do things like enabling and disabling counters and | |
1874 | * synchronising the counters. | |
1875 | * | |
1876 | * Locking rules: | |
1877 | * | |
3685c2a1 | 1878 | * 1. m_sb_lock before picking up per-cpu locks |
8d280b98 | 1879 | * 2. per-cpu locks always picked up via for_each_online_cpu() order |
3685c2a1 | 1880 | * 3. accurate counter sync requires m_sb_lock + per cpu locks |
8d280b98 | 1881 | * 4. modifying per-cpu counters requires holding per-cpu lock |
3685c2a1 ES |
1882 | * 5. modifying global counters requires holding m_sb_lock |
1883 | * 6. enabling or disabling a counter requires holding the m_sb_lock | |
8d280b98 DC |
1884 | * and _none_ of the per-cpu locks. |
1885 | * | |
1886 | * Disabled counters are only ever re-enabled by a balance operation | |
1887 | * that results in more free resources per CPU than a given threshold. | |
1888 | * To ensure counters don't remain disabled, they are rebalanced when | |
1889 | * the global resource goes above a higher threshold (i.e. some hysteresis | |
1890 | * is present to prevent thrashing). | |
e8234a68 DC |
1891 | */ |
1892 | ||
5a67e4c5 | 1893 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
1894 | /* |
1895 | * hot-plug CPU notifier support. | |
8d280b98 | 1896 | * |
5a67e4c5 CS |
1897 | * We need a notifier per filesystem as we need to be able to identify |
1898 | * the filesystem to balance the counters out. This is achieved by | |
1899 | * having a notifier block embedded in the xfs_mount_t and doing pointer | |
1900 | * magic to get the mount pointer from the notifier block address. | |
8d280b98 | 1901 | */ |
e8234a68 DC |
1902 | STATIC int |
1903 | xfs_icsb_cpu_notify( | |
1904 | struct notifier_block *nfb, | |
1905 | unsigned long action, | |
1906 | void *hcpu) | |
1907 | { | |
1908 | xfs_icsb_cnts_t *cntp; | |
1909 | xfs_mount_t *mp; | |
e8234a68 DC |
1910 | |
1911 | mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier); | |
1912 | cntp = (xfs_icsb_cnts_t *) | |
1913 | per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu); | |
1914 | switch (action) { | |
1915 | case CPU_UP_PREPARE: | |
8bb78442 | 1916 | case CPU_UP_PREPARE_FROZEN: |
e8234a68 DC |
1917 | /* Easy Case - initialize the area and locks, and |
1918 | * then rebalance when online does everything else for us. */ | |
01e1b69c | 1919 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 DC |
1920 | break; |
1921 | case CPU_ONLINE: | |
8bb78442 | 1922 | case CPU_ONLINE_FROZEN: |
03135cf7 | 1923 | xfs_icsb_lock(mp); |
45af6c6d CH |
1924 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
1925 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
1926 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
03135cf7 | 1927 | xfs_icsb_unlock(mp); |
e8234a68 DC |
1928 | break; |
1929 | case CPU_DEAD: | |
8bb78442 | 1930 | case CPU_DEAD_FROZEN: |
e8234a68 DC |
1931 | /* Disable all the counters, then fold the dead cpu's |
1932 | * count into the total on the global superblock and | |
1933 | * re-enable the counters. */ | |
03135cf7 | 1934 | xfs_icsb_lock(mp); |
3685c2a1 | 1935 | spin_lock(&mp->m_sb_lock); |
e8234a68 DC |
1936 | xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT); |
1937 | xfs_icsb_disable_counter(mp, XFS_SBS_IFREE); | |
1938 | xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS); | |
1939 | ||
1940 | mp->m_sb.sb_icount += cntp->icsb_icount; | |
1941 | mp->m_sb.sb_ifree += cntp->icsb_ifree; | |
1942 | mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks; | |
1943 | ||
01e1b69c | 1944 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 | 1945 | |
45af6c6d CH |
1946 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0); |
1947 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0); | |
1948 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0); | |
3685c2a1 | 1949 | spin_unlock(&mp->m_sb_lock); |
03135cf7 | 1950 | xfs_icsb_unlock(mp); |
e8234a68 DC |
1951 | break; |
1952 | } | |
1953 | ||
1954 | return NOTIFY_OK; | |
1955 | } | |
5a67e4c5 | 1956 | #endif /* CONFIG_HOTPLUG_CPU */ |
e8234a68 | 1957 | |
8d280b98 DC |
1958 | int |
1959 | xfs_icsb_init_counters( | |
1960 | xfs_mount_t *mp) | |
1961 | { | |
1962 | xfs_icsb_cnts_t *cntp; | |
1963 | int i; | |
1964 | ||
1965 | mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t); | |
1966 | if (mp->m_sb_cnts == NULL) | |
1967 | return -ENOMEM; | |
1968 | ||
5a67e4c5 | 1969 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
1970 | mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify; |
1971 | mp->m_icsb_notifier.priority = 0; | |
5a67e4c5 CS |
1972 | register_hotcpu_notifier(&mp->m_icsb_notifier); |
1973 | #endif /* CONFIG_HOTPLUG_CPU */ | |
e8234a68 | 1974 | |
8d280b98 DC |
1975 | for_each_online_cpu(i) { |
1976 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 1977 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
8d280b98 | 1978 | } |
20b64285 DC |
1979 | |
1980 | mutex_init(&mp->m_icsb_mutex); | |
1981 | ||
8d280b98 DC |
1982 | /* |
1983 | * start with all counters disabled so that the | |
1984 | * initial balance kicks us off correctly | |
1985 | */ | |
1986 | mp->m_icsb_counters = -1; | |
1987 | return 0; | |
1988 | } | |
1989 | ||
5478eead LM |
1990 | void |
1991 | xfs_icsb_reinit_counters( | |
1992 | xfs_mount_t *mp) | |
1993 | { | |
1994 | xfs_icsb_lock(mp); | |
1995 | /* | |
1996 | * start with all counters disabled so that the | |
1997 | * initial balance kicks us off correctly | |
1998 | */ | |
1999 | mp->m_icsb_counters = -1; | |
45af6c6d CH |
2000 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
2001 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
2002 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
5478eead LM |
2003 | xfs_icsb_unlock(mp); |
2004 | } | |
2005 | ||
c962fb79 | 2006 | void |
8d280b98 DC |
2007 | xfs_icsb_destroy_counters( |
2008 | xfs_mount_t *mp) | |
2009 | { | |
e8234a68 | 2010 | if (mp->m_sb_cnts) { |
5a67e4c5 | 2011 | unregister_hotcpu_notifier(&mp->m_icsb_notifier); |
8d280b98 | 2012 | free_percpu(mp->m_sb_cnts); |
e8234a68 | 2013 | } |
03135cf7 | 2014 | mutex_destroy(&mp->m_icsb_mutex); |
8d280b98 DC |
2015 | } |
2016 | ||
7989cb8e | 2017 | STATIC_INLINE void |
01e1b69c DC |
2018 | xfs_icsb_lock_cntr( |
2019 | xfs_icsb_cnts_t *icsbp) | |
2020 | { | |
2021 | while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) { | |
2022 | ndelay(1000); | |
2023 | } | |
2024 | } | |
2025 | ||
7989cb8e | 2026 | STATIC_INLINE void |
01e1b69c DC |
2027 | xfs_icsb_unlock_cntr( |
2028 | xfs_icsb_cnts_t *icsbp) | |
2029 | { | |
2030 | clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags); | |
2031 | } | |
2032 | ||
8d280b98 | 2033 | |
7989cb8e | 2034 | STATIC_INLINE void |
8d280b98 DC |
2035 | xfs_icsb_lock_all_counters( |
2036 | xfs_mount_t *mp) | |
2037 | { | |
2038 | xfs_icsb_cnts_t *cntp; | |
2039 | int i; | |
2040 | ||
2041 | for_each_online_cpu(i) { | |
2042 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2043 | xfs_icsb_lock_cntr(cntp); |
8d280b98 DC |
2044 | } |
2045 | } | |
2046 | ||
7989cb8e | 2047 | STATIC_INLINE void |
8d280b98 DC |
2048 | xfs_icsb_unlock_all_counters( |
2049 | xfs_mount_t *mp) | |
2050 | { | |
2051 | xfs_icsb_cnts_t *cntp; | |
2052 | int i; | |
2053 | ||
2054 | for_each_online_cpu(i) { | |
2055 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2056 | xfs_icsb_unlock_cntr(cntp); |
8d280b98 DC |
2057 | } |
2058 | } | |
2059 | ||
2060 | STATIC void | |
2061 | xfs_icsb_count( | |
2062 | xfs_mount_t *mp, | |
2063 | xfs_icsb_cnts_t *cnt, | |
2064 | int flags) | |
2065 | { | |
2066 | xfs_icsb_cnts_t *cntp; | |
2067 | int i; | |
2068 | ||
2069 | memset(cnt, 0, sizeof(xfs_icsb_cnts_t)); | |
2070 | ||
2071 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
2072 | xfs_icsb_lock_all_counters(mp); | |
2073 | ||
2074 | for_each_online_cpu(i) { | |
2075 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
2076 | cnt->icsb_icount += cntp->icsb_icount; | |
2077 | cnt->icsb_ifree += cntp->icsb_ifree; | |
2078 | cnt->icsb_fdblocks += cntp->icsb_fdblocks; | |
2079 | } | |
2080 | ||
2081 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
2082 | xfs_icsb_unlock_all_counters(mp); | |
2083 | } | |
2084 | ||
2085 | STATIC int | |
2086 | xfs_icsb_counter_disabled( | |
2087 | xfs_mount_t *mp, | |
2088 | xfs_sb_field_t field) | |
2089 | { | |
2090 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2091 | return test_bit(field, &mp->m_icsb_counters); | |
2092 | } | |
2093 | ||
36fbe6e6 | 2094 | STATIC void |
8d280b98 DC |
2095 | xfs_icsb_disable_counter( |
2096 | xfs_mount_t *mp, | |
2097 | xfs_sb_field_t field) | |
2098 | { | |
2099 | xfs_icsb_cnts_t cnt; | |
2100 | ||
2101 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2102 | ||
20b64285 DC |
2103 | /* |
2104 | * If we are already disabled, then there is nothing to do | |
2105 | * here. We check before locking all the counters to avoid | |
2106 | * the expensive lock operation when being called in the | |
2107 | * slow path and the counter is already disabled. This is | |
2108 | * safe because the only time we set or clear this state is under | |
2109 | * the m_icsb_mutex. | |
2110 | */ | |
2111 | if (xfs_icsb_counter_disabled(mp, field)) | |
36fbe6e6 | 2112 | return; |
20b64285 | 2113 | |
8d280b98 DC |
2114 | xfs_icsb_lock_all_counters(mp); |
2115 | if (!test_and_set_bit(field, &mp->m_icsb_counters)) { | |
2116 | /* drain back to superblock */ | |
2117 | ||
ce46193b | 2118 | xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT); |
8d280b98 DC |
2119 | switch(field) { |
2120 | case XFS_SBS_ICOUNT: | |
2121 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
2122 | break; | |
2123 | case XFS_SBS_IFREE: | |
2124 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
2125 | break; | |
2126 | case XFS_SBS_FDBLOCKS: | |
2127 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
2128 | break; | |
2129 | default: | |
2130 | BUG(); | |
2131 | } | |
2132 | } | |
2133 | ||
2134 | xfs_icsb_unlock_all_counters(mp); | |
8d280b98 DC |
2135 | } |
2136 | ||
2137 | STATIC void | |
2138 | xfs_icsb_enable_counter( | |
2139 | xfs_mount_t *mp, | |
2140 | xfs_sb_field_t field, | |
2141 | uint64_t count, | |
2142 | uint64_t resid) | |
2143 | { | |
2144 | xfs_icsb_cnts_t *cntp; | |
2145 | int i; | |
2146 | ||
2147 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2148 | ||
2149 | xfs_icsb_lock_all_counters(mp); | |
2150 | for_each_online_cpu(i) { | |
2151 | cntp = per_cpu_ptr(mp->m_sb_cnts, i); | |
2152 | switch (field) { | |
2153 | case XFS_SBS_ICOUNT: | |
2154 | cntp->icsb_icount = count + resid; | |
2155 | break; | |
2156 | case XFS_SBS_IFREE: | |
2157 | cntp->icsb_ifree = count + resid; | |
2158 | break; | |
2159 | case XFS_SBS_FDBLOCKS: | |
2160 | cntp->icsb_fdblocks = count + resid; | |
2161 | break; | |
2162 | default: | |
2163 | BUG(); | |
2164 | break; | |
2165 | } | |
2166 | resid = 0; | |
2167 | } | |
2168 | clear_bit(field, &mp->m_icsb_counters); | |
2169 | xfs_icsb_unlock_all_counters(mp); | |
2170 | } | |
2171 | ||
dbcabad1 | 2172 | void |
d4d90b57 | 2173 | xfs_icsb_sync_counters_locked( |
8d280b98 DC |
2174 | xfs_mount_t *mp, |
2175 | int flags) | |
2176 | { | |
2177 | xfs_icsb_cnts_t cnt; | |
8d280b98 | 2178 | |
8d280b98 DC |
2179 | xfs_icsb_count(mp, &cnt, flags); |
2180 | ||
8d280b98 DC |
2181 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT)) |
2182 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
2183 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE)) | |
2184 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
2185 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS)) | |
2186 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
8d280b98 DC |
2187 | } |
2188 | ||
2189 | /* | |
2190 | * Accurate update of per-cpu counters to incore superblock | |
2191 | */ | |
d4d90b57 | 2192 | void |
8d280b98 | 2193 | xfs_icsb_sync_counters( |
d4d90b57 CH |
2194 | xfs_mount_t *mp, |
2195 | int flags) | |
8d280b98 | 2196 | { |
d4d90b57 CH |
2197 | spin_lock(&mp->m_sb_lock); |
2198 | xfs_icsb_sync_counters_locked(mp, flags); | |
2199 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
2200 | } |
2201 | ||
2202 | /* | |
2203 | * Balance and enable/disable counters as necessary. | |
2204 | * | |
20b64285 DC |
2205 | * Thresholds for re-enabling counters are somewhat magic. inode counts are |
2206 | * chosen to be the same number as single on disk allocation chunk per CPU, and | |
2207 | * free blocks is something far enough zero that we aren't going thrash when we | |
2208 | * get near ENOSPC. We also need to supply a minimum we require per cpu to | |
2209 | * prevent looping endlessly when xfs_alloc_space asks for more than will | |
2210 | * be distributed to a single CPU but each CPU has enough blocks to be | |
2211 | * reenabled. | |
2212 | * | |
2213 | * Note that we can be called when counters are already disabled. | |
2214 | * xfs_icsb_disable_counter() optimises the counter locking in this case to | |
2215 | * prevent locking every per-cpu counter needlessly. | |
8d280b98 | 2216 | */ |
20b64285 DC |
2217 | |
2218 | #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64 | |
4be536de | 2219 | #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \ |
20b64285 | 2220 | (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp)) |
8d280b98 | 2221 | STATIC void |
45af6c6d | 2222 | xfs_icsb_balance_counter_locked( |
8d280b98 DC |
2223 | xfs_mount_t *mp, |
2224 | xfs_sb_field_t field, | |
20b64285 | 2225 | int min_per_cpu) |
8d280b98 | 2226 | { |
6fdf8ccc | 2227 | uint64_t count, resid; |
8d280b98 | 2228 | int weight = num_online_cpus(); |
20b64285 | 2229 | uint64_t min = (uint64_t)min_per_cpu; |
8d280b98 | 2230 | |
8d280b98 DC |
2231 | /* disable counter and sync counter */ |
2232 | xfs_icsb_disable_counter(mp, field); | |
2233 | ||
2234 | /* update counters - first CPU gets residual*/ | |
2235 | switch (field) { | |
2236 | case XFS_SBS_ICOUNT: | |
2237 | count = mp->m_sb.sb_icount; | |
2238 | resid = do_div(count, weight); | |
20b64285 | 2239 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 2240 | return; |
8d280b98 DC |
2241 | break; |
2242 | case XFS_SBS_IFREE: | |
2243 | count = mp->m_sb.sb_ifree; | |
2244 | resid = do_div(count, weight); | |
20b64285 | 2245 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 2246 | return; |
8d280b98 DC |
2247 | break; |
2248 | case XFS_SBS_FDBLOCKS: | |
2249 | count = mp->m_sb.sb_fdblocks; | |
2250 | resid = do_div(count, weight); | |
20b64285 | 2251 | if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp))) |
45af6c6d | 2252 | return; |
8d280b98 DC |
2253 | break; |
2254 | default: | |
2255 | BUG(); | |
6fdf8ccc | 2256 | count = resid = 0; /* quiet, gcc */ |
8d280b98 DC |
2257 | break; |
2258 | } | |
2259 | ||
2260 | xfs_icsb_enable_counter(mp, field, count, resid); | |
45af6c6d CH |
2261 | } |
2262 | ||
2263 | STATIC void | |
2264 | xfs_icsb_balance_counter( | |
2265 | xfs_mount_t *mp, | |
2266 | xfs_sb_field_t fields, | |
2267 | int min_per_cpu) | |
2268 | { | |
2269 | spin_lock(&mp->m_sb_lock); | |
2270 | xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu); | |
2271 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
2272 | } |
2273 | ||
a8272ce0 | 2274 | STATIC int |
20b64285 | 2275 | xfs_icsb_modify_counters( |
8d280b98 DC |
2276 | xfs_mount_t *mp, |
2277 | xfs_sb_field_t field, | |
20f4ebf2 | 2278 | int64_t delta, |
20b64285 | 2279 | int rsvd) |
8d280b98 DC |
2280 | { |
2281 | xfs_icsb_cnts_t *icsbp; | |
2282 | long long lcounter; /* long counter for 64 bit fields */ | |
007c61c6 | 2283 | int cpu, ret = 0; |
8d280b98 | 2284 | |
20b64285 | 2285 | might_sleep(); |
8d280b98 DC |
2286 | again: |
2287 | cpu = get_cpu(); | |
20b64285 DC |
2288 | icsbp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, cpu); |
2289 | ||
2290 | /* | |
2291 | * if the counter is disabled, go to slow path | |
2292 | */ | |
8d280b98 DC |
2293 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) |
2294 | goto slow_path; | |
20b64285 DC |
2295 | xfs_icsb_lock_cntr(icsbp); |
2296 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) { | |
2297 | xfs_icsb_unlock_cntr(icsbp); | |
2298 | goto slow_path; | |
2299 | } | |
8d280b98 DC |
2300 | |
2301 | switch (field) { | |
2302 | case XFS_SBS_ICOUNT: | |
2303 | lcounter = icsbp->icsb_icount; | |
2304 | lcounter += delta; | |
2305 | if (unlikely(lcounter < 0)) | |
20b64285 | 2306 | goto balance_counter; |
8d280b98 DC |
2307 | icsbp->icsb_icount = lcounter; |
2308 | break; | |
2309 | ||
2310 | case XFS_SBS_IFREE: | |
2311 | lcounter = icsbp->icsb_ifree; | |
2312 | lcounter += delta; | |
2313 | if (unlikely(lcounter < 0)) | |
20b64285 | 2314 | goto balance_counter; |
8d280b98 DC |
2315 | icsbp->icsb_ifree = lcounter; |
2316 | break; | |
2317 | ||
2318 | case XFS_SBS_FDBLOCKS: | |
2319 | BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0); | |
2320 | ||
4be536de | 2321 | lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
2322 | lcounter += delta; |
2323 | if (unlikely(lcounter < 0)) | |
20b64285 | 2324 | goto balance_counter; |
4be536de | 2325 | icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
2326 | break; |
2327 | default: | |
2328 | BUG(); | |
2329 | break; | |
2330 | } | |
01e1b69c | 2331 | xfs_icsb_unlock_cntr(icsbp); |
8d280b98 | 2332 | put_cpu(); |
8d280b98 DC |
2333 | return 0; |
2334 | ||
8d280b98 | 2335 | slow_path: |
8d280b98 DC |
2336 | put_cpu(); |
2337 | ||
20b64285 DC |
2338 | /* |
2339 | * serialise with a mutex so we don't burn lots of cpu on | |
2340 | * the superblock lock. We still need to hold the superblock | |
2341 | * lock, however, when we modify the global structures. | |
2342 | */ | |
03135cf7 | 2343 | xfs_icsb_lock(mp); |
20b64285 DC |
2344 | |
2345 | /* | |
2346 | * Now running atomically. | |
2347 | * | |
2348 | * If the counter is enabled, someone has beaten us to rebalancing. | |
2349 | * Drop the lock and try again in the fast path.... | |
2350 | */ | |
2351 | if (!(xfs_icsb_counter_disabled(mp, field))) { | |
03135cf7 | 2352 | xfs_icsb_unlock(mp); |
8d280b98 | 2353 | goto again; |
8d280b98 DC |
2354 | } |
2355 | ||
20b64285 DC |
2356 | /* |
2357 | * The counter is currently disabled. Because we are | |
2358 | * running atomically here, we know a rebalance cannot | |
2359 | * be in progress. Hence we can go straight to operating | |
2360 | * on the global superblock. We do not call xfs_mod_incore_sb() | |
3685c2a1 | 2361 | * here even though we need to get the m_sb_lock. Doing so |
20b64285 | 2362 | * will cause us to re-enter this function and deadlock. |
3685c2a1 | 2363 | * Hence we get the m_sb_lock ourselves and then call |
20b64285 DC |
2364 | * xfs_mod_incore_sb_unlocked() as the unlocked path operates |
2365 | * directly on the global counters. | |
2366 | */ | |
3685c2a1 | 2367 | spin_lock(&mp->m_sb_lock); |
8d280b98 | 2368 | ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); |
3685c2a1 | 2369 | spin_unlock(&mp->m_sb_lock); |
8d280b98 | 2370 | |
20b64285 DC |
2371 | /* |
2372 | * Now that we've modified the global superblock, we | |
2373 | * may be able to re-enable the distributed counters | |
2374 | * (e.g. lots of space just got freed). After that | |
2375 | * we are done. | |
2376 | */ | |
2377 | if (ret != ENOSPC) | |
45af6c6d | 2378 | xfs_icsb_balance_counter(mp, field, 0); |
03135cf7 | 2379 | xfs_icsb_unlock(mp); |
8d280b98 | 2380 | return ret; |
8d280b98 | 2381 | |
20b64285 DC |
2382 | balance_counter: |
2383 | xfs_icsb_unlock_cntr(icsbp); | |
2384 | put_cpu(); | |
8d280b98 | 2385 | |
20b64285 DC |
2386 | /* |
2387 | * We may have multiple threads here if multiple per-cpu | |
2388 | * counters run dry at the same time. This will mean we can | |
2389 | * do more balances than strictly necessary but it is not | |
2390 | * the common slowpath case. | |
2391 | */ | |
03135cf7 | 2392 | xfs_icsb_lock(mp); |
20b64285 DC |
2393 | |
2394 | /* | |
2395 | * running atomically. | |
2396 | * | |
2397 | * This will leave the counter in the correct state for future | |
2398 | * accesses. After the rebalance, we simply try again and our retry | |
2399 | * will either succeed through the fast path or slow path without | |
2400 | * another balance operation being required. | |
2401 | */ | |
45af6c6d | 2402 | xfs_icsb_balance_counter(mp, field, delta); |
03135cf7 | 2403 | xfs_icsb_unlock(mp); |
20b64285 | 2404 | goto again; |
8d280b98 | 2405 | } |
20b64285 | 2406 | |
8d280b98 | 2407 | #endif |