]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
a805bad5 | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
7b718769 | 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_bit.h" |
1da177e4 LT |
20 | #include "xfs_log.h" |
21 | #include "xfs_clnt.h" | |
a844f451 | 22 | #include "xfs_inum.h" |
1da177e4 LT |
23 | #include "xfs_trans.h" |
24 | #include "xfs_sb.h" | |
a844f451 | 25 | #include "xfs_ag.h" |
1da177e4 LT |
26 | #include "xfs_dir2.h" |
27 | #include "xfs_alloc.h" | |
28 | #include "xfs_dmapi.h" | |
29 | #include "xfs_quota.h" | |
30 | #include "xfs_mount.h" | |
1da177e4 | 31 | #include "xfs_bmap_btree.h" |
a844f451 | 32 | #include "xfs_alloc_btree.h" |
1da177e4 | 33 | #include "xfs_ialloc_btree.h" |
1da177e4 | 34 | #include "xfs_dir2_sf.h" |
a844f451 | 35 | #include "xfs_attr_sf.h" |
1da177e4 LT |
36 | #include "xfs_dinode.h" |
37 | #include "xfs_inode.h" | |
a844f451 NS |
38 | #include "xfs_btree.h" |
39 | #include "xfs_ialloc.h" | |
1da177e4 | 40 | #include "xfs_bmap.h" |
1da177e4 LT |
41 | #include "xfs_rtalloc.h" |
42 | #include "xfs_error.h" | |
43 | #include "xfs_itable.h" | |
44 | #include "xfs_rw.h" | |
45 | #include "xfs_acl.h" | |
1da177e4 LT |
46 | #include "xfs_attr.h" |
47 | #include "xfs_buf_item.h" | |
48 | #include "xfs_utils.h" | |
49 | #include "xfs_version.h" | |
1da177e4 LT |
50 | |
51 | #include <linux/namei.h> | |
52 | #include <linux/init.h> | |
53 | #include <linux/mount.h> | |
0829c360 | 54 | #include <linux/mempool.h> |
1da177e4 | 55 | #include <linux/writeback.h> |
4df08c52 | 56 | #include <linux/kthread.h> |
7dfb7103 | 57 | #include <linux/freezer.h> |
1da177e4 | 58 | |
7989cb8e DC |
59 | static struct quotactl_ops xfs_quotactl_operations; |
60 | static struct super_operations xfs_super_operations; | |
61 | static kmem_zone_t *xfs_vnode_zone; | |
62 | static kmem_zone_t *xfs_ioend_zone; | |
0829c360 | 63 | mempool_t *xfs_ioend_pool; |
1da177e4 LT |
64 | |
65 | STATIC struct xfs_mount_args * | |
66 | xfs_args_allocate( | |
764d1f89 NS |
67 | struct super_block *sb, |
68 | int silent) | |
1da177e4 LT |
69 | { |
70 | struct xfs_mount_args *args; | |
71 | ||
72 | args = kmem_zalloc(sizeof(struct xfs_mount_args), KM_SLEEP); | |
73 | args->logbufs = args->logbufsize = -1; | |
74 | strncpy(args->fsname, sb->s_id, MAXNAMELEN); | |
75 | ||
76 | /* Copy the already-parsed mount(2) flags we're interested in */ | |
1da177e4 LT |
77 | if (sb->s_flags & MS_DIRSYNC) |
78 | args->flags |= XFSMNT_DIRSYNC; | |
79 | if (sb->s_flags & MS_SYNCHRONOUS) | |
80 | args->flags |= XFSMNT_WSYNC; | |
764d1f89 NS |
81 | if (silent) |
82 | args->flags |= XFSMNT_QUIET; | |
1da177e4 LT |
83 | args->flags |= XFSMNT_32BITINODES; |
84 | ||
85 | return args; | |
86 | } | |
87 | ||
88 | __uint64_t | |
89 | xfs_max_file_offset( | |
90 | unsigned int blockshift) | |
91 | { | |
92 | unsigned int pagefactor = 1; | |
93 | unsigned int bitshift = BITS_PER_LONG - 1; | |
94 | ||
95 | /* Figure out maximum filesize, on Linux this can depend on | |
96 | * the filesystem blocksize (on 32 bit platforms). | |
97 | * __block_prepare_write does this in an [unsigned] long... | |
98 | * page->index << (PAGE_CACHE_SHIFT - bbits) | |
99 | * So, for page sized blocks (4K on 32 bit platforms), | |
100 | * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is | |
101 | * (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1) | |
102 | * but for smaller blocksizes it is less (bbits = log2 bsize). | |
103 | * Note1: get_block_t takes a long (implicit cast from above) | |
104 | * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch | |
105 | * can optionally convert the [unsigned] long from above into | |
106 | * an [unsigned] long long. | |
107 | */ | |
108 | ||
109 | #if BITS_PER_LONG == 32 | |
110 | # if defined(CONFIG_LBD) | |
111 | ASSERT(sizeof(sector_t) == 8); | |
112 | pagefactor = PAGE_CACHE_SIZE; | |
113 | bitshift = BITS_PER_LONG; | |
114 | # else | |
115 | pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift); | |
116 | # endif | |
117 | #endif | |
118 | ||
119 | return (((__uint64_t)pagefactor) << bitshift) - 1; | |
120 | } | |
121 | ||
7989cb8e | 122 | STATIC_INLINE void |
1da177e4 LT |
123 | xfs_set_inodeops( |
124 | struct inode *inode) | |
125 | { | |
0432dab2 CH |
126 | switch (inode->i_mode & S_IFMT) { |
127 | case S_IFREG: | |
416c6d5b | 128 | inode->i_op = &xfs_inode_operations; |
3562fd45 | 129 | inode->i_fop = &xfs_file_operations; |
e4c573bb | 130 | inode->i_mapping->a_ops = &xfs_address_space_operations; |
0432dab2 CH |
131 | break; |
132 | case S_IFDIR: | |
416c6d5b | 133 | inode->i_op = &xfs_dir_inode_operations; |
3562fd45 | 134 | inode->i_fop = &xfs_dir_file_operations; |
0432dab2 CH |
135 | break; |
136 | case S_IFLNK: | |
416c6d5b | 137 | inode->i_op = &xfs_symlink_inode_operations; |
1da177e4 | 138 | if (inode->i_blocks) |
e4c573bb | 139 | inode->i_mapping->a_ops = &xfs_address_space_operations; |
0432dab2 CH |
140 | break; |
141 | default: | |
416c6d5b | 142 | inode->i_op = &xfs_inode_operations; |
1da177e4 | 143 | init_special_inode(inode, inode->i_mode, inode->i_rdev); |
0432dab2 | 144 | break; |
1da177e4 LT |
145 | } |
146 | } | |
147 | ||
7989cb8e | 148 | STATIC_INLINE void |
1da177e4 LT |
149 | xfs_revalidate_inode( |
150 | xfs_mount_t *mp, | |
67fcaa73 | 151 | bhv_vnode_t *vp, |
1da177e4 LT |
152 | xfs_inode_t *ip) |
153 | { | |
ec86dc02 | 154 | struct inode *inode = vn_to_inode(vp); |
1da177e4 | 155 | |
0432dab2 | 156 | inode->i_mode = ip->i_d.di_mode; |
1da177e4 LT |
157 | inode->i_nlink = ip->i_d.di_nlink; |
158 | inode->i_uid = ip->i_d.di_uid; | |
159 | inode->i_gid = ip->i_d.di_gid; | |
0432dab2 CH |
160 | |
161 | switch (inode->i_mode & S_IFMT) { | |
162 | case S_IFBLK: | |
163 | case S_IFCHR: | |
164 | inode->i_rdev = | |
165 | MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff, | |
166 | sysv_minor(ip->i_df.if_u2.if_rdev)); | |
167 | break; | |
168 | default: | |
1da177e4 | 169 | inode->i_rdev = 0; |
0432dab2 | 170 | break; |
1da177e4 | 171 | } |
0432dab2 | 172 | |
1da177e4 LT |
173 | inode->i_generation = ip->i_d.di_gen; |
174 | i_size_write(inode, ip->i_d.di_size); | |
175 | inode->i_blocks = | |
176 | XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks); | |
177 | inode->i_atime.tv_sec = ip->i_d.di_atime.t_sec; | |
178 | inode->i_atime.tv_nsec = ip->i_d.di_atime.t_nsec; | |
179 | inode->i_mtime.tv_sec = ip->i_d.di_mtime.t_sec; | |
180 | inode->i_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec; | |
181 | inode->i_ctime.tv_sec = ip->i_d.di_ctime.t_sec; | |
182 | inode->i_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec; | |
183 | if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE) | |
184 | inode->i_flags |= S_IMMUTABLE; | |
185 | else | |
186 | inode->i_flags &= ~S_IMMUTABLE; | |
187 | if (ip->i_d.di_flags & XFS_DIFLAG_APPEND) | |
188 | inode->i_flags |= S_APPEND; | |
189 | else | |
190 | inode->i_flags &= ~S_APPEND; | |
191 | if (ip->i_d.di_flags & XFS_DIFLAG_SYNC) | |
192 | inode->i_flags |= S_SYNC; | |
193 | else | |
194 | inode->i_flags &= ~S_SYNC; | |
195 | if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME) | |
196 | inode->i_flags |= S_NOATIME; | |
197 | else | |
198 | inode->i_flags &= ~S_NOATIME; | |
199 | vp->v_flag &= ~VMODIFIED; | |
200 | } | |
201 | ||
202 | void | |
203 | xfs_initialize_vnode( | |
204 | bhv_desc_t *bdp, | |
67fcaa73 | 205 | bhv_vnode_t *vp, |
1da177e4 LT |
206 | bhv_desc_t *inode_bhv, |
207 | int unlock) | |
208 | { | |
209 | xfs_inode_t *ip = XFS_BHVTOI(inode_bhv); | |
ec86dc02 | 210 | struct inode *inode = vn_to_inode(vp); |
1da177e4 LT |
211 | |
212 | if (!inode_bhv->bd_vobj) { | |
213 | vp->v_vfsp = bhvtovfs(bdp); | |
214 | bhv_desc_init(inode_bhv, ip, vp, &xfs_vnodeops); | |
215 | bhv_insert(VN_BHV_HEAD(vp), inode_bhv); | |
216 | } | |
217 | ||
218 | /* | |
219 | * We need to set the ops vectors, and unlock the inode, but if | |
220 | * we have been called during the new inode create process, it is | |
221 | * too early to fill in the Linux inode. We will get called a | |
222 | * second time once the inode is properly set up, and then we can | |
223 | * finish our work. | |
224 | */ | |
225 | if (ip->i_d.di_mode != 0 && unlock && (inode->i_state & I_NEW)) { | |
1da177e4 LT |
226 | xfs_revalidate_inode(XFS_BHVTOM(bdp), vp, ip); |
227 | xfs_set_inodeops(inode); | |
ec86dc02 | 228 | |
7a18c386 | 229 | xfs_iflags_clear(ip, XFS_INEW); |
1da177e4 LT |
230 | barrier(); |
231 | ||
232 | unlock_new_inode(inode); | |
233 | } | |
234 | } | |
235 | ||
236 | int | |
237 | xfs_blkdev_get( | |
238 | xfs_mount_t *mp, | |
239 | const char *name, | |
240 | struct block_device **bdevp) | |
241 | { | |
242 | int error = 0; | |
243 | ||
244 | *bdevp = open_bdev_excl(name, 0, mp); | |
245 | if (IS_ERR(*bdevp)) { | |
246 | error = PTR_ERR(*bdevp); | |
247 | printk("XFS: Invalid device [%s], error=%d\n", name, error); | |
248 | } | |
249 | ||
250 | return -error; | |
251 | } | |
252 | ||
253 | void | |
254 | xfs_blkdev_put( | |
255 | struct block_device *bdev) | |
256 | { | |
257 | if (bdev) | |
258 | close_bdev_excl(bdev); | |
259 | } | |
260 | ||
f538d4da CH |
261 | /* |
262 | * Try to write out the superblock using barriers. | |
263 | */ | |
264 | STATIC int | |
265 | xfs_barrier_test( | |
266 | xfs_mount_t *mp) | |
267 | { | |
268 | xfs_buf_t *sbp = xfs_getsb(mp, 0); | |
269 | int error; | |
270 | ||
271 | XFS_BUF_UNDONE(sbp); | |
272 | XFS_BUF_UNREAD(sbp); | |
273 | XFS_BUF_UNDELAYWRITE(sbp); | |
274 | XFS_BUF_WRITE(sbp); | |
275 | XFS_BUF_UNASYNC(sbp); | |
276 | XFS_BUF_ORDERED(sbp); | |
277 | ||
278 | xfsbdstrat(mp, sbp); | |
279 | error = xfs_iowait(sbp); | |
280 | ||
281 | /* | |
282 | * Clear all the flags we set and possible error state in the | |
283 | * buffer. We only did the write to try out whether barriers | |
284 | * worked and shouldn't leave any traces in the superblock | |
285 | * buffer. | |
286 | */ | |
287 | XFS_BUF_DONE(sbp); | |
288 | XFS_BUF_ERROR(sbp, 0); | |
289 | XFS_BUF_UNORDERED(sbp); | |
290 | ||
291 | xfs_buf_relse(sbp); | |
292 | return error; | |
293 | } | |
294 | ||
295 | void | |
296 | xfs_mountfs_check_barriers(xfs_mount_t *mp) | |
297 | { | |
298 | int error; | |
299 | ||
300 | if (mp->m_logdev_targp != mp->m_ddev_targp) { | |
301 | xfs_fs_cmn_err(CE_NOTE, mp, | |
302 | "Disabling barriers, not supported with external log device"); | |
303 | mp->m_flags &= ~XFS_MOUNT_BARRIER; | |
4ef19ddd | 304 | return; |
f538d4da CH |
305 | } |
306 | ||
ce8e922c | 307 | if (mp->m_ddev_targp->bt_bdev->bd_disk->queue->ordered == |
f538d4da CH |
308 | QUEUE_ORDERED_NONE) { |
309 | xfs_fs_cmn_err(CE_NOTE, mp, | |
310 | "Disabling barriers, not supported by the underlying device"); | |
311 | mp->m_flags &= ~XFS_MOUNT_BARRIER; | |
4ef19ddd | 312 | return; |
f538d4da CH |
313 | } |
314 | ||
b2ea401b NS |
315 | if (xfs_readonly_buftarg(mp->m_ddev_targp)) { |
316 | xfs_fs_cmn_err(CE_NOTE, mp, | |
317 | "Disabling barriers, underlying device is readonly"); | |
318 | mp->m_flags &= ~XFS_MOUNT_BARRIER; | |
319 | return; | |
320 | } | |
321 | ||
f538d4da CH |
322 | error = xfs_barrier_test(mp); |
323 | if (error) { | |
324 | xfs_fs_cmn_err(CE_NOTE, mp, | |
325 | "Disabling barriers, trial barrier write failed"); | |
326 | mp->m_flags &= ~XFS_MOUNT_BARRIER; | |
4ef19ddd | 327 | return; |
f538d4da CH |
328 | } |
329 | } | |
330 | ||
331 | void | |
332 | xfs_blkdev_issue_flush( | |
333 | xfs_buftarg_t *buftarg) | |
334 | { | |
ce8e922c | 335 | blkdev_issue_flush(buftarg->bt_bdev, NULL); |
f538d4da | 336 | } |
1da177e4 LT |
337 | |
338 | STATIC struct inode * | |
a50cd269 | 339 | xfs_fs_alloc_inode( |
1da177e4 LT |
340 | struct super_block *sb) |
341 | { | |
67fcaa73 | 342 | bhv_vnode_t *vp; |
1da177e4 | 343 | |
8758280f NS |
344 | vp = kmem_zone_alloc(xfs_vnode_zone, KM_SLEEP); |
345 | if (unlikely(!vp)) | |
1da177e4 | 346 | return NULL; |
ec86dc02 | 347 | return vn_to_inode(vp); |
1da177e4 LT |
348 | } |
349 | ||
350 | STATIC void | |
a50cd269 | 351 | xfs_fs_destroy_inode( |
1da177e4 LT |
352 | struct inode *inode) |
353 | { | |
ec86dc02 | 354 | kmem_zone_free(xfs_vnode_zone, vn_from_inode(inode)); |
1da177e4 LT |
355 | } |
356 | ||
357 | STATIC void | |
a50cd269 | 358 | xfs_fs_inode_init_once( |
8758280f NS |
359 | void *vnode, |
360 | kmem_zone_t *zonep, | |
1da177e4 LT |
361 | unsigned long flags) |
362 | { | |
50953fe9 | 363 | if (flags & SLAB_CTOR_CONSTRUCTOR) |
67fcaa73 | 364 | inode_init_once(vn_to_inode((bhv_vnode_t *)vnode)); |
1da177e4 LT |
365 | } |
366 | ||
367 | STATIC int | |
8758280f | 368 | xfs_init_zones(void) |
1da177e4 | 369 | { |
67fcaa73 | 370 | xfs_vnode_zone = kmem_zone_init_flags(sizeof(bhv_vnode_t), "xfs_vnode", |
e0cc2325 NS |
371 | KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | |
372 | KM_ZONE_SPREAD, | |
a50cd269 | 373 | xfs_fs_inode_init_once); |
0829c360 CH |
374 | if (!xfs_vnode_zone) |
375 | goto out; | |
376 | ||
377 | xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend"); | |
378 | if (!xfs_ioend_zone) | |
379 | goto out_destroy_vnode_zone; | |
380 | ||
93d2341c MD |
381 | xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE, |
382 | xfs_ioend_zone); | |
0829c360 CH |
383 | if (!xfs_ioend_pool) |
384 | goto out_free_ioend_zone; | |
1da177e4 | 385 | return 0; |
0829c360 | 386 | |
0829c360 CH |
387 | out_free_ioend_zone: |
388 | kmem_zone_destroy(xfs_ioend_zone); | |
389 | out_destroy_vnode_zone: | |
390 | kmem_zone_destroy(xfs_vnode_zone); | |
391 | out: | |
392 | return -ENOMEM; | |
1da177e4 LT |
393 | } |
394 | ||
395 | STATIC void | |
8758280f | 396 | xfs_destroy_zones(void) |
1da177e4 | 397 | { |
0829c360 CH |
398 | mempool_destroy(xfs_ioend_pool); |
399 | kmem_zone_destroy(xfs_vnode_zone); | |
400 | kmem_zone_destroy(xfs_ioend_zone); | |
1da177e4 LT |
401 | } |
402 | ||
403 | /* | |
404 | * Attempt to flush the inode, this will actually fail | |
405 | * if the inode is pinned, but we dirty the inode again | |
406 | * at the point when it is unpinned after a log write, | |
8758280f | 407 | * since this is when the inode itself becomes flushable. |
1da177e4 LT |
408 | */ |
409 | STATIC int | |
a50cd269 | 410 | xfs_fs_write_inode( |
1da177e4 LT |
411 | struct inode *inode, |
412 | int sync) | |
413 | { | |
67fcaa73 | 414 | bhv_vnode_t *vp = vn_from_inode(inode); |
1da177e4 LT |
415 | int error = 0, flags = FLUSH_INODE; |
416 | ||
417 | if (vp) { | |
418 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); | |
419 | if (sync) | |
420 | flags |= FLUSH_SYNC; | |
67fcaa73 NS |
421 | error = bhv_vop_iflush(vp, flags); |
422 | if (error == EAGAIN) | |
423 | error = sync? bhv_vop_iflush(vp, flags | FLUSH_LOG) : 0; | |
1da177e4 | 424 | } |
1da177e4 LT |
425 | return -error; |
426 | } | |
427 | ||
428 | STATIC void | |
a50cd269 | 429 | xfs_fs_clear_inode( |
1da177e4 LT |
430 | struct inode *inode) |
431 | { | |
67fcaa73 | 432 | bhv_vnode_t *vp = vn_from_inode(inode); |
1da177e4 | 433 | |
a50cd269 | 434 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); |
56d433e4 | 435 | |
56d433e4 CH |
436 | XFS_STATS_INC(vn_rele); |
437 | XFS_STATS_INC(vn_remove); | |
438 | XFS_STATS_INC(vn_reclaim); | |
439 | XFS_STATS_DEC(vn_active); | |
440 | ||
02ba71de CH |
441 | /* |
442 | * This can happen because xfs_iget_core calls xfs_idestroy if we | |
443 | * find an inode with di_mode == 0 but without IGET_CREATE set. | |
444 | */ | |
67fcaa73 NS |
445 | if (VNHEAD(vp)) |
446 | bhv_vop_inactive(vp, NULL); | |
1da177e4 | 447 | |
56d433e4 CH |
448 | VN_LOCK(vp); |
449 | vp->v_flag &= ~VMODIFIED; | |
450 | VN_UNLOCK(vp, 0); | |
451 | ||
67fcaa73 NS |
452 | if (VNHEAD(vp)) |
453 | if (bhv_vop_reclaim(vp)) | |
454 | panic("%s: cannot reclaim 0x%p\n", __FUNCTION__, vp); | |
56d433e4 | 455 | |
67fcaa73 | 456 | ASSERT(VNHEAD(vp) == NULL); |
56d433e4 CH |
457 | |
458 | #ifdef XFS_VNODE_TRACE | |
459 | ktrace_free(vp->v_trace); | |
460 | #endif | |
461 | } | |
1da177e4 LT |
462 | |
463 | /* | |
464 | * Enqueue a work item to be picked up by the vfs xfssyncd thread. | |
465 | * Doing this has two advantages: | |
466 | * - It saves on stack space, which is tight in certain situations | |
467 | * - It can be used (with care) as a mechanism to avoid deadlocks. | |
468 | * Flushing while allocating in a full filesystem requires both. | |
469 | */ | |
470 | STATIC void | |
471 | xfs_syncd_queue_work( | |
b83bd138 | 472 | struct bhv_vfs *vfs, |
1da177e4 | 473 | void *data, |
b83bd138 | 474 | void (*syncer)(bhv_vfs_t *, void *)) |
1da177e4 | 475 | { |
b83bd138 | 476 | struct bhv_vfs_sync_work *work; |
1da177e4 | 477 | |
b83bd138 | 478 | work = kmem_alloc(sizeof(struct bhv_vfs_sync_work), KM_SLEEP); |
1da177e4 LT |
479 | INIT_LIST_HEAD(&work->w_list); |
480 | work->w_syncer = syncer; | |
481 | work->w_data = data; | |
482 | work->w_vfs = vfs; | |
483 | spin_lock(&vfs->vfs_sync_lock); | |
484 | list_add_tail(&work->w_list, &vfs->vfs_sync_list); | |
485 | spin_unlock(&vfs->vfs_sync_lock); | |
486 | wake_up_process(vfs->vfs_sync_task); | |
487 | } | |
488 | ||
489 | /* | |
490 | * Flush delayed allocate data, attempting to free up reserved space | |
491 | * from existing allocations. At this point a new allocation attempt | |
492 | * has failed with ENOSPC and we are in the process of scratching our | |
493 | * heads, looking about for more room... | |
494 | */ | |
495 | STATIC void | |
496 | xfs_flush_inode_work( | |
b83bd138 | 497 | bhv_vfs_t *vfs, |
1da177e4 LT |
498 | void *inode) |
499 | { | |
500 | filemap_flush(((struct inode *)inode)->i_mapping); | |
501 | iput((struct inode *)inode); | |
502 | } | |
503 | ||
504 | void | |
505 | xfs_flush_inode( | |
506 | xfs_inode_t *ip) | |
507 | { | |
ec86dc02 | 508 | struct inode *inode = vn_to_inode(XFS_ITOV(ip)); |
b83bd138 | 509 | struct bhv_vfs *vfs = XFS_MTOVFS(ip->i_mount); |
1da177e4 LT |
510 | |
511 | igrab(inode); | |
512 | xfs_syncd_queue_work(vfs, inode, xfs_flush_inode_work); | |
041e0e3b | 513 | delay(msecs_to_jiffies(500)); |
1da177e4 LT |
514 | } |
515 | ||
516 | /* | |
517 | * This is the "bigger hammer" version of xfs_flush_inode_work... | |
518 | * (IOW, "If at first you don't succeed, use a Bigger Hammer"). | |
519 | */ | |
520 | STATIC void | |
521 | xfs_flush_device_work( | |
b83bd138 | 522 | bhv_vfs_t *vfs, |
1da177e4 LT |
523 | void *inode) |
524 | { | |
525 | sync_blockdev(vfs->vfs_super->s_bdev); | |
526 | iput((struct inode *)inode); | |
527 | } | |
528 | ||
529 | void | |
530 | xfs_flush_device( | |
531 | xfs_inode_t *ip) | |
532 | { | |
ec86dc02 | 533 | struct inode *inode = vn_to_inode(XFS_ITOV(ip)); |
b83bd138 | 534 | struct bhv_vfs *vfs = XFS_MTOVFS(ip->i_mount); |
1da177e4 LT |
535 | |
536 | igrab(inode); | |
537 | xfs_syncd_queue_work(vfs, inode, xfs_flush_device_work); | |
041e0e3b | 538 | delay(msecs_to_jiffies(500)); |
1da177e4 LT |
539 | xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC); |
540 | } | |
541 | ||
1da177e4 LT |
542 | STATIC void |
543 | vfs_sync_worker( | |
b83bd138 | 544 | bhv_vfs_t *vfsp, |
1da177e4 LT |
545 | void *unused) |
546 | { | |
547 | int error; | |
548 | ||
549 | if (!(vfsp->vfs_flag & VFS_RDONLY)) | |
b83bd138 NS |
550 | error = bhv_vfs_sync(vfsp, SYNC_FSDATA | SYNC_BDFLUSH | \ |
551 | SYNC_ATTR | SYNC_REFCACHE, NULL); | |
1da177e4 | 552 | vfsp->vfs_sync_seq++; |
1da177e4 LT |
553 | wake_up(&vfsp->vfs_wait_single_sync_task); |
554 | } | |
555 | ||
556 | STATIC int | |
557 | xfssyncd( | |
558 | void *arg) | |
559 | { | |
560 | long timeleft; | |
b83bd138 NS |
561 | bhv_vfs_t *vfsp = (bhv_vfs_t *) arg; |
562 | bhv_vfs_sync_work_t *work, *n; | |
4df08c52 | 563 | LIST_HEAD (tmp); |
1da177e4 | 564 | |
041e0e3b | 565 | timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10); |
1da177e4 | 566 | for (;;) { |
041e0e3b | 567 | timeleft = schedule_timeout_interruptible(timeleft); |
1da177e4 | 568 | /* swsusp */ |
3e1d1d28 | 569 | try_to_freeze(); |
71df099d | 570 | if (kthread_should_stop() && list_empty(&vfsp->vfs_sync_list)) |
1da177e4 LT |
571 | break; |
572 | ||
573 | spin_lock(&vfsp->vfs_sync_lock); | |
574 | /* | |
575 | * We can get woken by laptop mode, to do a sync - | |
576 | * that's the (only!) case where the list would be | |
577 | * empty with time remaining. | |
578 | */ | |
579 | if (!timeleft || list_empty(&vfsp->vfs_sync_list)) { | |
580 | if (!timeleft) | |
041e0e3b NA |
581 | timeleft = xfs_syncd_centisecs * |
582 | msecs_to_jiffies(10); | |
1da177e4 LT |
583 | INIT_LIST_HEAD(&vfsp->vfs_sync_work.w_list); |
584 | list_add_tail(&vfsp->vfs_sync_work.w_list, | |
585 | &vfsp->vfs_sync_list); | |
586 | } | |
587 | list_for_each_entry_safe(work, n, &vfsp->vfs_sync_list, w_list) | |
588 | list_move(&work->w_list, &tmp); | |
589 | spin_unlock(&vfsp->vfs_sync_lock); | |
590 | ||
591 | list_for_each_entry_safe(work, n, &tmp, w_list) { | |
592 | (*work->w_syncer)(vfsp, work->w_data); | |
593 | list_del(&work->w_list); | |
594 | if (work == &vfsp->vfs_sync_work) | |
595 | continue; | |
b83bd138 | 596 | kmem_free(work, sizeof(struct bhv_vfs_sync_work)); |
1da177e4 LT |
597 | } |
598 | } | |
599 | ||
1da177e4 LT |
600 | return 0; |
601 | } | |
602 | ||
603 | STATIC int | |
a50cd269 | 604 | xfs_fs_start_syncd( |
b83bd138 | 605 | bhv_vfs_t *vfsp) |
1da177e4 | 606 | { |
4df08c52 CH |
607 | vfsp->vfs_sync_work.w_syncer = vfs_sync_worker; |
608 | vfsp->vfs_sync_work.w_vfs = vfsp; | |
609 | vfsp->vfs_sync_task = kthread_run(xfssyncd, vfsp, "xfssyncd"); | |
610 | if (IS_ERR(vfsp->vfs_sync_task)) | |
611 | return -PTR_ERR(vfsp->vfs_sync_task); | |
1da177e4 LT |
612 | return 0; |
613 | } | |
614 | ||
615 | STATIC void | |
a50cd269 | 616 | xfs_fs_stop_syncd( |
b83bd138 | 617 | bhv_vfs_t *vfsp) |
1da177e4 | 618 | { |
4df08c52 | 619 | kthread_stop(vfsp->vfs_sync_task); |
1da177e4 LT |
620 | } |
621 | ||
622 | STATIC void | |
a50cd269 | 623 | xfs_fs_put_super( |
1da177e4 LT |
624 | struct super_block *sb) |
625 | { | |
b83bd138 | 626 | bhv_vfs_t *vfsp = vfs_from_sb(sb); |
1da177e4 LT |
627 | int error; |
628 | ||
a50cd269 | 629 | xfs_fs_stop_syncd(vfsp); |
b83bd138 NS |
630 | bhv_vfs_sync(vfsp, SYNC_ATTR | SYNC_DELWRI, NULL); |
631 | error = bhv_vfs_unmount(vfsp, 0, NULL); | |
1da177e4 | 632 | if (error) { |
b83bd138 NS |
633 | printk("XFS: unmount got error=%d\n", error); |
634 | printk("%s: vfs=0x%p left dangling!\n", __FUNCTION__, vfsp); | |
635 | } else { | |
636 | vfs_deallocate(vfsp); | |
1da177e4 | 637 | } |
1da177e4 LT |
638 | } |
639 | ||
640 | STATIC void | |
a50cd269 | 641 | xfs_fs_write_super( |
1da177e4 LT |
642 | struct super_block *sb) |
643 | { | |
b83bd138 NS |
644 | if (!(sb->s_flags & MS_RDONLY)) |
645 | bhv_vfs_sync(vfs_from_sb(sb), SYNC_FSDATA, NULL); | |
1da177e4 LT |
646 | sb->s_dirt = 0; |
647 | } | |
648 | ||
649 | STATIC int | |
a50cd269 | 650 | xfs_fs_sync_super( |
1da177e4 LT |
651 | struct super_block *sb, |
652 | int wait) | |
653 | { | |
b83bd138 NS |
654 | bhv_vfs_t *vfsp = vfs_from_sb(sb); |
655 | int error; | |
656 | int flags; | |
1da177e4 | 657 | |
2823945f DC |
658 | if (unlikely(sb->s_frozen == SB_FREEZE_WRITE)) { |
659 | /* | |
660 | * First stage of freeze - no more writers will make progress | |
661 | * now we are here, so we flush delwri and delalloc buffers | |
662 | * here, then wait for all I/O to complete. Data is frozen at | |
663 | * that point. Metadata is not frozen, transactions can still | |
664 | * occur here so don't bother flushing the buftarg (i.e | |
665 | * SYNC_QUIESCE) because it'll just get dirty again. | |
666 | */ | |
3c0dc77b | 667 | flags = SYNC_FSDATA | SYNC_DELWRI | SYNC_WAIT | SYNC_IOWAIT; |
2823945f | 668 | } else |
f898d6c0 | 669 | flags = SYNC_FSDATA | (wait ? SYNC_WAIT : 0); |
1da177e4 | 670 | |
b83bd138 | 671 | error = bhv_vfs_sync(vfsp, flags, NULL); |
1da177e4 LT |
672 | sb->s_dirt = 0; |
673 | ||
674 | if (unlikely(laptop_mode)) { | |
675 | int prev_sync_seq = vfsp->vfs_sync_seq; | |
676 | ||
677 | /* | |
678 | * The disk must be active because we're syncing. | |
679 | * We schedule xfssyncd now (now that the disk is | |
680 | * active) instead of later (when it might not be). | |
681 | */ | |
682 | wake_up_process(vfsp->vfs_sync_task); | |
683 | /* | |
684 | * We have to wait for the sync iteration to complete. | |
685 | * If we don't, the disk activity caused by the sync | |
686 | * will come after the sync is completed, and that | |
687 | * triggers another sync from laptop mode. | |
688 | */ | |
689 | wait_event(vfsp->vfs_wait_single_sync_task, | |
690 | vfsp->vfs_sync_seq != prev_sync_seq); | |
691 | } | |
692 | ||
693 | return -error; | |
694 | } | |
695 | ||
696 | STATIC int | |
a50cd269 | 697 | xfs_fs_statfs( |
726c3342 | 698 | struct dentry *dentry, |
1da177e4 LT |
699 | struct kstatfs *statp) |
700 | { | |
d6938d1b DH |
701 | return -bhv_vfs_statvfs(vfs_from_sb(dentry->d_sb), statp, |
702 | vn_from_inode(dentry->d_inode)); | |
1da177e4 LT |
703 | } |
704 | ||
705 | STATIC int | |
a50cd269 | 706 | xfs_fs_remount( |
1da177e4 LT |
707 | struct super_block *sb, |
708 | int *flags, | |
709 | char *options) | |
710 | { | |
b83bd138 | 711 | bhv_vfs_t *vfsp = vfs_from_sb(sb); |
764d1f89 | 712 | struct xfs_mount_args *args = xfs_args_allocate(sb, 0); |
1da177e4 LT |
713 | int error; |
714 | ||
b83bd138 | 715 | error = bhv_vfs_parseargs(vfsp, options, args, 1); |
1da177e4 | 716 | if (!error) |
b83bd138 | 717 | error = bhv_vfs_mntupdate(vfsp, flags, args); |
1da177e4 LT |
718 | kmem_free(args, sizeof(*args)); |
719 | return -error; | |
720 | } | |
721 | ||
722 | STATIC void | |
a50cd269 | 723 | xfs_fs_lockfs( |
1da177e4 LT |
724 | struct super_block *sb) |
725 | { | |
b83bd138 | 726 | bhv_vfs_freeze(vfs_from_sb(sb)); |
1da177e4 LT |
727 | } |
728 | ||
729 | STATIC int | |
a50cd269 | 730 | xfs_fs_show_options( |
1da177e4 LT |
731 | struct seq_file *m, |
732 | struct vfsmount *mnt) | |
733 | { | |
b83bd138 | 734 | return -bhv_vfs_showargs(vfs_from_sb(mnt->mnt_sb), m); |
1da177e4 LT |
735 | } |
736 | ||
ee34807a | 737 | STATIC int |
a50cd269 | 738 | xfs_fs_quotasync( |
ee34807a NS |
739 | struct super_block *sb, |
740 | int type) | |
741 | { | |
b83bd138 | 742 | return -bhv_vfs_quotactl(vfs_from_sb(sb), Q_XQUOTASYNC, 0, NULL); |
ee34807a NS |
743 | } |
744 | ||
1da177e4 | 745 | STATIC int |
a50cd269 | 746 | xfs_fs_getxstate( |
1da177e4 LT |
747 | struct super_block *sb, |
748 | struct fs_quota_stat *fqs) | |
749 | { | |
b83bd138 | 750 | return -bhv_vfs_quotactl(vfs_from_sb(sb), Q_XGETQSTAT, 0, (caddr_t)fqs); |
1da177e4 LT |
751 | } |
752 | ||
753 | STATIC int | |
a50cd269 | 754 | xfs_fs_setxstate( |
1da177e4 LT |
755 | struct super_block *sb, |
756 | unsigned int flags, | |
757 | int op) | |
758 | { | |
b83bd138 | 759 | return -bhv_vfs_quotactl(vfs_from_sb(sb), op, 0, (caddr_t)&flags); |
1da177e4 LT |
760 | } |
761 | ||
762 | STATIC int | |
a50cd269 | 763 | xfs_fs_getxquota( |
1da177e4 LT |
764 | struct super_block *sb, |
765 | int type, | |
766 | qid_t id, | |
767 | struct fs_disk_quota *fdq) | |
768 | { | |
b83bd138 NS |
769 | return -bhv_vfs_quotactl(vfs_from_sb(sb), |
770 | (type == USRQUOTA) ? Q_XGETQUOTA : | |
771 | ((type == GRPQUOTA) ? Q_XGETGQUOTA : | |
772 | Q_XGETPQUOTA), id, (caddr_t)fdq); | |
1da177e4 LT |
773 | } |
774 | ||
775 | STATIC int | |
a50cd269 | 776 | xfs_fs_setxquota( |
1da177e4 LT |
777 | struct super_block *sb, |
778 | int type, | |
779 | qid_t id, | |
780 | struct fs_disk_quota *fdq) | |
781 | { | |
b83bd138 NS |
782 | return -bhv_vfs_quotactl(vfs_from_sb(sb), |
783 | (type == USRQUOTA) ? Q_XSETQLIM : | |
784 | ((type == GRPQUOTA) ? Q_XSETGQLIM : | |
785 | Q_XSETPQLIM), id, (caddr_t)fdq); | |
1da177e4 LT |
786 | } |
787 | ||
788 | STATIC int | |
a50cd269 | 789 | xfs_fs_fill_super( |
1da177e4 LT |
790 | struct super_block *sb, |
791 | void *data, | |
792 | int silent) | |
793 | { | |
67fcaa73 | 794 | struct bhv_vnode *rootvp; |
b83bd138 | 795 | struct bhv_vfs *vfsp = vfs_allocate(sb); |
764d1f89 | 796 | struct xfs_mount_args *args = xfs_args_allocate(sb, silent); |
1da177e4 | 797 | struct kstatfs statvfs; |
b83bd138 | 798 | int error; |
1da177e4 | 799 | |
1da177e4 LT |
800 | bhv_insert_all_vfsops(vfsp); |
801 | ||
b83bd138 | 802 | error = bhv_vfs_parseargs(vfsp, (char *)data, args, 0); |
1da177e4 LT |
803 | if (error) { |
804 | bhv_remove_all_vfsops(vfsp, 1); | |
805 | goto fail_vfsop; | |
806 | } | |
807 | ||
808 | sb_min_blocksize(sb, BBSIZE); | |
a50cd269 | 809 | sb->s_export_op = &xfs_export_operations; |
a50cd269 NS |
810 | sb->s_qcop = &xfs_quotactl_operations; |
811 | sb->s_op = &xfs_super_operations; | |
1da177e4 | 812 | |
b83bd138 | 813 | error = bhv_vfs_mount(vfsp, args, NULL); |
1da177e4 LT |
814 | if (error) { |
815 | bhv_remove_all_vfsops(vfsp, 1); | |
816 | goto fail_vfsop; | |
817 | } | |
818 | ||
b83bd138 | 819 | error = bhv_vfs_statvfs(vfsp, &statvfs, NULL); |
1da177e4 LT |
820 | if (error) |
821 | goto fail_unmount; | |
822 | ||
823 | sb->s_dirt = 1; | |
824 | sb->s_magic = statvfs.f_type; | |
825 | sb->s_blocksize = statvfs.f_bsize; | |
826 | sb->s_blocksize_bits = ffs(statvfs.f_bsize) - 1; | |
827 | sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits); | |
828 | sb->s_time_gran = 1; | |
829 | set_posix_acl_flag(sb); | |
830 | ||
b83bd138 | 831 | error = bhv_vfs_root(vfsp, &rootvp); |
1da177e4 LT |
832 | if (error) |
833 | goto fail_unmount; | |
834 | ||
ec86dc02 | 835 | sb->s_root = d_alloc_root(vn_to_inode(rootvp)); |
1da177e4 LT |
836 | if (!sb->s_root) { |
837 | error = ENOMEM; | |
838 | goto fail_vnrele; | |
839 | } | |
840 | if (is_bad_inode(sb->s_root->d_inode)) { | |
841 | error = EINVAL; | |
842 | goto fail_vnrele; | |
843 | } | |
a50cd269 | 844 | if ((error = xfs_fs_start_syncd(vfsp))) |
1da177e4 LT |
845 | goto fail_vnrele; |
846 | vn_trace_exit(rootvp, __FUNCTION__, (inst_t *)__return_address); | |
847 | ||
848 | kmem_free(args, sizeof(*args)); | |
849 | return 0; | |
850 | ||
851 | fail_vnrele: | |
852 | if (sb->s_root) { | |
853 | dput(sb->s_root); | |
854 | sb->s_root = NULL; | |
855 | } else { | |
856 | VN_RELE(rootvp); | |
857 | } | |
858 | ||
859 | fail_unmount: | |
b83bd138 | 860 | bhv_vfs_unmount(vfsp, 0, NULL); |
1da177e4 LT |
861 | |
862 | fail_vfsop: | |
863 | vfs_deallocate(vfsp); | |
864 | kmem_free(args, sizeof(*args)); | |
865 | return -error; | |
866 | } | |
867 | ||
454e2398 | 868 | STATIC int |
a50cd269 | 869 | xfs_fs_get_sb( |
1da177e4 LT |
870 | struct file_system_type *fs_type, |
871 | int flags, | |
872 | const char *dev_name, | |
454e2398 DH |
873 | void *data, |
874 | struct vfsmount *mnt) | |
1da177e4 | 875 | { |
454e2398 DH |
876 | return get_sb_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super, |
877 | mnt); | |
a50cd269 NS |
878 | } |
879 | ||
7989cb8e | 880 | static struct super_operations xfs_super_operations = { |
a50cd269 NS |
881 | .alloc_inode = xfs_fs_alloc_inode, |
882 | .destroy_inode = xfs_fs_destroy_inode, | |
883 | .write_inode = xfs_fs_write_inode, | |
884 | .clear_inode = xfs_fs_clear_inode, | |
885 | .put_super = xfs_fs_put_super, | |
886 | .write_super = xfs_fs_write_super, | |
887 | .sync_fs = xfs_fs_sync_super, | |
888 | .write_super_lockfs = xfs_fs_lockfs, | |
889 | .statfs = xfs_fs_statfs, | |
890 | .remount_fs = xfs_fs_remount, | |
891 | .show_options = xfs_fs_show_options, | |
1da177e4 LT |
892 | }; |
893 | ||
7989cb8e | 894 | static struct quotactl_ops xfs_quotactl_operations = { |
a50cd269 NS |
895 | .quota_sync = xfs_fs_quotasync, |
896 | .get_xstate = xfs_fs_getxstate, | |
897 | .set_xstate = xfs_fs_setxstate, | |
898 | .get_xquota = xfs_fs_getxquota, | |
899 | .set_xquota = xfs_fs_setxquota, | |
1da177e4 LT |
900 | }; |
901 | ||
5085b607 | 902 | static struct file_system_type xfs_fs_type = { |
1da177e4 LT |
903 | .owner = THIS_MODULE, |
904 | .name = "xfs", | |
a50cd269 | 905 | .get_sb = xfs_fs_get_sb, |
1da177e4 LT |
906 | .kill_sb = kill_block_super, |
907 | .fs_flags = FS_REQUIRES_DEV, | |
908 | }; | |
909 | ||
910 | ||
911 | STATIC int __init | |
912 | init_xfs_fs( void ) | |
913 | { | |
914 | int error; | |
915 | struct sysinfo si; | |
916 | static char message[] __initdata = KERN_INFO \ | |
917 | XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled\n"; | |
918 | ||
919 | printk(message); | |
920 | ||
921 | si_meminfo(&si); | |
922 | xfs_physmem = si.totalram; | |
923 | ||
924 | ktrace_init(64); | |
925 | ||
8758280f | 926 | error = xfs_init_zones(); |
1da177e4 | 927 | if (error < 0) |
0829c360 | 928 | goto undo_zones; |
1da177e4 | 929 | |
ce8e922c | 930 | error = xfs_buf_init(); |
1da177e4 | 931 | if (error < 0) |
ce8e922c | 932 | goto undo_buffers; |
1da177e4 LT |
933 | |
934 | vn_init(); | |
935 | xfs_init(); | |
936 | uuid_init(); | |
937 | vfs_initquota(); | |
938 | ||
939 | error = register_filesystem(&xfs_fs_type); | |
940 | if (error) | |
941 | goto undo_register; | |
1da177e4 LT |
942 | return 0; |
943 | ||
944 | undo_register: | |
ce8e922c | 945 | xfs_buf_terminate(); |
1da177e4 | 946 | |
ce8e922c | 947 | undo_buffers: |
8758280f | 948 | xfs_destroy_zones(); |
1da177e4 | 949 | |
0829c360 | 950 | undo_zones: |
1da177e4 LT |
951 | return error; |
952 | } | |
953 | ||
954 | STATIC void __exit | |
955 | exit_xfs_fs( void ) | |
956 | { | |
957 | vfs_exitquota(); | |
1da177e4 LT |
958 | unregister_filesystem(&xfs_fs_type); |
959 | xfs_cleanup(); | |
ce8e922c | 960 | xfs_buf_terminate(); |
8758280f | 961 | xfs_destroy_zones(); |
1da177e4 LT |
962 | ktrace_uninit(); |
963 | } | |
964 | ||
965 | module_init(init_xfs_fs); | |
966 | module_exit(exit_xfs_fs); | |
967 | ||
968 | MODULE_AUTHOR("Silicon Graphics, Inc."); | |
969 | MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled"); | |
970 | MODULE_LICENSE("GPL"); |