]>
Commit | Line | Data |
---|---|---|
68988114 DC |
1 | /* |
2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. | |
c24b5dfa | 3 | * Copyright (c) 2012 Red Hat, Inc. |
68988114 DC |
4 | * All Rights Reserved. |
5 | * | |
6 | * This program is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU General Public License as | |
8 | * published by the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope that it would be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License | |
16 | * along with this program; if not, write the Free Software Foundation, | |
17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
18 | */ | |
19 | #include "xfs.h" | |
20 | #include "xfs_fs.h" | |
70a9883c | 21 | #include "xfs_shared.h" |
239880ef DC |
22 | #include "xfs_format.h" |
23 | #include "xfs_log_format.h" | |
24 | #include "xfs_trans_resv.h" | |
68988114 | 25 | #include "xfs_bit.h" |
68988114 | 26 | #include "xfs_mount.h" |
57062787 | 27 | #include "xfs_da_format.h" |
68988114 DC |
28 | #include "xfs_inode.h" |
29 | #include "xfs_btree.h" | |
239880ef | 30 | #include "xfs_trans.h" |
68988114 DC |
31 | #include "xfs_extfree_item.h" |
32 | #include "xfs_alloc.h" | |
33 | #include "xfs_bmap.h" | |
34 | #include "xfs_bmap_util.h" | |
a4fbe6ab | 35 | #include "xfs_bmap_btree.h" |
68988114 DC |
36 | #include "xfs_rtalloc.h" |
37 | #include "xfs_error.h" | |
38 | #include "xfs_quota.h" | |
39 | #include "xfs_trans_space.h" | |
40 | #include "xfs_trace.h" | |
c24b5dfa | 41 | #include "xfs_icache.h" |
239880ef | 42 | #include "xfs_log.h" |
68988114 DC |
43 | |
44 | /* Kernel only BMAP related definitions and functions */ | |
45 | ||
46 | /* | |
47 | * Convert the given file system block to a disk block. We have to treat it | |
48 | * differently based on whether the file is a real time file or not, because the | |
49 | * bmap code does. | |
50 | */ | |
51 | xfs_daddr_t | |
52 | xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb) | |
53 | { | |
54 | return (XFS_IS_REALTIME_INODE(ip) ? \ | |
55 | (xfs_daddr_t)XFS_FSB_TO_BB((ip)->i_mount, (fsb)) : \ | |
56 | XFS_FSB_TO_DADDR((ip)->i_mount, (fsb))); | |
57 | } | |
58 | ||
3fbbbea3 DC |
59 | /* |
60 | * Routine to zero an extent on disk allocated to the specific inode. | |
61 | * | |
62 | * The VFS functions take a linearised filesystem block offset, so we have to | |
63 | * convert the sparse xfs fsb to the right format first. | |
64 | * VFS types are real funky, too. | |
65 | */ | |
66 | int | |
67 | xfs_zero_extent( | |
68 | struct xfs_inode *ip, | |
69 | xfs_fsblock_t start_fsb, | |
70 | xfs_off_t count_fsb) | |
71 | { | |
72 | struct xfs_mount *mp = ip->i_mount; | |
73 | xfs_daddr_t sector = xfs_fsb_to_db(ip, start_fsb); | |
74 | sector_t block = XFS_BB_TO_FSBT(mp, sector); | |
3fbbbea3 | 75 | |
3dc29161 MW |
76 | return blkdev_issue_zeroout(xfs_find_bdev_for_inode(VFS_I(ip)), |
77 | block << (mp->m_super->s_blocksize_bits - 9), | |
78 | count_fsb << (mp->m_super->s_blocksize_bits - 9), | |
79 | GFP_NOFS, true); | |
3fbbbea3 DC |
80 | } |
81 | ||
e66a4c67 DW |
82 | /* Sort bmap items by AG. */ |
83 | static int | |
84 | xfs_bmap_free_list_cmp( | |
85 | void *priv, | |
86 | struct list_head *a, | |
87 | struct list_head *b) | |
88 | { | |
89 | struct xfs_mount *mp = priv; | |
90 | struct xfs_bmap_free_item *ra; | |
91 | struct xfs_bmap_free_item *rb; | |
92 | ||
93 | ra = container_of(a, struct xfs_bmap_free_item, xbfi_list); | |
94 | rb = container_of(b, struct xfs_bmap_free_item, xbfi_list); | |
95 | return XFS_FSB_TO_AGNO(mp, ra->xbfi_startblock) - | |
96 | XFS_FSB_TO_AGNO(mp, rb->xbfi_startblock); | |
97 | } | |
98 | ||
68988114 DC |
99 | /* |
100 | * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi | |
101 | * caller. Frees all the extents that need freeing, which must be done | |
102 | * last due to locking considerations. We never free any extents in | |
103 | * the first transaction. | |
104 | * | |
f6106efa ES |
105 | * If an inode *ip is provided, rejoin it to the transaction if |
106 | * the transaction was committed. | |
68988114 DC |
107 | */ |
108 | int /* error */ | |
109 | xfs_bmap_finish( | |
8d99fe92 BF |
110 | struct xfs_trans **tp, /* transaction pointer addr */ |
111 | struct xfs_bmap_free *flist, /* i/o: list extents to free */ | |
f6106efa | 112 | struct xfs_inode *ip) |
68988114 | 113 | { |
8d99fe92 BF |
114 | struct xfs_efd_log_item *efd; /* extent free data */ |
115 | struct xfs_efi_log_item *efi; /* extent free intention */ | |
116 | int error; /* error return value */ | |
f6106efa | 117 | int committed;/* xact committed or not */ |
8d99fe92 | 118 | struct xfs_bmap_free_item *free; /* free extent item */ |
68988114 DC |
119 | |
120 | ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES); | |
f6106efa | 121 | if (flist->xbf_count == 0) |
68988114 | 122 | return 0; |
f6106efa | 123 | |
e66a4c67 DW |
124 | list_sort((*tp)->t_mountp, &flist->xbf_flist, xfs_bmap_free_list_cmp); |
125 | ||
2e6db6c4 | 126 | efi = xfs_trans_get_efi(*tp, flist->xbf_count); |
e66a4c67 | 127 | list_for_each_entry(free, &flist->xbf_flist, xbfi_list) |
2e6db6c4 | 128 | xfs_trans_log_efi_extent(*tp, efi, free->xbfi_startblock, |
68988114 | 129 | free->xbfi_blockcount); |
3d3c8b52 | 130 | |
f6106efa | 131 | error = __xfs_trans_roll(tp, ip, &committed); |
8d99fe92 BF |
132 | if (error) { |
133 | /* | |
134 | * If the transaction was committed, drop the EFD reference | |
135 | * since we're bailing out of here. The other reference is | |
136 | * dropped when the EFI hits the AIL. | |
137 | * | |
138 | * If the transaction was not committed, the EFI is freed by the | |
139 | * EFI item unlock handler on abort. Also, we have a new | |
140 | * transaction so we should return committed=1 even though we're | |
141 | * returning an error. | |
142 | */ | |
f6106efa | 143 | if (committed) { |
8d99fe92 BF |
144 | xfs_efi_release(efi); |
145 | xfs_force_shutdown((*tp)->t_mountp, | |
0c871f9a | 146 | SHUTDOWN_META_IO_ERROR); |
8d99fe92 | 147 | } |
68988114 | 148 | return error; |
8d99fe92 | 149 | } |
68988114 | 150 | |
6bc43af3 BF |
151 | /* |
152 | * Get an EFD and free each extent in the list, logging to the EFD in | |
153 | * the process. The remaining bmap free list is cleaned up by the caller | |
154 | * on error. | |
155 | */ | |
2e6db6c4 | 156 | efd = xfs_trans_get_efd(*tp, efi, flist->xbf_count); |
e66a4c67 DW |
157 | while (!list_empty(&flist->xbf_flist)) { |
158 | free = list_first_entry(&flist->xbf_flist, | |
159 | struct xfs_bmap_free_item, xbfi_list); | |
6bc43af3 BF |
160 | error = xfs_trans_free_extent(*tp, efd, free->xbfi_startblock, |
161 | free->xbfi_blockcount); | |
8d99fe92 BF |
162 | if (error) |
163 | return error; | |
164 | ||
e66a4c67 | 165 | xfs_bmap_del_free(flist, free); |
68988114 | 166 | } |
8d99fe92 | 167 | |
68988114 DC |
168 | return 0; |
169 | } | |
170 | ||
171 | int | |
172 | xfs_bmap_rtalloc( | |
173 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ | |
174 | { | |
175 | xfs_alloctype_t atype = 0; /* type for allocation routines */ | |
176 | int error; /* error return value */ | |
177 | xfs_mount_t *mp; /* mount point structure */ | |
178 | xfs_extlen_t prod = 0; /* product factor for allocators */ | |
179 | xfs_extlen_t ralen = 0; /* realtime allocation length */ | |
180 | xfs_extlen_t align; /* minimum allocation alignment */ | |
181 | xfs_rtblock_t rtb; | |
182 | ||
183 | mp = ap->ip->i_mount; | |
184 | align = xfs_get_extsz_hint(ap->ip); | |
185 | prod = align / mp->m_sb.sb_rextsize; | |
186 | error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, | |
187 | align, 1, ap->eof, 0, | |
188 | ap->conv, &ap->offset, &ap->length); | |
189 | if (error) | |
190 | return error; | |
191 | ASSERT(ap->length); | |
192 | ASSERT(ap->length % mp->m_sb.sb_rextsize == 0); | |
193 | ||
194 | /* | |
195 | * If the offset & length are not perfectly aligned | |
196 | * then kill prod, it will just get us in trouble. | |
197 | */ | |
198 | if (do_mod(ap->offset, align) || ap->length % align) | |
199 | prod = 1; | |
200 | /* | |
201 | * Set ralen to be the actual requested length in rtextents. | |
202 | */ | |
203 | ralen = ap->length / mp->m_sb.sb_rextsize; | |
204 | /* | |
205 | * If the old value was close enough to MAXEXTLEN that | |
206 | * we rounded up to it, cut it back so it's valid again. | |
207 | * Note that if it's a really large request (bigger than | |
208 | * MAXEXTLEN), we don't hear about that number, and can't | |
209 | * adjust the starting point to match it. | |
210 | */ | |
211 | if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN) | |
212 | ralen = MAXEXTLEN / mp->m_sb.sb_rextsize; | |
213 | ||
214 | /* | |
4b680afb | 215 | * Lock out modifications to both the RT bitmap and summary inodes |
68988114 | 216 | */ |
f4a0660d | 217 | xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP); |
68988114 | 218 | xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL); |
f4a0660d | 219 | xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM); |
4b680afb | 220 | xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL); |
68988114 DC |
221 | |
222 | /* | |
223 | * If it's an allocation to an empty file at offset 0, | |
224 | * pick an extent that will space things out in the rt area. | |
225 | */ | |
226 | if (ap->eof && ap->offset == 0) { | |
227 | xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */ | |
228 | ||
229 | error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx); | |
230 | if (error) | |
231 | return error; | |
232 | ap->blkno = rtx * mp->m_sb.sb_rextsize; | |
233 | } else { | |
234 | ap->blkno = 0; | |
235 | } | |
236 | ||
237 | xfs_bmap_adjacent(ap); | |
238 | ||
239 | /* | |
240 | * Realtime allocation, done through xfs_rtallocate_extent. | |
241 | */ | |
242 | atype = ap->blkno == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO; | |
243 | do_div(ap->blkno, mp->m_sb.sb_rextsize); | |
244 | rtb = ap->blkno; | |
245 | ap->length = ralen; | |
246 | if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length, | |
247 | &ralen, atype, ap->wasdel, prod, &rtb))) | |
248 | return error; | |
249 | if (rtb == NULLFSBLOCK && prod > 1 && | |
250 | (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, | |
251 | ap->length, &ralen, atype, | |
252 | ap->wasdel, 1, &rtb))) | |
253 | return error; | |
254 | ap->blkno = rtb; | |
255 | if (ap->blkno != NULLFSBLOCK) { | |
256 | ap->blkno *= mp->m_sb.sb_rextsize; | |
257 | ralen *= mp->m_sb.sb_rextsize; | |
258 | ap->length = ralen; | |
259 | ap->ip->i_d.di_nblocks += ralen; | |
260 | xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE); | |
261 | if (ap->wasdel) | |
262 | ap->ip->i_delayed_blks -= ralen; | |
263 | /* | |
264 | * Adjust the disk quota also. This was reserved | |
265 | * earlier. | |
266 | */ | |
267 | xfs_trans_mod_dquot_byino(ap->tp, ap->ip, | |
268 | ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT : | |
269 | XFS_TRANS_DQ_RTBCOUNT, (long) ralen); | |
3fbbbea3 DC |
270 | |
271 | /* Zero the extent if we were asked to do so */ | |
272 | if (ap->userdata & XFS_ALLOC_USERDATA_ZERO) { | |
273 | error = xfs_zero_extent(ap->ip, ap->blkno, ap->length); | |
274 | if (error) | |
275 | return error; | |
276 | } | |
68988114 DC |
277 | } else { |
278 | ap->length = 0; | |
279 | } | |
280 | return 0; | |
281 | } | |
282 | ||
68988114 DC |
283 | /* |
284 | * Check if the endoff is outside the last extent. If so the caller will grow | |
285 | * the allocation to a stripe unit boundary. All offsets are considered outside | |
286 | * the end of file for an empty fork, so 1 is returned in *eof in that case. | |
287 | */ | |
288 | int | |
289 | xfs_bmap_eof( | |
290 | struct xfs_inode *ip, | |
291 | xfs_fileoff_t endoff, | |
292 | int whichfork, | |
293 | int *eof) | |
294 | { | |
295 | struct xfs_bmbt_irec rec; | |
296 | int error; | |
297 | ||
298 | error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof); | |
299 | if (error || *eof) | |
300 | return error; | |
301 | ||
302 | *eof = endoff >= rec.br_startoff + rec.br_blockcount; | |
303 | return 0; | |
304 | } | |
305 | ||
306 | /* | |
307 | * Extent tree block counting routines. | |
308 | */ | |
309 | ||
310 | /* | |
311 | * Count leaf blocks given a range of extent records. | |
312 | */ | |
313 | STATIC void | |
314 | xfs_bmap_count_leaves( | |
315 | xfs_ifork_t *ifp, | |
316 | xfs_extnum_t idx, | |
317 | int numrecs, | |
318 | int *count) | |
319 | { | |
320 | int b; | |
321 | ||
322 | for (b = 0; b < numrecs; b++) { | |
323 | xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b); | |
324 | *count += xfs_bmbt_get_blockcount(frp); | |
325 | } | |
326 | } | |
327 | ||
328 | /* | |
329 | * Count leaf blocks given a range of extent records originally | |
330 | * in btree format. | |
331 | */ | |
332 | STATIC void | |
333 | xfs_bmap_disk_count_leaves( | |
334 | struct xfs_mount *mp, | |
335 | struct xfs_btree_block *block, | |
336 | int numrecs, | |
337 | int *count) | |
338 | { | |
339 | int b; | |
340 | xfs_bmbt_rec_t *frp; | |
341 | ||
342 | for (b = 1; b <= numrecs; b++) { | |
343 | frp = XFS_BMBT_REC_ADDR(mp, block, b); | |
344 | *count += xfs_bmbt_disk_get_blockcount(frp); | |
345 | } | |
346 | } | |
347 | ||
348 | /* | |
349 | * Recursively walks each level of a btree | |
8be11e92 | 350 | * to count total fsblocks in use. |
68988114 DC |
351 | */ |
352 | STATIC int /* error */ | |
353 | xfs_bmap_count_tree( | |
354 | xfs_mount_t *mp, /* file system mount point */ | |
355 | xfs_trans_t *tp, /* transaction pointer */ | |
356 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
357 | xfs_fsblock_t blockno, /* file system block number */ | |
358 | int levelin, /* level in btree */ | |
359 | int *count) /* Count of blocks */ | |
360 | { | |
361 | int error; | |
362 | xfs_buf_t *bp, *nbp; | |
363 | int level = levelin; | |
364 | __be64 *pp; | |
365 | xfs_fsblock_t bno = blockno; | |
366 | xfs_fsblock_t nextbno; | |
367 | struct xfs_btree_block *block, *nextblock; | |
368 | int numrecs; | |
369 | ||
370 | error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF, | |
371 | &xfs_bmbt_buf_ops); | |
372 | if (error) | |
373 | return error; | |
374 | *count += 1; | |
375 | block = XFS_BUF_TO_BLOCK(bp); | |
376 | ||
377 | if (--level) { | |
378 | /* Not at node above leaves, count this level of nodes */ | |
379 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); | |
380 | while (nextbno != NULLFSBLOCK) { | |
381 | error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp, | |
382 | XFS_BMAP_BTREE_REF, | |
383 | &xfs_bmbt_buf_ops); | |
384 | if (error) | |
385 | return error; | |
386 | *count += 1; | |
387 | nextblock = XFS_BUF_TO_BLOCK(nbp); | |
388 | nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib); | |
389 | xfs_trans_brelse(tp, nbp); | |
390 | } | |
391 | ||
392 | /* Dive to the next level */ | |
393 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]); | |
394 | bno = be64_to_cpu(*pp); | |
395 | if (unlikely((error = | |
396 | xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) { | |
397 | xfs_trans_brelse(tp, bp); | |
398 | XFS_ERROR_REPORT("xfs_bmap_count_tree(1)", | |
399 | XFS_ERRLEVEL_LOW, mp); | |
2451337d | 400 | return -EFSCORRUPTED; |
68988114 DC |
401 | } |
402 | xfs_trans_brelse(tp, bp); | |
403 | } else { | |
404 | /* count all level 1 nodes and their leaves */ | |
405 | for (;;) { | |
406 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); | |
407 | numrecs = be16_to_cpu(block->bb_numrecs); | |
408 | xfs_bmap_disk_count_leaves(mp, block, numrecs, count); | |
409 | xfs_trans_brelse(tp, bp); | |
410 | if (nextbno == NULLFSBLOCK) | |
411 | break; | |
412 | bno = nextbno; | |
413 | error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, | |
414 | XFS_BMAP_BTREE_REF, | |
415 | &xfs_bmbt_buf_ops); | |
416 | if (error) | |
417 | return error; | |
418 | *count += 1; | |
419 | block = XFS_BUF_TO_BLOCK(bp); | |
420 | } | |
421 | } | |
422 | return 0; | |
423 | } | |
424 | ||
425 | /* | |
426 | * Count fsblocks of the given fork. | |
427 | */ | |
0d5a75e9 | 428 | static int /* error */ |
68988114 DC |
429 | xfs_bmap_count_blocks( |
430 | xfs_trans_t *tp, /* transaction pointer */ | |
431 | xfs_inode_t *ip, /* incore inode */ | |
432 | int whichfork, /* data or attr fork */ | |
433 | int *count) /* out: count of blocks */ | |
434 | { | |
435 | struct xfs_btree_block *block; /* current btree block */ | |
436 | xfs_fsblock_t bno; /* block # of "block" */ | |
437 | xfs_ifork_t *ifp; /* fork structure */ | |
438 | int level; /* btree level, for checking */ | |
439 | xfs_mount_t *mp; /* file system mount structure */ | |
440 | __be64 *pp; /* pointer to block address */ | |
441 | ||
442 | bno = NULLFSBLOCK; | |
443 | mp = ip->i_mount; | |
444 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
445 | if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) { | |
446 | xfs_bmap_count_leaves(ifp, 0, | |
447 | ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t), | |
448 | count); | |
449 | return 0; | |
450 | } | |
451 | ||
452 | /* | |
453 | * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out. | |
454 | */ | |
455 | block = ifp->if_broot; | |
456 | level = be16_to_cpu(block->bb_level); | |
457 | ASSERT(level > 0); | |
458 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes); | |
459 | bno = be64_to_cpu(*pp); | |
d5cf09ba | 460 | ASSERT(bno != NULLFSBLOCK); |
68988114 DC |
461 | ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount); |
462 | ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks); | |
463 | ||
464 | if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) { | |
465 | XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW, | |
466 | mp); | |
2451337d | 467 | return -EFSCORRUPTED; |
68988114 DC |
468 | } |
469 | ||
470 | return 0; | |
471 | } | |
472 | ||
473 | /* | |
474 | * returns 1 for success, 0 if we failed to map the extent. | |
475 | */ | |
476 | STATIC int | |
477 | xfs_getbmapx_fix_eof_hole( | |
478 | xfs_inode_t *ip, /* xfs incore inode pointer */ | |
479 | struct getbmapx *out, /* output structure */ | |
480 | int prealloced, /* this is a file with | |
481 | * preallocated data space */ | |
482 | __int64_t end, /* last block requested */ | |
483 | xfs_fsblock_t startblock) | |
484 | { | |
485 | __int64_t fixlen; | |
486 | xfs_mount_t *mp; /* file system mount point */ | |
487 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
488 | xfs_extnum_t lastx; /* last extent pointer */ | |
489 | xfs_fileoff_t fileblock; | |
490 | ||
491 | if (startblock == HOLESTARTBLOCK) { | |
492 | mp = ip->i_mount; | |
493 | out->bmv_block = -1; | |
494 | fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip))); | |
495 | fixlen -= out->bmv_offset; | |
496 | if (prealloced && out->bmv_offset + out->bmv_length == end) { | |
497 | /* Came to hole at EOF. Trim it. */ | |
498 | if (fixlen <= 0) | |
499 | return 0; | |
500 | out->bmv_length = fixlen; | |
501 | } | |
502 | } else { | |
503 | if (startblock == DELAYSTARTBLOCK) | |
504 | out->bmv_block = -2; | |
505 | else | |
506 | out->bmv_block = xfs_fsb_to_db(ip, startblock); | |
507 | fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset); | |
508 | ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK); | |
509 | if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) && | |
510 | (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1)) | |
511 | out->bmv_oflags |= BMV_OF_LAST; | |
512 | } | |
513 | ||
514 | return 1; | |
515 | } | |
516 | ||
517 | /* | |
518 | * Get inode's extents as described in bmv, and format for output. | |
519 | * Calls formatter to fill the user's buffer until all extents | |
520 | * are mapped, until the passed-in bmv->bmv_count slots have | |
521 | * been filled, or until the formatter short-circuits the loop, | |
522 | * if it is tracking filled-in extents on its own. | |
523 | */ | |
524 | int /* error code */ | |
525 | xfs_getbmap( | |
526 | xfs_inode_t *ip, | |
527 | struct getbmapx *bmv, /* user bmap structure */ | |
528 | xfs_bmap_format_t formatter, /* format to user */ | |
529 | void *arg) /* formatter arg */ | |
530 | { | |
531 | __int64_t bmvend; /* last block requested */ | |
532 | int error = 0; /* return value */ | |
533 | __int64_t fixlen; /* length for -1 case */ | |
534 | int i; /* extent number */ | |
535 | int lock; /* lock state */ | |
536 | xfs_bmbt_irec_t *map; /* buffer for user's data */ | |
537 | xfs_mount_t *mp; /* file system mount point */ | |
538 | int nex; /* # of user extents can do */ | |
539 | int nexleft; /* # of user extents left */ | |
540 | int subnex; /* # of bmapi's can do */ | |
541 | int nmap; /* number of map entries */ | |
542 | struct getbmapx *out; /* output structure */ | |
543 | int whichfork; /* data or attr fork */ | |
544 | int prealloced; /* this is a file with | |
545 | * preallocated data space */ | |
546 | int iflags; /* interface flags */ | |
547 | int bmapi_flags; /* flags for xfs_bmapi */ | |
548 | int cur_ext = 0; | |
549 | ||
550 | mp = ip->i_mount; | |
551 | iflags = bmv->bmv_iflags; | |
552 | whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK; | |
553 | ||
554 | if (whichfork == XFS_ATTR_FORK) { | |
555 | if (XFS_IFORK_Q(ip)) { | |
556 | if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS && | |
557 | ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE && | |
558 | ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL) | |
2451337d | 559 | return -EINVAL; |
68988114 DC |
560 | } else if (unlikely( |
561 | ip->i_d.di_aformat != 0 && | |
562 | ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) { | |
563 | XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW, | |
564 | ip->i_mount); | |
2451337d | 565 | return -EFSCORRUPTED; |
68988114 DC |
566 | } |
567 | ||
568 | prealloced = 0; | |
569 | fixlen = 1LL << 32; | |
570 | } else { | |
571 | if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS && | |
572 | ip->i_d.di_format != XFS_DINODE_FMT_BTREE && | |
573 | ip->i_d.di_format != XFS_DINODE_FMT_LOCAL) | |
2451337d | 574 | return -EINVAL; |
68988114 DC |
575 | |
576 | if (xfs_get_extsz_hint(ip) || | |
577 | ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){ | |
578 | prealloced = 1; | |
579 | fixlen = mp->m_super->s_maxbytes; | |
580 | } else { | |
581 | prealloced = 0; | |
582 | fixlen = XFS_ISIZE(ip); | |
583 | } | |
584 | } | |
585 | ||
586 | if (bmv->bmv_length == -1) { | |
587 | fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen)); | |
588 | bmv->bmv_length = | |
589 | max_t(__int64_t, fixlen - bmv->bmv_offset, 0); | |
590 | } else if (bmv->bmv_length == 0) { | |
591 | bmv->bmv_entries = 0; | |
592 | return 0; | |
593 | } else if (bmv->bmv_length < 0) { | |
2451337d | 594 | return -EINVAL; |
68988114 DC |
595 | } |
596 | ||
597 | nex = bmv->bmv_count - 1; | |
598 | if (nex <= 0) | |
2451337d | 599 | return -EINVAL; |
68988114 DC |
600 | bmvend = bmv->bmv_offset + bmv->bmv_length; |
601 | ||
602 | ||
603 | if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx)) | |
2451337d | 604 | return -ENOMEM; |
fdd3ccee DC |
605 | out = kmem_zalloc_large(bmv->bmv_count * sizeof(struct getbmapx), 0); |
606 | if (!out) | |
2451337d | 607 | return -ENOMEM; |
68988114 DC |
608 | |
609 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
efa70be1 CH |
610 | if (whichfork == XFS_DATA_FORK) { |
611 | if (!(iflags & BMV_IF_DELALLOC) && | |
612 | (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) { | |
2451337d | 613 | error = filemap_write_and_wait(VFS_I(ip)->i_mapping); |
68988114 DC |
614 | if (error) |
615 | goto out_unlock_iolock; | |
efa70be1 CH |
616 | |
617 | /* | |
618 | * Even after flushing the inode, there can still be | |
619 | * delalloc blocks on the inode beyond EOF due to | |
620 | * speculative preallocation. These are not removed | |
621 | * until the release function is called or the inode | |
622 | * is inactivated. Hence we cannot assert here that | |
623 | * ip->i_delayed_blks == 0. | |
624 | */ | |
68988114 | 625 | } |
68988114 | 626 | |
efa70be1 CH |
627 | lock = xfs_ilock_data_map_shared(ip); |
628 | } else { | |
629 | lock = xfs_ilock_attr_map_shared(ip); | |
630 | } | |
68988114 DC |
631 | |
632 | /* | |
633 | * Don't let nex be bigger than the number of extents | |
634 | * we can have assuming alternating holes and real extents. | |
635 | */ | |
636 | if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1) | |
637 | nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1; | |
638 | ||
639 | bmapi_flags = xfs_bmapi_aflag(whichfork); | |
640 | if (!(iflags & BMV_IF_PREALLOC)) | |
641 | bmapi_flags |= XFS_BMAPI_IGSTATE; | |
642 | ||
643 | /* | |
644 | * Allocate enough space to handle "subnex" maps at a time. | |
645 | */ | |
2451337d | 646 | error = -ENOMEM; |
68988114 DC |
647 | subnex = 16; |
648 | map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS); | |
649 | if (!map) | |
650 | goto out_unlock_ilock; | |
651 | ||
652 | bmv->bmv_entries = 0; | |
653 | ||
654 | if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 && | |
655 | (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) { | |
656 | error = 0; | |
657 | goto out_free_map; | |
658 | } | |
659 | ||
660 | nexleft = nex; | |
661 | ||
662 | do { | |
663 | nmap = (nexleft > subnex) ? subnex : nexleft; | |
664 | error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset), | |
665 | XFS_BB_TO_FSB(mp, bmv->bmv_length), | |
666 | map, &nmap, bmapi_flags); | |
667 | if (error) | |
668 | goto out_free_map; | |
669 | ASSERT(nmap <= subnex); | |
670 | ||
671 | for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) { | |
672 | out[cur_ext].bmv_oflags = 0; | |
673 | if (map[i].br_state == XFS_EXT_UNWRITTEN) | |
674 | out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC; | |
675 | else if (map[i].br_startblock == DELAYSTARTBLOCK) | |
676 | out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC; | |
677 | out[cur_ext].bmv_offset = | |
678 | XFS_FSB_TO_BB(mp, map[i].br_startoff); | |
679 | out[cur_ext].bmv_length = | |
680 | XFS_FSB_TO_BB(mp, map[i].br_blockcount); | |
681 | out[cur_ext].bmv_unused1 = 0; | |
682 | out[cur_ext].bmv_unused2 = 0; | |
683 | ||
684 | /* | |
685 | * delayed allocation extents that start beyond EOF can | |
686 | * occur due to speculative EOF allocation when the | |
687 | * delalloc extent is larger than the largest freespace | |
688 | * extent at conversion time. These extents cannot be | |
689 | * converted by data writeback, so can exist here even | |
690 | * if we are not supposed to be finding delalloc | |
691 | * extents. | |
692 | */ | |
693 | if (map[i].br_startblock == DELAYSTARTBLOCK && | |
694 | map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip))) | |
695 | ASSERT((iflags & BMV_IF_DELALLOC) != 0); | |
696 | ||
697 | if (map[i].br_startblock == HOLESTARTBLOCK && | |
698 | whichfork == XFS_ATTR_FORK) { | |
699 | /* came to the end of attribute fork */ | |
700 | out[cur_ext].bmv_oflags |= BMV_OF_LAST; | |
701 | goto out_free_map; | |
702 | } | |
703 | ||
704 | if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext], | |
705 | prealloced, bmvend, | |
706 | map[i].br_startblock)) | |
707 | goto out_free_map; | |
708 | ||
709 | bmv->bmv_offset = | |
710 | out[cur_ext].bmv_offset + | |
711 | out[cur_ext].bmv_length; | |
712 | bmv->bmv_length = | |
713 | max_t(__int64_t, 0, bmvend - bmv->bmv_offset); | |
714 | ||
715 | /* | |
716 | * In case we don't want to return the hole, | |
717 | * don't increase cur_ext so that we can reuse | |
718 | * it in the next loop. | |
719 | */ | |
720 | if ((iflags & BMV_IF_NO_HOLES) && | |
721 | map[i].br_startblock == HOLESTARTBLOCK) { | |
722 | memset(&out[cur_ext], 0, sizeof(out[cur_ext])); | |
723 | continue; | |
724 | } | |
725 | ||
726 | nexleft--; | |
727 | bmv->bmv_entries++; | |
728 | cur_ext++; | |
729 | } | |
730 | } while (nmap && nexleft && bmv->bmv_length); | |
731 | ||
732 | out_free_map: | |
733 | kmem_free(map); | |
734 | out_unlock_ilock: | |
01f4f327 | 735 | xfs_iunlock(ip, lock); |
68988114 DC |
736 | out_unlock_iolock: |
737 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
738 | ||
739 | for (i = 0; i < cur_ext; i++) { | |
740 | int full = 0; /* user array is full */ | |
741 | ||
742 | /* format results & advance arg */ | |
743 | error = formatter(&arg, &out[i], &full); | |
744 | if (error || full) | |
745 | break; | |
746 | } | |
747 | ||
fdd3ccee | 748 | kmem_free(out); |
68988114 DC |
749 | return error; |
750 | } | |
751 | ||
752 | /* | |
753 | * dead simple method of punching delalyed allocation blocks from a range in | |
754 | * the inode. Walks a block at a time so will be slow, but is only executed in | |
ad4809bf | 755 | * rare error cases so the overhead is not critical. This will always punch out |
68988114 DC |
756 | * both the start and end blocks, even if the ranges only partially overlap |
757 | * them, so it is up to the caller to ensure that partial blocks are not | |
758 | * passed in. | |
759 | */ | |
760 | int | |
761 | xfs_bmap_punch_delalloc_range( | |
762 | struct xfs_inode *ip, | |
763 | xfs_fileoff_t start_fsb, | |
764 | xfs_fileoff_t length) | |
765 | { | |
766 | xfs_fileoff_t remaining = length; | |
767 | int error = 0; | |
768 | ||
769 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
770 | ||
771 | do { | |
772 | int done; | |
773 | xfs_bmbt_irec_t imap; | |
774 | int nimaps = 1; | |
775 | xfs_fsblock_t firstblock; | |
776 | xfs_bmap_free_t flist; | |
777 | ||
778 | /* | |
779 | * Map the range first and check that it is a delalloc extent | |
780 | * before trying to unmap the range. Otherwise we will be | |
781 | * trying to remove a real extent (which requires a | |
782 | * transaction) or a hole, which is probably a bad idea... | |
783 | */ | |
784 | error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps, | |
785 | XFS_BMAPI_ENTIRE); | |
786 | ||
787 | if (error) { | |
788 | /* something screwed, just bail */ | |
789 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
790 | xfs_alert(ip->i_mount, | |
791 | "Failed delalloc mapping lookup ino %lld fsb %lld.", | |
792 | ip->i_ino, start_fsb); | |
793 | } | |
794 | break; | |
795 | } | |
796 | if (!nimaps) { | |
797 | /* nothing there */ | |
798 | goto next_block; | |
799 | } | |
800 | if (imap.br_startblock != DELAYSTARTBLOCK) { | |
801 | /* been converted, ignore */ | |
802 | goto next_block; | |
803 | } | |
804 | WARN_ON(imap.br_blockcount == 0); | |
805 | ||
806 | /* | |
807 | * Note: while we initialise the firstblock/flist pair, they | |
808 | * should never be used because blocks should never be | |
809 | * allocated or freed for a delalloc extent and hence we need | |
810 | * don't cancel or finish them after the xfs_bunmapi() call. | |
811 | */ | |
812 | xfs_bmap_init(&flist, &firstblock); | |
813 | error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock, | |
814 | &flist, &done); | |
815 | if (error) | |
816 | break; | |
817 | ||
e66a4c67 | 818 | ASSERT(!flist.xbf_count && list_empty(&flist.xbf_flist)); |
68988114 DC |
819 | next_block: |
820 | start_fsb++; | |
821 | remaining--; | |
822 | } while(remaining > 0); | |
823 | ||
824 | return error; | |
825 | } | |
c24b5dfa DC |
826 | |
827 | /* | |
828 | * Test whether it is appropriate to check an inode for and free post EOF | |
829 | * blocks. The 'force' parameter determines whether we should also consider | |
830 | * regular files that are marked preallocated or append-only. | |
831 | */ | |
832 | bool | |
833 | xfs_can_free_eofblocks(struct xfs_inode *ip, bool force) | |
834 | { | |
835 | /* prealloc/delalloc exists only on regular files */ | |
c19b3b05 | 836 | if (!S_ISREG(VFS_I(ip)->i_mode)) |
c24b5dfa DC |
837 | return false; |
838 | ||
839 | /* | |
840 | * Zero sized files with no cached pages and delalloc blocks will not | |
841 | * have speculative prealloc/delalloc blocks to remove. | |
842 | */ | |
843 | if (VFS_I(ip)->i_size == 0 && | |
2667c6f9 | 844 | VFS_I(ip)->i_mapping->nrpages == 0 && |
c24b5dfa DC |
845 | ip->i_delayed_blks == 0) |
846 | return false; | |
847 | ||
848 | /* If we haven't read in the extent list, then don't do it now. */ | |
849 | if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) | |
850 | return false; | |
851 | ||
852 | /* | |
853 | * Do not free real preallocated or append-only files unless the file | |
854 | * has delalloc blocks and we are forced to remove them. | |
855 | */ | |
856 | if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) | |
857 | if (!force || ip->i_delayed_blks == 0) | |
858 | return false; | |
859 | ||
860 | return true; | |
861 | } | |
862 | ||
863 | /* | |
864 | * This is called by xfs_inactive to free any blocks beyond eof | |
865 | * when the link count isn't zero and by xfs_dm_punch_hole() when | |
866 | * punching a hole to EOF. | |
867 | */ | |
868 | int | |
869 | xfs_free_eofblocks( | |
870 | xfs_mount_t *mp, | |
871 | xfs_inode_t *ip, | |
872 | bool need_iolock) | |
873 | { | |
874 | xfs_trans_t *tp; | |
875 | int error; | |
876 | xfs_fileoff_t end_fsb; | |
877 | xfs_fileoff_t last_fsb; | |
878 | xfs_filblks_t map_len; | |
879 | int nimaps; | |
880 | xfs_bmbt_irec_t imap; | |
881 | ||
882 | /* | |
883 | * Figure out if there are any blocks beyond the end | |
884 | * of the file. If not, then there is nothing to do. | |
885 | */ | |
886 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip)); | |
887 | last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); | |
888 | if (last_fsb <= end_fsb) | |
889 | return 0; | |
890 | map_len = last_fsb - end_fsb; | |
891 | ||
892 | nimaps = 1; | |
893 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
894 | error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0); | |
895 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
896 | ||
897 | if (!error && (nimaps != 0) && | |
898 | (imap.br_startblock != HOLESTARTBLOCK || | |
899 | ip->i_delayed_blks)) { | |
900 | /* | |
901 | * Attach the dquots to the inode up front. | |
902 | */ | |
903 | error = xfs_qm_dqattach(ip, 0); | |
904 | if (error) | |
905 | return error; | |
906 | ||
907 | /* | |
908 | * There are blocks after the end of file. | |
909 | * Free them up now by truncating the file to | |
910 | * its current size. | |
911 | */ | |
c24b5dfa | 912 | if (need_iolock) { |
253f4911 | 913 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) |
2451337d | 914 | return -EAGAIN; |
c24b5dfa DC |
915 | } |
916 | ||
253f4911 CH |
917 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, |
918 | &tp); | |
c24b5dfa DC |
919 | if (error) { |
920 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
c24b5dfa DC |
921 | if (need_iolock) |
922 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
923 | return error; | |
924 | } | |
925 | ||
926 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
927 | xfs_trans_ijoin(tp, ip, 0); | |
928 | ||
929 | /* | |
930 | * Do not update the on-disk file size. If we update the | |
931 | * on-disk file size and then the system crashes before the | |
932 | * contents of the file are flushed to disk then the files | |
933 | * may be full of holes (ie NULL files bug). | |
934 | */ | |
935 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, | |
936 | XFS_ISIZE(ip)); | |
937 | if (error) { | |
938 | /* | |
939 | * If we get an error at this point we simply don't | |
940 | * bother truncating the file. | |
941 | */ | |
4906e215 | 942 | xfs_trans_cancel(tp); |
c24b5dfa | 943 | } else { |
70393313 | 944 | error = xfs_trans_commit(tp); |
c24b5dfa DC |
945 | if (!error) |
946 | xfs_inode_clear_eofblocks_tag(ip); | |
947 | } | |
948 | ||
949 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
950 | if (need_iolock) | |
951 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
952 | } | |
953 | return error; | |
954 | } | |
955 | ||
83aee9e4 | 956 | int |
c24b5dfa | 957 | xfs_alloc_file_space( |
83aee9e4 | 958 | struct xfs_inode *ip, |
c24b5dfa DC |
959 | xfs_off_t offset, |
960 | xfs_off_t len, | |
5f8aca8b | 961 | int alloc_type) |
c24b5dfa DC |
962 | { |
963 | xfs_mount_t *mp = ip->i_mount; | |
964 | xfs_off_t count; | |
965 | xfs_filblks_t allocated_fsb; | |
966 | xfs_filblks_t allocatesize_fsb; | |
967 | xfs_extlen_t extsz, temp; | |
968 | xfs_fileoff_t startoffset_fsb; | |
969 | xfs_fsblock_t firstfsb; | |
970 | int nimaps; | |
971 | int quota_flag; | |
972 | int rt; | |
973 | xfs_trans_t *tp; | |
974 | xfs_bmbt_irec_t imaps[1], *imapp; | |
975 | xfs_bmap_free_t free_list; | |
976 | uint qblocks, resblks, resrtextents; | |
c24b5dfa DC |
977 | int error; |
978 | ||
979 | trace_xfs_alloc_file_space(ip); | |
980 | ||
981 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 982 | return -EIO; |
c24b5dfa DC |
983 | |
984 | error = xfs_qm_dqattach(ip, 0); | |
985 | if (error) | |
986 | return error; | |
987 | ||
988 | if (len <= 0) | |
2451337d | 989 | return -EINVAL; |
c24b5dfa DC |
990 | |
991 | rt = XFS_IS_REALTIME_INODE(ip); | |
992 | extsz = xfs_get_extsz_hint(ip); | |
993 | ||
994 | count = len; | |
995 | imapp = &imaps[0]; | |
996 | nimaps = 1; | |
997 | startoffset_fsb = XFS_B_TO_FSBT(mp, offset); | |
998 | allocatesize_fsb = XFS_B_TO_FSB(mp, count); | |
999 | ||
1000 | /* | |
1001 | * Allocate file space until done or until there is an error | |
1002 | */ | |
1003 | while (allocatesize_fsb && !error) { | |
1004 | xfs_fileoff_t s, e; | |
1005 | ||
1006 | /* | |
1007 | * Determine space reservations for data/realtime. | |
1008 | */ | |
1009 | if (unlikely(extsz)) { | |
1010 | s = startoffset_fsb; | |
1011 | do_div(s, extsz); | |
1012 | s *= extsz; | |
1013 | e = startoffset_fsb + allocatesize_fsb; | |
1014 | if ((temp = do_mod(startoffset_fsb, extsz))) | |
1015 | e += temp; | |
1016 | if ((temp = do_mod(e, extsz))) | |
1017 | e += extsz - temp; | |
1018 | } else { | |
1019 | s = 0; | |
1020 | e = allocatesize_fsb; | |
1021 | } | |
1022 | ||
1023 | /* | |
1024 | * The transaction reservation is limited to a 32-bit block | |
1025 | * count, hence we need to limit the number of blocks we are | |
1026 | * trying to reserve to avoid an overflow. We can't allocate | |
1027 | * more than @nimaps extents, and an extent is limited on disk | |
1028 | * to MAXEXTLEN (21 bits), so use that to enforce the limit. | |
1029 | */ | |
1030 | resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps)); | |
1031 | if (unlikely(rt)) { | |
1032 | resrtextents = qblocks = resblks; | |
1033 | resrtextents /= mp->m_sb.sb_rextsize; | |
1034 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
1035 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
1036 | } else { | |
1037 | resrtextents = 0; | |
1038 | resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks); | |
1039 | quota_flag = XFS_QMOPT_RES_REGBLKS; | |
1040 | } | |
1041 | ||
1042 | /* | |
1043 | * Allocate and setup the transaction. | |
1044 | */ | |
253f4911 CH |
1045 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, |
1046 | resrtextents, 0, &tp); | |
1047 | ||
c24b5dfa DC |
1048 | /* |
1049 | * Check for running out of space | |
1050 | */ | |
1051 | if (error) { | |
1052 | /* | |
1053 | * Free the transaction structure. | |
1054 | */ | |
2451337d | 1055 | ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp)); |
c24b5dfa DC |
1056 | break; |
1057 | } | |
1058 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1059 | error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, | |
1060 | 0, quota_flag); | |
1061 | if (error) | |
1062 | goto error1; | |
1063 | ||
1064 | xfs_trans_ijoin(tp, ip, 0); | |
1065 | ||
1066 | xfs_bmap_init(&free_list, &firstfsb); | |
1067 | error = xfs_bmapi_write(tp, ip, startoffset_fsb, | |
1068 | allocatesize_fsb, alloc_type, &firstfsb, | |
dbd5c8c9 | 1069 | resblks, imapp, &nimaps, &free_list); |
f6106efa | 1070 | if (error) |
c24b5dfa | 1071 | goto error0; |
c24b5dfa DC |
1072 | |
1073 | /* | |
1074 | * Complete the transaction | |
1075 | */ | |
f6106efa ES |
1076 | error = xfs_bmap_finish(&tp, &free_list, NULL); |
1077 | if (error) | |
c24b5dfa | 1078 | goto error0; |
c24b5dfa | 1079 | |
70393313 | 1080 | error = xfs_trans_commit(tp); |
c24b5dfa | 1081 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
f6106efa | 1082 | if (error) |
c24b5dfa | 1083 | break; |
c24b5dfa DC |
1084 | |
1085 | allocated_fsb = imapp->br_blockcount; | |
1086 | ||
1087 | if (nimaps == 0) { | |
2451337d | 1088 | error = -ENOSPC; |
c24b5dfa DC |
1089 | break; |
1090 | } | |
1091 | ||
1092 | startoffset_fsb += allocated_fsb; | |
1093 | allocatesize_fsb -= allocated_fsb; | |
1094 | } | |
1095 | ||
1096 | return error; | |
1097 | ||
1098 | error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */ | |
1099 | xfs_bmap_cancel(&free_list); | |
1100 | xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag); | |
1101 | ||
1102 | error1: /* Just cancel transaction */ | |
4906e215 | 1103 | xfs_trans_cancel(tp); |
c24b5dfa DC |
1104 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1105 | return error; | |
1106 | } | |
1107 | ||
bdb0d04f CH |
1108 | static int |
1109 | xfs_unmap_extent( | |
1110 | struct xfs_inode *ip, | |
1111 | xfs_fileoff_t startoffset_fsb, | |
1112 | xfs_filblks_t len_fsb, | |
1113 | int *done) | |
c24b5dfa | 1114 | { |
bdb0d04f CH |
1115 | struct xfs_mount *mp = ip->i_mount; |
1116 | struct xfs_trans *tp; | |
1117 | struct xfs_bmap_free free_list; | |
1118 | xfs_fsblock_t firstfsb; | |
1119 | uint resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
1120 | int error; | |
c24b5dfa | 1121 | |
bdb0d04f CH |
1122 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp); |
1123 | if (error) { | |
1124 | ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
1125 | return error; | |
1126 | } | |
c24b5dfa | 1127 | |
bdb0d04f CH |
1128 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
1129 | error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot, ip->i_gdquot, | |
1130 | ip->i_pdquot, resblks, 0, XFS_QMOPT_RES_REGBLKS); | |
1131 | if (error) | |
1132 | goto out_trans_cancel; | |
c24b5dfa | 1133 | |
bdb0d04f | 1134 | xfs_trans_ijoin(tp, ip, 0); |
4f317369 | 1135 | |
bdb0d04f CH |
1136 | xfs_bmap_init(&free_list, &firstfsb); |
1137 | error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, &firstfsb, | |
1138 | &free_list, done); | |
1139 | if (error) | |
1140 | goto out_bmap_cancel; | |
4f317369 | 1141 | |
bdb0d04f CH |
1142 | error = xfs_bmap_finish(&tp, &free_list, NULL); |
1143 | if (error) | |
1144 | goto out_bmap_cancel; | |
4f317369 | 1145 | |
bdb0d04f CH |
1146 | error = xfs_trans_commit(tp); |
1147 | out_unlock: | |
1148 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1149 | return error; | |
4f69f578 | 1150 | |
bdb0d04f CH |
1151 | out_bmap_cancel: |
1152 | xfs_bmap_cancel(&free_list); | |
1153 | out_trans_cancel: | |
1154 | xfs_trans_cancel(tp); | |
1155 | goto out_unlock; | |
1156 | } | |
4f69f578 | 1157 | |
bdb0d04f CH |
1158 | static int |
1159 | xfs_adjust_extent_unmap_boundaries( | |
1160 | struct xfs_inode *ip, | |
1161 | xfs_fileoff_t *startoffset_fsb, | |
1162 | xfs_fileoff_t *endoffset_fsb) | |
1163 | { | |
1164 | struct xfs_mount *mp = ip->i_mount; | |
1165 | struct xfs_bmbt_irec imap; | |
1166 | int nimap, error; | |
1167 | xfs_extlen_t mod = 0; | |
4f69f578 | 1168 | |
bdb0d04f CH |
1169 | nimap = 1; |
1170 | error = xfs_bmapi_read(ip, *startoffset_fsb, 1, &imap, &nimap, 0); | |
1171 | if (error) | |
1172 | return error; | |
83a0adc3 | 1173 | |
bdb0d04f CH |
1174 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { |
1175 | xfs_daddr_t block; | |
8c156125 | 1176 | |
bdb0d04f CH |
1177 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); |
1178 | block = imap.br_startblock; | |
1179 | mod = do_div(block, mp->m_sb.sb_rextsize); | |
1180 | if (mod) | |
1181 | *startoffset_fsb += mp->m_sb.sb_rextsize - mod; | |
1182 | } | |
83a0adc3 | 1183 | |
bdb0d04f CH |
1184 | nimap = 1; |
1185 | error = xfs_bmapi_read(ip, *endoffset_fsb - 1, 1, &imap, &nimap, 0); | |
1186 | if (error) | |
1187 | return error; | |
1188 | ||
1189 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
1190 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
1191 | mod++; | |
1192 | if (mod && mod != mp->m_sb.sb_rextsize) | |
1193 | *endoffset_fsb -= mod; | |
c24b5dfa | 1194 | } |
bdb0d04f CH |
1195 | |
1196 | return 0; | |
1197 | } | |
1198 | ||
1199 | static int | |
1200 | xfs_flush_unmap_range( | |
1201 | struct xfs_inode *ip, | |
1202 | xfs_off_t offset, | |
1203 | xfs_off_t len) | |
1204 | { | |
1205 | struct xfs_mount *mp = ip->i_mount; | |
1206 | struct inode *inode = VFS_I(ip); | |
1207 | xfs_off_t rounding, start, end; | |
1208 | int error; | |
1209 | ||
1210 | /* wait for the completion of any pending DIOs */ | |
1211 | inode_dio_wait(inode); | |
1212 | ||
1213 | rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_SIZE); | |
1214 | start = round_down(offset, rounding); | |
1215 | end = round_up(offset + len, rounding) - 1; | |
1216 | ||
1217 | error = filemap_write_and_wait_range(inode->i_mapping, start, end); | |
1218 | if (error) | |
1219 | return error; | |
1220 | truncate_pagecache_range(inode, start, end); | |
1221 | return 0; | |
c24b5dfa DC |
1222 | } |
1223 | ||
83aee9e4 | 1224 | int |
c24b5dfa | 1225 | xfs_free_file_space( |
83aee9e4 | 1226 | struct xfs_inode *ip, |
c24b5dfa | 1227 | xfs_off_t offset, |
5f8aca8b | 1228 | xfs_off_t len) |
c24b5dfa | 1229 | { |
bdb0d04f | 1230 | struct xfs_mount *mp = ip->i_mount; |
c24b5dfa | 1231 | xfs_fileoff_t startoffset_fsb; |
bdb0d04f | 1232 | xfs_fileoff_t endoffset_fsb; |
3c2bdc91 | 1233 | int done = 0, error; |
c24b5dfa DC |
1234 | |
1235 | trace_xfs_free_file_space(ip); | |
1236 | ||
1237 | error = xfs_qm_dqattach(ip, 0); | |
1238 | if (error) | |
1239 | return error; | |
1240 | ||
c24b5dfa | 1241 | if (len <= 0) /* if nothing being freed */ |
bdb0d04f | 1242 | return 0; |
c24b5dfa | 1243 | |
bdb0d04f | 1244 | error = xfs_flush_unmap_range(ip, offset, len); |
c24b5dfa | 1245 | if (error) |
bdb0d04f CH |
1246 | return error; |
1247 | ||
1248 | startoffset_fsb = XFS_B_TO_FSB(mp, offset); | |
1249 | endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len); | |
c24b5dfa DC |
1250 | |
1251 | /* | |
bdb0d04f CH |
1252 | * Need to zero the stuff we're not freeing, on disk. If it's a RT file |
1253 | * and we can't use unwritten extents then we actually need to ensure | |
1254 | * to zero the whole extent, otherwise we just need to take of block | |
1255 | * boundaries, and xfs_bunmapi will handle the rest. | |
c24b5dfa | 1256 | */ |
bdb0d04f CH |
1257 | if (XFS_IS_REALTIME_INODE(ip) && |
1258 | !xfs_sb_version_hasextflgbit(&mp->m_sb)) { | |
1259 | error = xfs_adjust_extent_unmap_boundaries(ip, &startoffset_fsb, | |
1260 | &endoffset_fsb); | |
c24b5dfa | 1261 | if (error) |
bdb0d04f | 1262 | return error; |
c24b5dfa DC |
1263 | } |
1264 | ||
3c2bdc91 CH |
1265 | if (endoffset_fsb > startoffset_fsb) { |
1266 | while (!done) { | |
1267 | error = xfs_unmap_extent(ip, startoffset_fsb, | |
1268 | endoffset_fsb - startoffset_fsb, &done); | |
1269 | if (error) | |
1270 | return error; | |
c24b5dfa | 1271 | } |
c24b5dfa DC |
1272 | } |
1273 | ||
3c2bdc91 CH |
1274 | /* |
1275 | * Now that we've unmap all full blocks we'll have to zero out any | |
1276 | * partial block at the beginning and/or end. xfs_zero_range is | |
1277 | * smart enough to skip any holes, including those we just created. | |
1278 | */ | |
1279 | return xfs_zero_range(ip, offset, len, NULL); | |
c24b5dfa DC |
1280 | } |
1281 | ||
5d11fb4b BF |
1282 | /* |
1283 | * Preallocate and zero a range of a file. This mechanism has the allocation | |
1284 | * semantics of fallocate and in addition converts data in the range to zeroes. | |
1285 | */ | |
865e9446 | 1286 | int |
c24b5dfa DC |
1287 | xfs_zero_file_space( |
1288 | struct xfs_inode *ip, | |
1289 | xfs_off_t offset, | |
5f8aca8b | 1290 | xfs_off_t len) |
c24b5dfa DC |
1291 | { |
1292 | struct xfs_mount *mp = ip->i_mount; | |
5d11fb4b | 1293 | uint blksize; |
c24b5dfa DC |
1294 | int error; |
1295 | ||
897b73b6 DC |
1296 | trace_xfs_zero_file_space(ip); |
1297 | ||
5d11fb4b | 1298 | blksize = 1 << mp->m_sb.sb_blocklog; |
c24b5dfa DC |
1299 | |
1300 | /* | |
5d11fb4b BF |
1301 | * Punch a hole and prealloc the range. We use hole punch rather than |
1302 | * unwritten extent conversion for two reasons: | |
1303 | * | |
1304 | * 1.) Hole punch handles partial block zeroing for us. | |
1305 | * | |
1306 | * 2.) If prealloc returns ENOSPC, the file range is still zero-valued | |
1307 | * by virtue of the hole punch. | |
c24b5dfa | 1308 | */ |
5d11fb4b BF |
1309 | error = xfs_free_file_space(ip, offset, len); |
1310 | if (error) | |
1311 | goto out; | |
c24b5dfa | 1312 | |
5d11fb4b BF |
1313 | error = xfs_alloc_file_space(ip, round_down(offset, blksize), |
1314 | round_up(offset + len, blksize) - | |
1315 | round_down(offset, blksize), | |
1316 | XFS_BMAPI_PREALLOC); | |
5f8aca8b | 1317 | out: |
c24b5dfa DC |
1318 | return error; |
1319 | ||
1320 | } | |
1321 | ||
e1d8fb88 | 1322 | /* |
a904b1ca NJ |
1323 | * @next_fsb will keep track of the extent currently undergoing shift. |
1324 | * @stop_fsb will keep track of the extent at which we have to stop. | |
1325 | * If we are shifting left, we will start with block (offset + len) and | |
1326 | * shift each extent till last extent. | |
1327 | * If we are shifting right, we will start with last extent inside file space | |
1328 | * and continue until we reach the block corresponding to offset. | |
e1d8fb88 | 1329 | */ |
72c1a739 | 1330 | static int |
a904b1ca NJ |
1331 | xfs_shift_file_space( |
1332 | struct xfs_inode *ip, | |
1333 | xfs_off_t offset, | |
1334 | xfs_off_t len, | |
1335 | enum shift_direction direction) | |
e1d8fb88 NJ |
1336 | { |
1337 | int done = 0; | |
1338 | struct xfs_mount *mp = ip->i_mount; | |
1339 | struct xfs_trans *tp; | |
1340 | int error; | |
e1d8fb88 NJ |
1341 | struct xfs_bmap_free free_list; |
1342 | xfs_fsblock_t first_block; | |
a904b1ca | 1343 | xfs_fileoff_t stop_fsb; |
2c845f5a | 1344 | xfs_fileoff_t next_fsb; |
e1d8fb88 NJ |
1345 | xfs_fileoff_t shift_fsb; |
1346 | ||
a904b1ca | 1347 | ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT); |
e1d8fb88 | 1348 | |
a904b1ca NJ |
1349 | if (direction == SHIFT_LEFT) { |
1350 | next_fsb = XFS_B_TO_FSB(mp, offset + len); | |
1351 | stop_fsb = XFS_B_TO_FSB(mp, VFS_I(ip)->i_size); | |
1352 | } else { | |
1353 | /* | |
1354 | * If right shift, delegate the work of initialization of | |
1355 | * next_fsb to xfs_bmap_shift_extent as it has ilock held. | |
1356 | */ | |
1357 | next_fsb = NULLFSBLOCK; | |
1358 | stop_fsb = XFS_B_TO_FSB(mp, offset); | |
1359 | } | |
e1d8fb88 | 1360 | |
e1d8fb88 NJ |
1361 | shift_fsb = XFS_B_TO_FSB(mp, len); |
1362 | ||
f71721d0 BF |
1363 | /* |
1364 | * Trim eofblocks to avoid shifting uninitialized post-eof preallocation | |
1365 | * into the accessible region of the file. | |
1366 | */ | |
41b9d726 BF |
1367 | if (xfs_can_free_eofblocks(ip, true)) { |
1368 | error = xfs_free_eofblocks(mp, ip, false); | |
1369 | if (error) | |
1370 | return error; | |
1371 | } | |
1669a8ca | 1372 | |
f71721d0 BF |
1373 | /* |
1374 | * Writeback and invalidate cache for the remainder of the file as we're | |
a904b1ca | 1375 | * about to shift down every extent from offset to EOF. |
f71721d0 BF |
1376 | */ |
1377 | error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, | |
a904b1ca | 1378 | offset, -1); |
f71721d0 BF |
1379 | if (error) |
1380 | return error; | |
1381 | error = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, | |
09cbfeaf | 1382 | offset >> PAGE_SHIFT, -1); |
e1d8fb88 NJ |
1383 | if (error) |
1384 | return error; | |
1385 | ||
a904b1ca NJ |
1386 | /* |
1387 | * The extent shiting code works on extent granularity. So, if | |
1388 | * stop_fsb is not the starting block of extent, we need to split | |
1389 | * the extent at stop_fsb. | |
1390 | */ | |
1391 | if (direction == SHIFT_RIGHT) { | |
1392 | error = xfs_bmap_split_extent(ip, stop_fsb); | |
1393 | if (error) | |
1394 | return error; | |
1395 | } | |
1396 | ||
e1d8fb88 | 1397 | while (!error && !done) { |
e1d8fb88 NJ |
1398 | /* |
1399 | * We would need to reserve permanent block for transaction. | |
1400 | * This will come into picture when after shifting extent into | |
1401 | * hole we found that adjacent extents can be merged which | |
1402 | * may lead to freeing of a block during record update. | |
1403 | */ | |
253f4911 CH |
1404 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, |
1405 | XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp); | |
1406 | if (error) | |
e1d8fb88 | 1407 | break; |
e1d8fb88 NJ |
1408 | |
1409 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1410 | error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot, | |
1411 | ip->i_gdquot, ip->i_pdquot, | |
1412 | XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, | |
1413 | XFS_QMOPT_RES_REGBLKS); | |
1414 | if (error) | |
d4a97a04 | 1415 | goto out_trans_cancel; |
e1d8fb88 | 1416 | |
a904b1ca | 1417 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
e1d8fb88 NJ |
1418 | |
1419 | xfs_bmap_init(&free_list, &first_block); | |
1420 | ||
1421 | /* | |
1422 | * We are using the write transaction in which max 2 bmbt | |
1423 | * updates are allowed | |
1424 | */ | |
a904b1ca NJ |
1425 | error = xfs_bmap_shift_extents(tp, ip, &next_fsb, shift_fsb, |
1426 | &done, stop_fsb, &first_block, &free_list, | |
1427 | direction, XFS_BMAP_MAX_SHIFT_EXTENTS); | |
e1d8fb88 | 1428 | if (error) |
d4a97a04 | 1429 | goto out_bmap_cancel; |
e1d8fb88 | 1430 | |
f6106efa | 1431 | error = xfs_bmap_finish(&tp, &free_list, NULL); |
e1d8fb88 | 1432 | if (error) |
d4a97a04 | 1433 | goto out_bmap_cancel; |
e1d8fb88 | 1434 | |
70393313 | 1435 | error = xfs_trans_commit(tp); |
e1d8fb88 NJ |
1436 | } |
1437 | ||
1438 | return error; | |
1439 | ||
d4a97a04 BF |
1440 | out_bmap_cancel: |
1441 | xfs_bmap_cancel(&free_list); | |
1442 | out_trans_cancel: | |
4906e215 | 1443 | xfs_trans_cancel(tp); |
e1d8fb88 NJ |
1444 | return error; |
1445 | } | |
1446 | ||
a904b1ca NJ |
1447 | /* |
1448 | * xfs_collapse_file_space() | |
1449 | * This routine frees disk space and shift extent for the given file. | |
1450 | * The first thing we do is to free data blocks in the specified range | |
1451 | * by calling xfs_free_file_space(). It would also sync dirty data | |
1452 | * and invalidate page cache over the region on which collapse range | |
1453 | * is working. And Shift extent records to the left to cover a hole. | |
1454 | * RETURNS: | |
1455 | * 0 on success | |
1456 | * errno on error | |
1457 | * | |
1458 | */ | |
1459 | int | |
1460 | xfs_collapse_file_space( | |
1461 | struct xfs_inode *ip, | |
1462 | xfs_off_t offset, | |
1463 | xfs_off_t len) | |
1464 | { | |
1465 | int error; | |
1466 | ||
1467 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
1468 | trace_xfs_collapse_file_space(ip); | |
1469 | ||
1470 | error = xfs_free_file_space(ip, offset, len); | |
1471 | if (error) | |
1472 | return error; | |
1473 | ||
1474 | return xfs_shift_file_space(ip, offset, len, SHIFT_LEFT); | |
1475 | } | |
1476 | ||
1477 | /* | |
1478 | * xfs_insert_file_space() | |
1479 | * This routine create hole space by shifting extents for the given file. | |
1480 | * The first thing we do is to sync dirty data and invalidate page cache | |
1481 | * over the region on which insert range is working. And split an extent | |
1482 | * to two extents at given offset by calling xfs_bmap_split_extent. | |
1483 | * And shift all extent records which are laying between [offset, | |
1484 | * last allocated extent] to the right to reserve hole range. | |
1485 | * RETURNS: | |
1486 | * 0 on success | |
1487 | * errno on error | |
1488 | */ | |
1489 | int | |
1490 | xfs_insert_file_space( | |
1491 | struct xfs_inode *ip, | |
1492 | loff_t offset, | |
1493 | loff_t len) | |
1494 | { | |
1495 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
1496 | trace_xfs_insert_file_space(ip); | |
1497 | ||
1498 | return xfs_shift_file_space(ip, offset, len, SHIFT_RIGHT); | |
1499 | } | |
1500 | ||
a133d952 DC |
1501 | /* |
1502 | * We need to check that the format of the data fork in the temporary inode is | |
1503 | * valid for the target inode before doing the swap. This is not a problem with | |
1504 | * attr1 because of the fixed fork offset, but attr2 has a dynamically sized | |
1505 | * data fork depending on the space the attribute fork is taking so we can get | |
1506 | * invalid formats on the target inode. | |
1507 | * | |
1508 | * E.g. target has space for 7 extents in extent format, temp inode only has | |
1509 | * space for 6. If we defragment down to 7 extents, then the tmp format is a | |
1510 | * btree, but when swapped it needs to be in extent format. Hence we can't just | |
1511 | * blindly swap data forks on attr2 filesystems. | |
1512 | * | |
1513 | * Note that we check the swap in both directions so that we don't end up with | |
1514 | * a corrupt temporary inode, either. | |
1515 | * | |
1516 | * Note that fixing the way xfs_fsr sets up the attribute fork in the source | |
1517 | * inode will prevent this situation from occurring, so all we do here is | |
1518 | * reject and log the attempt. basically we are putting the responsibility on | |
1519 | * userspace to get this right. | |
1520 | */ | |
1521 | static int | |
1522 | xfs_swap_extents_check_format( | |
1523 | xfs_inode_t *ip, /* target inode */ | |
1524 | xfs_inode_t *tip) /* tmp inode */ | |
1525 | { | |
1526 | ||
1527 | /* Should never get a local format */ | |
1528 | if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL || | |
1529 | tip->i_d.di_format == XFS_DINODE_FMT_LOCAL) | |
2451337d | 1530 | return -EINVAL; |
a133d952 DC |
1531 | |
1532 | /* | |
1533 | * if the target inode has less extents that then temporary inode then | |
1534 | * why did userspace call us? | |
1535 | */ | |
1536 | if (ip->i_d.di_nextents < tip->i_d.di_nextents) | |
2451337d | 1537 | return -EINVAL; |
a133d952 DC |
1538 | |
1539 | /* | |
1540 | * if the target inode is in extent form and the temp inode is in btree | |
1541 | * form then we will end up with the target inode in the wrong format | |
1542 | * as we already know there are less extents in the temp inode. | |
1543 | */ | |
1544 | if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS && | |
1545 | tip->i_d.di_format == XFS_DINODE_FMT_BTREE) | |
2451337d | 1546 | return -EINVAL; |
a133d952 DC |
1547 | |
1548 | /* Check temp in extent form to max in target */ | |
1549 | if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS && | |
1550 | XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) > | |
1551 | XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)) | |
2451337d | 1552 | return -EINVAL; |
a133d952 DC |
1553 | |
1554 | /* Check target in extent form to max in temp */ | |
1555 | if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS && | |
1556 | XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) > | |
1557 | XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK)) | |
2451337d | 1558 | return -EINVAL; |
a133d952 DC |
1559 | |
1560 | /* | |
1561 | * If we are in a btree format, check that the temp root block will fit | |
1562 | * in the target and that it has enough extents to be in btree format | |
1563 | * in the target. | |
1564 | * | |
1565 | * Note that we have to be careful to allow btree->extent conversions | |
1566 | * (a common defrag case) which will occur when the temp inode is in | |
1567 | * extent format... | |
1568 | */ | |
1569 | if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) { | |
1570 | if (XFS_IFORK_BOFF(ip) && | |
1571 | XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip)) | |
2451337d | 1572 | return -EINVAL; |
a133d952 DC |
1573 | if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <= |
1574 | XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)) | |
2451337d | 1575 | return -EINVAL; |
a133d952 DC |
1576 | } |
1577 | ||
1578 | /* Reciprocal target->temp btree format checks */ | |
1579 | if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) { | |
1580 | if (XFS_IFORK_BOFF(tip) && | |
1581 | XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip)) | |
2451337d | 1582 | return -EINVAL; |
a133d952 DC |
1583 | if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <= |
1584 | XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK)) | |
2451337d | 1585 | return -EINVAL; |
a133d952 DC |
1586 | } |
1587 | ||
1588 | return 0; | |
1589 | } | |
1590 | ||
7abbb8f9 | 1591 | static int |
4ef897a2 DC |
1592 | xfs_swap_extent_flush( |
1593 | struct xfs_inode *ip) | |
1594 | { | |
1595 | int error; | |
1596 | ||
1597 | error = filemap_write_and_wait(VFS_I(ip)->i_mapping); | |
1598 | if (error) | |
1599 | return error; | |
1600 | truncate_pagecache_range(VFS_I(ip), 0, -1); | |
1601 | ||
1602 | /* Verify O_DIRECT for ftmp */ | |
1603 | if (VFS_I(ip)->i_mapping->nrpages) | |
1604 | return -EINVAL; | |
4ef897a2 DC |
1605 | return 0; |
1606 | } | |
1607 | ||
a133d952 DC |
1608 | int |
1609 | xfs_swap_extents( | |
1610 | xfs_inode_t *ip, /* target inode */ | |
1611 | xfs_inode_t *tip, /* tmp inode */ | |
1612 | xfs_swapext_t *sxp) | |
1613 | { | |
1614 | xfs_mount_t *mp = ip->i_mount; | |
1615 | xfs_trans_t *tp; | |
1616 | xfs_bstat_t *sbp = &sxp->sx_stat; | |
1617 | xfs_ifork_t *tempifp, *ifp, *tifp; | |
1618 | int src_log_flags, target_log_flags; | |
1619 | int error = 0; | |
1620 | int aforkblks = 0; | |
1621 | int taforkblks = 0; | |
1622 | __uint64_t tmp; | |
81217683 | 1623 | int lock_flags; |
a133d952 | 1624 | |
a133d952 DC |
1625 | tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL); |
1626 | if (!tempifp) { | |
2451337d | 1627 | error = -ENOMEM; |
a133d952 DC |
1628 | goto out; |
1629 | } | |
1630 | ||
1631 | /* | |
723cac48 DC |
1632 | * Lock the inodes against other IO, page faults and truncate to |
1633 | * begin with. Then we can ensure the inodes are flushed and have no | |
1634 | * page cache safely. Once we have done this we can take the ilocks and | |
1635 | * do the rest of the checks. | |
a133d952 | 1636 | */ |
723cac48 | 1637 | lock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; |
a133d952 | 1638 | xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL); |
723cac48 | 1639 | xfs_lock_two_inodes(ip, tip, XFS_MMAPLOCK_EXCL); |
a133d952 DC |
1640 | |
1641 | /* Verify that both files have the same format */ | |
c19b3b05 | 1642 | if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) { |
2451337d | 1643 | error = -EINVAL; |
a133d952 DC |
1644 | goto out_unlock; |
1645 | } | |
1646 | ||
1647 | /* Verify both files are either real-time or non-realtime */ | |
1648 | if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) { | |
2451337d | 1649 | error = -EINVAL; |
a133d952 DC |
1650 | goto out_unlock; |
1651 | } | |
1652 | ||
4ef897a2 DC |
1653 | error = xfs_swap_extent_flush(ip); |
1654 | if (error) | |
1655 | goto out_unlock; | |
1656 | error = xfs_swap_extent_flush(tip); | |
a133d952 DC |
1657 | if (error) |
1658 | goto out_unlock; | |
a133d952 | 1659 | |
253f4911 CH |
1660 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp); |
1661 | if (error) | |
a133d952 | 1662 | goto out_unlock; |
723cac48 DC |
1663 | |
1664 | /* | |
1665 | * Lock and join the inodes to the tansaction so that transaction commit | |
1666 | * or cancel will unlock the inodes from this point onwards. | |
1667 | */ | |
4ef897a2 DC |
1668 | xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL); |
1669 | lock_flags |= XFS_ILOCK_EXCL; | |
723cac48 DC |
1670 | xfs_trans_ijoin(tp, ip, lock_flags); |
1671 | xfs_trans_ijoin(tp, tip, lock_flags); | |
1672 | ||
a133d952 DC |
1673 | |
1674 | /* Verify all data are being swapped */ | |
1675 | if (sxp->sx_offset != 0 || | |
1676 | sxp->sx_length != ip->i_d.di_size || | |
1677 | sxp->sx_length != tip->i_d.di_size) { | |
2451337d | 1678 | error = -EFAULT; |
4ef897a2 | 1679 | goto out_trans_cancel; |
a133d952 DC |
1680 | } |
1681 | ||
1682 | trace_xfs_swap_extent_before(ip, 0); | |
1683 | trace_xfs_swap_extent_before(tip, 1); | |
1684 | ||
1685 | /* check inode formats now that data is flushed */ | |
1686 | error = xfs_swap_extents_check_format(ip, tip); | |
1687 | if (error) { | |
1688 | xfs_notice(mp, | |
1689 | "%s: inode 0x%llx format is incompatible for exchanging.", | |
1690 | __func__, ip->i_ino); | |
4ef897a2 | 1691 | goto out_trans_cancel; |
a133d952 DC |
1692 | } |
1693 | ||
1694 | /* | |
1695 | * Compare the current change & modify times with that | |
1696 | * passed in. If they differ, we abort this swap. | |
1697 | * This is the mechanism used to ensure the calling | |
1698 | * process that the file was not changed out from | |
1699 | * under it. | |
1700 | */ | |
1701 | if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) || | |
1702 | (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) || | |
1703 | (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) || | |
1704 | (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) { | |
2451337d | 1705 | error = -EBUSY; |
81217683 | 1706 | goto out_trans_cancel; |
a133d952 | 1707 | } |
a133d952 DC |
1708 | /* |
1709 | * Count the number of extended attribute blocks | |
1710 | */ | |
1711 | if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) && | |
1712 | (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) { | |
1713 | error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks); | |
1714 | if (error) | |
1715 | goto out_trans_cancel; | |
1716 | } | |
1717 | if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) && | |
1718 | (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) { | |
1719 | error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, | |
1720 | &taforkblks); | |
1721 | if (error) | |
1722 | goto out_trans_cancel; | |
1723 | } | |
1724 | ||
21b5c978 DC |
1725 | /* |
1726 | * Before we've swapped the forks, lets set the owners of the forks | |
1727 | * appropriately. We have to do this as we are demand paging the btree | |
1728 | * buffers, and so the validation done on read will expect the owner | |
1729 | * field to be correctly set. Once we change the owners, we can swap the | |
1730 | * inode forks. | |
1731 | * | |
1732 | * Note the trickiness in setting the log flags - we set the owner log | |
1733 | * flag on the opposite inode (i.e. the inode we are setting the new | |
1734 | * owner to be) because once we swap the forks and log that, log | |
1735 | * recovery is going to see the fork as owned by the swapped inode, | |
1736 | * not the pre-swapped inodes. | |
1737 | */ | |
1738 | src_log_flags = XFS_ILOG_CORE; | |
1739 | target_log_flags = XFS_ILOG_CORE; | |
1740 | if (ip->i_d.di_version == 3 && | |
1741 | ip->i_d.di_format == XFS_DINODE_FMT_BTREE) { | |
638f4416 DC |
1742 | target_log_flags |= XFS_ILOG_DOWNER; |
1743 | error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, | |
1744 | tip->i_ino, NULL); | |
21b5c978 DC |
1745 | if (error) |
1746 | goto out_trans_cancel; | |
1747 | } | |
1748 | ||
1749 | if (tip->i_d.di_version == 3 && | |
1750 | tip->i_d.di_format == XFS_DINODE_FMT_BTREE) { | |
638f4416 DC |
1751 | src_log_flags |= XFS_ILOG_DOWNER; |
1752 | error = xfs_bmbt_change_owner(tp, tip, XFS_DATA_FORK, | |
1753 | ip->i_ino, NULL); | |
21b5c978 DC |
1754 | if (error) |
1755 | goto out_trans_cancel; | |
1756 | } | |
1757 | ||
a133d952 DC |
1758 | /* |
1759 | * Swap the data forks of the inodes | |
1760 | */ | |
1761 | ifp = &ip->i_df; | |
1762 | tifp = &tip->i_df; | |
1763 | *tempifp = *ifp; /* struct copy */ | |
1764 | *ifp = *tifp; /* struct copy */ | |
1765 | *tifp = *tempifp; /* struct copy */ | |
1766 | ||
1767 | /* | |
1768 | * Fix the on-disk inode values | |
1769 | */ | |
1770 | tmp = (__uint64_t)ip->i_d.di_nblocks; | |
1771 | ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks; | |
1772 | tip->i_d.di_nblocks = tmp + taforkblks - aforkblks; | |
1773 | ||
1774 | tmp = (__uint64_t) ip->i_d.di_nextents; | |
1775 | ip->i_d.di_nextents = tip->i_d.di_nextents; | |
1776 | tip->i_d.di_nextents = tmp; | |
1777 | ||
1778 | tmp = (__uint64_t) ip->i_d.di_format; | |
1779 | ip->i_d.di_format = tip->i_d.di_format; | |
1780 | tip->i_d.di_format = tmp; | |
1781 | ||
1782 | /* | |
1783 | * The extents in the source inode could still contain speculative | |
1784 | * preallocation beyond EOF (e.g. the file is open but not modified | |
1785 | * while defrag is in progress). In that case, we need to copy over the | |
1786 | * number of delalloc blocks the data fork in the source inode is | |
1787 | * tracking beyond EOF so that when the fork is truncated away when the | |
1788 | * temporary inode is unlinked we don't underrun the i_delayed_blks | |
1789 | * counter on that inode. | |
1790 | */ | |
1791 | ASSERT(tip->i_delayed_blks == 0); | |
1792 | tip->i_delayed_blks = ip->i_delayed_blks; | |
1793 | ip->i_delayed_blks = 0; | |
1794 | ||
a133d952 DC |
1795 | switch (ip->i_d.di_format) { |
1796 | case XFS_DINODE_FMT_EXTENTS: | |
1797 | /* If the extents fit in the inode, fix the | |
1798 | * pointer. Otherwise it's already NULL or | |
1799 | * pointing to the extent. | |
1800 | */ | |
1801 | if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) { | |
1802 | ifp->if_u1.if_extents = | |
1803 | ifp->if_u2.if_inline_ext; | |
1804 | } | |
1805 | src_log_flags |= XFS_ILOG_DEXT; | |
1806 | break; | |
1807 | case XFS_DINODE_FMT_BTREE: | |
21b5c978 | 1808 | ASSERT(ip->i_d.di_version < 3 || |
638f4416 | 1809 | (src_log_flags & XFS_ILOG_DOWNER)); |
a133d952 DC |
1810 | src_log_flags |= XFS_ILOG_DBROOT; |
1811 | break; | |
1812 | } | |
1813 | ||
a133d952 DC |
1814 | switch (tip->i_d.di_format) { |
1815 | case XFS_DINODE_FMT_EXTENTS: | |
1816 | /* If the extents fit in the inode, fix the | |
1817 | * pointer. Otherwise it's already NULL or | |
1818 | * pointing to the extent. | |
1819 | */ | |
1820 | if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) { | |
1821 | tifp->if_u1.if_extents = | |
1822 | tifp->if_u2.if_inline_ext; | |
1823 | } | |
1824 | target_log_flags |= XFS_ILOG_DEXT; | |
1825 | break; | |
1826 | case XFS_DINODE_FMT_BTREE: | |
1827 | target_log_flags |= XFS_ILOG_DBROOT; | |
21b5c978 | 1828 | ASSERT(tip->i_d.di_version < 3 || |
638f4416 | 1829 | (target_log_flags & XFS_ILOG_DOWNER)); |
a133d952 DC |
1830 | break; |
1831 | } | |
1832 | ||
a133d952 DC |
1833 | xfs_trans_log_inode(tp, ip, src_log_flags); |
1834 | xfs_trans_log_inode(tp, tip, target_log_flags); | |
1835 | ||
1836 | /* | |
1837 | * If this is a synchronous mount, make sure that the | |
1838 | * transaction goes to disk before returning to the user. | |
1839 | */ | |
1840 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
1841 | xfs_trans_set_sync(tp); | |
1842 | ||
70393313 | 1843 | error = xfs_trans_commit(tp); |
a133d952 DC |
1844 | |
1845 | trace_xfs_swap_extent_after(ip, 0); | |
1846 | trace_xfs_swap_extent_after(tip, 1); | |
1847 | out: | |
1848 | kmem_free(tempifp); | |
1849 | return error; | |
1850 | ||
1851 | out_unlock: | |
81217683 DC |
1852 | xfs_iunlock(ip, lock_flags); |
1853 | xfs_iunlock(tip, lock_flags); | |
a133d952 DC |
1854 | goto out; |
1855 | ||
1856 | out_trans_cancel: | |
4906e215 | 1857 | xfs_trans_cancel(tp); |
723cac48 | 1858 | goto out; |
a133d952 | 1859 | } |