]>
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" |
3ab78df2 | 28 | #include "xfs_defer.h" |
68988114 DC |
29 | #include "xfs_inode.h" |
30 | #include "xfs_btree.h" | |
239880ef | 31 | #include "xfs_trans.h" |
68988114 DC |
32 | #include "xfs_extfree_item.h" |
33 | #include "xfs_alloc.h" | |
34 | #include "xfs_bmap.h" | |
35 | #include "xfs_bmap_util.h" | |
a4fbe6ab | 36 | #include "xfs_bmap_btree.h" |
68988114 DC |
37 | #include "xfs_rtalloc.h" |
38 | #include "xfs_error.h" | |
39 | #include "xfs_quota.h" | |
40 | #include "xfs_trans_space.h" | |
41 | #include "xfs_trace.h" | |
c24b5dfa | 42 | #include "xfs_icache.h" |
239880ef | 43 | #include "xfs_log.h" |
9c194644 | 44 | #include "xfs_rmap_btree.h" |
f86f4037 DW |
45 | #include "xfs_iomap.h" |
46 | #include "xfs_reflink.h" | |
47 | #include "xfs_refcount.h" | |
68988114 DC |
48 | |
49 | /* Kernel only BMAP related definitions and functions */ | |
50 | ||
51 | /* | |
52 | * Convert the given file system block to a disk block. We have to treat it | |
53 | * differently based on whether the file is a real time file or not, because the | |
54 | * bmap code does. | |
55 | */ | |
56 | xfs_daddr_t | |
57 | xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb) | |
58 | { | |
59 | return (XFS_IS_REALTIME_INODE(ip) ? \ | |
60 | (xfs_daddr_t)XFS_FSB_TO_BB((ip)->i_mount, (fsb)) : \ | |
61 | XFS_FSB_TO_DADDR((ip)->i_mount, (fsb))); | |
62 | } | |
63 | ||
3fbbbea3 DC |
64 | /* |
65 | * Routine to zero an extent on disk allocated to the specific inode. | |
66 | * | |
67 | * The VFS functions take a linearised filesystem block offset, so we have to | |
68 | * convert the sparse xfs fsb to the right format first. | |
69 | * VFS types are real funky, too. | |
70 | */ | |
71 | int | |
72 | xfs_zero_extent( | |
73 | struct xfs_inode *ip, | |
74 | xfs_fsblock_t start_fsb, | |
75 | xfs_off_t count_fsb) | |
76 | { | |
77 | struct xfs_mount *mp = ip->i_mount; | |
78 | xfs_daddr_t sector = xfs_fsb_to_db(ip, start_fsb); | |
79 | sector_t block = XFS_BB_TO_FSBT(mp, sector); | |
3fbbbea3 | 80 | |
3dc29161 MW |
81 | return blkdev_issue_zeroout(xfs_find_bdev_for_inode(VFS_I(ip)), |
82 | block << (mp->m_super->s_blocksize_bits - 9), | |
83 | count_fsb << (mp->m_super->s_blocksize_bits - 9), | |
84 | GFP_NOFS, true); | |
3fbbbea3 DC |
85 | } |
86 | ||
68988114 DC |
87 | int |
88 | xfs_bmap_rtalloc( | |
89 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ | |
90 | { | |
68988114 DC |
91 | int error; /* error return value */ |
92 | xfs_mount_t *mp; /* mount point structure */ | |
93 | xfs_extlen_t prod = 0; /* product factor for allocators */ | |
94 | xfs_extlen_t ralen = 0; /* realtime allocation length */ | |
95 | xfs_extlen_t align; /* minimum allocation alignment */ | |
96 | xfs_rtblock_t rtb; | |
97 | ||
98 | mp = ap->ip->i_mount; | |
99 | align = xfs_get_extsz_hint(ap->ip); | |
100 | prod = align / mp->m_sb.sb_rextsize; | |
101 | error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, | |
102 | align, 1, ap->eof, 0, | |
103 | ap->conv, &ap->offset, &ap->length); | |
104 | if (error) | |
105 | return error; | |
106 | ASSERT(ap->length); | |
107 | ASSERT(ap->length % mp->m_sb.sb_rextsize == 0); | |
108 | ||
109 | /* | |
110 | * If the offset & length are not perfectly aligned | |
111 | * then kill prod, it will just get us in trouble. | |
112 | */ | |
113 | if (do_mod(ap->offset, align) || ap->length % align) | |
114 | prod = 1; | |
115 | /* | |
116 | * Set ralen to be the actual requested length in rtextents. | |
117 | */ | |
118 | ralen = ap->length / mp->m_sb.sb_rextsize; | |
119 | /* | |
120 | * If the old value was close enough to MAXEXTLEN that | |
121 | * we rounded up to it, cut it back so it's valid again. | |
122 | * Note that if it's a really large request (bigger than | |
123 | * MAXEXTLEN), we don't hear about that number, and can't | |
124 | * adjust the starting point to match it. | |
125 | */ | |
126 | if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN) | |
127 | ralen = MAXEXTLEN / mp->m_sb.sb_rextsize; | |
128 | ||
129 | /* | |
4b680afb | 130 | * Lock out modifications to both the RT bitmap and summary inodes |
68988114 | 131 | */ |
f4a0660d | 132 | xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP); |
68988114 | 133 | xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL); |
f4a0660d | 134 | xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM); |
4b680afb | 135 | xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL); |
68988114 DC |
136 | |
137 | /* | |
138 | * If it's an allocation to an empty file at offset 0, | |
139 | * pick an extent that will space things out in the rt area. | |
140 | */ | |
141 | if (ap->eof && ap->offset == 0) { | |
142 | xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */ | |
143 | ||
144 | error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx); | |
145 | if (error) | |
146 | return error; | |
147 | ap->blkno = rtx * mp->m_sb.sb_rextsize; | |
148 | } else { | |
149 | ap->blkno = 0; | |
150 | } | |
151 | ||
152 | xfs_bmap_adjacent(ap); | |
153 | ||
154 | /* | |
155 | * Realtime allocation, done through xfs_rtallocate_extent. | |
156 | */ | |
68988114 DC |
157 | do_div(ap->blkno, mp->m_sb.sb_rextsize); |
158 | rtb = ap->blkno; | |
159 | ap->length = ralen; | |
089ec2f8 CH |
160 | error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length, |
161 | &ralen, ap->wasdel, prod, &rtb); | |
162 | if (error) | |
68988114 | 163 | return error; |
089ec2f8 | 164 | |
68988114 DC |
165 | ap->blkno = rtb; |
166 | if (ap->blkno != NULLFSBLOCK) { | |
167 | ap->blkno *= mp->m_sb.sb_rextsize; | |
168 | ralen *= mp->m_sb.sb_rextsize; | |
169 | ap->length = ralen; | |
170 | ap->ip->i_d.di_nblocks += ralen; | |
171 | xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE); | |
172 | if (ap->wasdel) | |
173 | ap->ip->i_delayed_blks -= ralen; | |
174 | /* | |
175 | * Adjust the disk quota also. This was reserved | |
176 | * earlier. | |
177 | */ | |
178 | xfs_trans_mod_dquot_byino(ap->tp, ap->ip, | |
179 | ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT : | |
180 | XFS_TRANS_DQ_RTBCOUNT, (long) ralen); | |
3fbbbea3 DC |
181 | |
182 | /* Zero the extent if we were asked to do so */ | |
292378ed | 183 | if (ap->datatype & XFS_ALLOC_USERDATA_ZERO) { |
3fbbbea3 DC |
184 | error = xfs_zero_extent(ap->ip, ap->blkno, ap->length); |
185 | if (error) | |
186 | return error; | |
187 | } | |
68988114 DC |
188 | } else { |
189 | ap->length = 0; | |
190 | } | |
191 | return 0; | |
192 | } | |
193 | ||
68988114 DC |
194 | /* |
195 | * Check if the endoff is outside the last extent. If so the caller will grow | |
196 | * the allocation to a stripe unit boundary. All offsets are considered outside | |
197 | * the end of file for an empty fork, so 1 is returned in *eof in that case. | |
198 | */ | |
199 | int | |
200 | xfs_bmap_eof( | |
201 | struct xfs_inode *ip, | |
202 | xfs_fileoff_t endoff, | |
203 | int whichfork, | |
204 | int *eof) | |
205 | { | |
206 | struct xfs_bmbt_irec rec; | |
207 | int error; | |
208 | ||
209 | error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof); | |
210 | if (error || *eof) | |
211 | return error; | |
212 | ||
213 | *eof = endoff >= rec.br_startoff + rec.br_blockcount; | |
214 | return 0; | |
215 | } | |
216 | ||
217 | /* | |
218 | * Extent tree block counting routines. | |
219 | */ | |
220 | ||
221 | /* | |
222 | * Count leaf blocks given a range of extent records. | |
223 | */ | |
224 | STATIC void | |
225 | xfs_bmap_count_leaves( | |
226 | xfs_ifork_t *ifp, | |
227 | xfs_extnum_t idx, | |
228 | int numrecs, | |
229 | int *count) | |
230 | { | |
231 | int b; | |
232 | ||
233 | for (b = 0; b < numrecs; b++) { | |
234 | xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b); | |
235 | *count += xfs_bmbt_get_blockcount(frp); | |
236 | } | |
237 | } | |
238 | ||
239 | /* | |
240 | * Count leaf blocks given a range of extent records originally | |
241 | * in btree format. | |
242 | */ | |
243 | STATIC void | |
244 | xfs_bmap_disk_count_leaves( | |
245 | struct xfs_mount *mp, | |
246 | struct xfs_btree_block *block, | |
247 | int numrecs, | |
248 | int *count) | |
249 | { | |
250 | int b; | |
251 | xfs_bmbt_rec_t *frp; | |
252 | ||
253 | for (b = 1; b <= numrecs; b++) { | |
254 | frp = XFS_BMBT_REC_ADDR(mp, block, b); | |
255 | *count += xfs_bmbt_disk_get_blockcount(frp); | |
256 | } | |
257 | } | |
258 | ||
259 | /* | |
260 | * Recursively walks each level of a btree | |
8be11e92 | 261 | * to count total fsblocks in use. |
68988114 DC |
262 | */ |
263 | STATIC int /* error */ | |
264 | xfs_bmap_count_tree( | |
265 | xfs_mount_t *mp, /* file system mount point */ | |
266 | xfs_trans_t *tp, /* transaction pointer */ | |
267 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
268 | xfs_fsblock_t blockno, /* file system block number */ | |
269 | int levelin, /* level in btree */ | |
270 | int *count) /* Count of blocks */ | |
271 | { | |
272 | int error; | |
273 | xfs_buf_t *bp, *nbp; | |
274 | int level = levelin; | |
275 | __be64 *pp; | |
276 | xfs_fsblock_t bno = blockno; | |
277 | xfs_fsblock_t nextbno; | |
278 | struct xfs_btree_block *block, *nextblock; | |
279 | int numrecs; | |
280 | ||
281 | error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF, | |
282 | &xfs_bmbt_buf_ops); | |
283 | if (error) | |
284 | return error; | |
285 | *count += 1; | |
286 | block = XFS_BUF_TO_BLOCK(bp); | |
287 | ||
288 | if (--level) { | |
289 | /* Not at node above leaves, count this level of nodes */ | |
290 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); | |
291 | while (nextbno != NULLFSBLOCK) { | |
292 | error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp, | |
293 | XFS_BMAP_BTREE_REF, | |
294 | &xfs_bmbt_buf_ops); | |
295 | if (error) | |
296 | return error; | |
297 | *count += 1; | |
298 | nextblock = XFS_BUF_TO_BLOCK(nbp); | |
299 | nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib); | |
300 | xfs_trans_brelse(tp, nbp); | |
301 | } | |
302 | ||
303 | /* Dive to the next level */ | |
304 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]); | |
305 | bno = be64_to_cpu(*pp); | |
306 | if (unlikely((error = | |
307 | xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) { | |
308 | xfs_trans_brelse(tp, bp); | |
309 | XFS_ERROR_REPORT("xfs_bmap_count_tree(1)", | |
310 | XFS_ERRLEVEL_LOW, mp); | |
2451337d | 311 | return -EFSCORRUPTED; |
68988114 DC |
312 | } |
313 | xfs_trans_brelse(tp, bp); | |
314 | } else { | |
315 | /* count all level 1 nodes and their leaves */ | |
316 | for (;;) { | |
317 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); | |
318 | numrecs = be16_to_cpu(block->bb_numrecs); | |
319 | xfs_bmap_disk_count_leaves(mp, block, numrecs, count); | |
320 | xfs_trans_brelse(tp, bp); | |
321 | if (nextbno == NULLFSBLOCK) | |
322 | break; | |
323 | bno = nextbno; | |
324 | error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, | |
325 | XFS_BMAP_BTREE_REF, | |
326 | &xfs_bmbt_buf_ops); | |
327 | if (error) | |
328 | return error; | |
329 | *count += 1; | |
330 | block = XFS_BUF_TO_BLOCK(bp); | |
331 | } | |
332 | } | |
333 | return 0; | |
334 | } | |
335 | ||
336 | /* | |
337 | * Count fsblocks of the given fork. | |
338 | */ | |
0d5a75e9 | 339 | static int /* error */ |
68988114 DC |
340 | xfs_bmap_count_blocks( |
341 | xfs_trans_t *tp, /* transaction pointer */ | |
342 | xfs_inode_t *ip, /* incore inode */ | |
343 | int whichfork, /* data or attr fork */ | |
344 | int *count) /* out: count of blocks */ | |
345 | { | |
346 | struct xfs_btree_block *block; /* current btree block */ | |
347 | xfs_fsblock_t bno; /* block # of "block" */ | |
348 | xfs_ifork_t *ifp; /* fork structure */ | |
349 | int level; /* btree level, for checking */ | |
350 | xfs_mount_t *mp; /* file system mount structure */ | |
351 | __be64 *pp; /* pointer to block address */ | |
352 | ||
353 | bno = NULLFSBLOCK; | |
354 | mp = ip->i_mount; | |
355 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
356 | if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) { | |
5d829300 | 357 | xfs_bmap_count_leaves(ifp, 0, xfs_iext_count(ifp), count); |
68988114 DC |
358 | return 0; |
359 | } | |
360 | ||
361 | /* | |
362 | * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out. | |
363 | */ | |
364 | block = ifp->if_broot; | |
365 | level = be16_to_cpu(block->bb_level); | |
366 | ASSERT(level > 0); | |
367 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes); | |
368 | bno = be64_to_cpu(*pp); | |
d5cf09ba | 369 | ASSERT(bno != NULLFSBLOCK); |
68988114 DC |
370 | ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount); |
371 | ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks); | |
372 | ||
373 | if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) { | |
374 | XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW, | |
375 | mp); | |
2451337d | 376 | return -EFSCORRUPTED; |
68988114 DC |
377 | } |
378 | ||
379 | return 0; | |
380 | } | |
381 | ||
382 | /* | |
383 | * returns 1 for success, 0 if we failed to map the extent. | |
384 | */ | |
385 | STATIC int | |
386 | xfs_getbmapx_fix_eof_hole( | |
387 | xfs_inode_t *ip, /* xfs incore inode pointer */ | |
f86f4037 | 388 | int whichfork, |
68988114 DC |
389 | struct getbmapx *out, /* output structure */ |
390 | int prealloced, /* this is a file with | |
391 | * preallocated data space */ | |
392 | __int64_t end, /* last block requested */ | |
f86f4037 DW |
393 | xfs_fsblock_t startblock, |
394 | bool moretocome) | |
68988114 DC |
395 | { |
396 | __int64_t fixlen; | |
397 | xfs_mount_t *mp; /* file system mount point */ | |
398 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
399 | xfs_extnum_t lastx; /* last extent pointer */ | |
400 | xfs_fileoff_t fileblock; | |
401 | ||
402 | if (startblock == HOLESTARTBLOCK) { | |
403 | mp = ip->i_mount; | |
404 | out->bmv_block = -1; | |
405 | fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip))); | |
406 | fixlen -= out->bmv_offset; | |
407 | if (prealloced && out->bmv_offset + out->bmv_length == end) { | |
408 | /* Came to hole at EOF. Trim it. */ | |
409 | if (fixlen <= 0) | |
410 | return 0; | |
411 | out->bmv_length = fixlen; | |
412 | } | |
413 | } else { | |
414 | if (startblock == DELAYSTARTBLOCK) | |
415 | out->bmv_block = -2; | |
416 | else | |
417 | out->bmv_block = xfs_fsb_to_db(ip, startblock); | |
418 | fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset); | |
f86f4037 DW |
419 | ifp = XFS_IFORK_PTR(ip, whichfork); |
420 | if (!moretocome && | |
421 | xfs_iext_bno_to_ext(ifp, fileblock, &lastx) && | |
5d829300 | 422 | (lastx == xfs_iext_count(ifp) - 1)) |
68988114 DC |
423 | out->bmv_oflags |= BMV_OF_LAST; |
424 | } | |
425 | ||
426 | return 1; | |
427 | } | |
428 | ||
f86f4037 DW |
429 | /* Adjust the reported bmap around shared/unshared extent transitions. */ |
430 | STATIC int | |
431 | xfs_getbmap_adjust_shared( | |
432 | struct xfs_inode *ip, | |
433 | int whichfork, | |
434 | struct xfs_bmbt_irec *map, | |
435 | struct getbmapx *out, | |
436 | struct xfs_bmbt_irec *next_map) | |
437 | { | |
438 | struct xfs_mount *mp = ip->i_mount; | |
439 | xfs_agnumber_t agno; | |
440 | xfs_agblock_t agbno; | |
441 | xfs_agblock_t ebno; | |
442 | xfs_extlen_t elen; | |
443 | xfs_extlen_t nlen; | |
444 | int error; | |
445 | ||
446 | next_map->br_startblock = NULLFSBLOCK; | |
447 | next_map->br_startoff = NULLFILEOFF; | |
448 | next_map->br_blockcount = 0; | |
449 | ||
450 | /* Only written data blocks can be shared. */ | |
451 | if (!xfs_is_reflink_inode(ip) || whichfork != XFS_DATA_FORK || | |
452 | map->br_startblock == DELAYSTARTBLOCK || | |
453 | map->br_startblock == HOLESTARTBLOCK || | |
454 | ISUNWRITTEN(map)) | |
455 | return 0; | |
456 | ||
457 | agno = XFS_FSB_TO_AGNO(mp, map->br_startblock); | |
458 | agbno = XFS_FSB_TO_AGBNO(mp, map->br_startblock); | |
459 | error = xfs_reflink_find_shared(mp, agno, agbno, map->br_blockcount, | |
460 | &ebno, &elen, true); | |
461 | if (error) | |
462 | return error; | |
463 | ||
464 | if (ebno == NULLAGBLOCK) { | |
465 | /* No shared blocks at all. */ | |
466 | return 0; | |
467 | } else if (agbno == ebno) { | |
468 | /* | |
469 | * Shared extent at (agbno, elen). Shrink the reported | |
470 | * extent length and prepare to move the start of map[i] | |
471 | * to agbno+elen, with the aim of (re)formatting the new | |
472 | * map[i] the next time through the inner loop. | |
473 | */ | |
474 | out->bmv_length = XFS_FSB_TO_BB(mp, elen); | |
475 | out->bmv_oflags |= BMV_OF_SHARED; | |
476 | if (elen != map->br_blockcount) { | |
477 | *next_map = *map; | |
478 | next_map->br_startblock += elen; | |
479 | next_map->br_startoff += elen; | |
480 | next_map->br_blockcount -= elen; | |
481 | } | |
482 | map->br_blockcount -= elen; | |
483 | } else { | |
484 | /* | |
485 | * There's an unshared extent (agbno, ebno - agbno) | |
486 | * followed by shared extent at (ebno, elen). Shrink | |
487 | * the reported extent length to cover only the unshared | |
488 | * extent and prepare to move up the start of map[i] to | |
489 | * ebno, with the aim of (re)formatting the new map[i] | |
490 | * the next time through the inner loop. | |
491 | */ | |
492 | *next_map = *map; | |
493 | nlen = ebno - agbno; | |
494 | out->bmv_length = XFS_FSB_TO_BB(mp, nlen); | |
495 | next_map->br_startblock += nlen; | |
496 | next_map->br_startoff += nlen; | |
497 | next_map->br_blockcount -= nlen; | |
498 | map->br_blockcount -= nlen; | |
499 | } | |
500 | ||
501 | return 0; | |
502 | } | |
503 | ||
68988114 DC |
504 | /* |
505 | * Get inode's extents as described in bmv, and format for output. | |
506 | * Calls formatter to fill the user's buffer until all extents | |
507 | * are mapped, until the passed-in bmv->bmv_count slots have | |
508 | * been filled, or until the formatter short-circuits the loop, | |
509 | * if it is tracking filled-in extents on its own. | |
510 | */ | |
511 | int /* error code */ | |
512 | xfs_getbmap( | |
513 | xfs_inode_t *ip, | |
514 | struct getbmapx *bmv, /* user bmap structure */ | |
515 | xfs_bmap_format_t formatter, /* format to user */ | |
516 | void *arg) /* formatter arg */ | |
517 | { | |
518 | __int64_t bmvend; /* last block requested */ | |
519 | int error = 0; /* return value */ | |
520 | __int64_t fixlen; /* length for -1 case */ | |
521 | int i; /* extent number */ | |
522 | int lock; /* lock state */ | |
523 | xfs_bmbt_irec_t *map; /* buffer for user's data */ | |
524 | xfs_mount_t *mp; /* file system mount point */ | |
525 | int nex; /* # of user extents can do */ | |
68988114 DC |
526 | int subnex; /* # of bmapi's can do */ |
527 | int nmap; /* number of map entries */ | |
528 | struct getbmapx *out; /* output structure */ | |
529 | int whichfork; /* data or attr fork */ | |
530 | int prealloced; /* this is a file with | |
531 | * preallocated data space */ | |
532 | int iflags; /* interface flags */ | |
533 | int bmapi_flags; /* flags for xfs_bmapi */ | |
534 | int cur_ext = 0; | |
f86f4037 | 535 | struct xfs_bmbt_irec inject_map; |
68988114 DC |
536 | |
537 | mp = ip->i_mount; | |
538 | iflags = bmv->bmv_iflags; | |
68988114 | 539 | |
f86f4037 DW |
540 | #ifndef DEBUG |
541 | /* Only allow CoW fork queries if we're debugging. */ | |
542 | if (iflags & BMV_IF_COWFORK) | |
543 | return -EINVAL; | |
544 | #endif | |
545 | if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK)) | |
546 | return -EINVAL; | |
547 | ||
548 | if (iflags & BMV_IF_ATTRFORK) | |
549 | whichfork = XFS_ATTR_FORK; | |
550 | else if (iflags & BMV_IF_COWFORK) | |
551 | whichfork = XFS_COW_FORK; | |
552 | else | |
553 | whichfork = XFS_DATA_FORK; | |
554 | ||
555 | switch (whichfork) { | |
556 | case XFS_ATTR_FORK: | |
68988114 DC |
557 | if (XFS_IFORK_Q(ip)) { |
558 | if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS && | |
559 | ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE && | |
560 | ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL) | |
2451337d | 561 | return -EINVAL; |
68988114 DC |
562 | } else if (unlikely( |
563 | ip->i_d.di_aformat != 0 && | |
564 | ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) { | |
565 | XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW, | |
566 | ip->i_mount); | |
2451337d | 567 | return -EFSCORRUPTED; |
68988114 DC |
568 | } |
569 | ||
570 | prealloced = 0; | |
571 | fixlen = 1LL << 32; | |
f86f4037 DW |
572 | break; |
573 | case XFS_COW_FORK: | |
574 | if (ip->i_cformat != XFS_DINODE_FMT_EXTENTS) | |
575 | return -EINVAL; | |
576 | ||
f7ca3522 DW |
577 | if (xfs_get_cowextsz_hint(ip)) { |
578 | prealloced = 1; | |
579 | fixlen = mp->m_super->s_maxbytes; | |
580 | } else { | |
581 | prealloced = 0; | |
582 | fixlen = XFS_ISIZE(ip); | |
583 | } | |
f86f4037 DW |
584 | break; |
585 | default: | |
68988114 DC |
586 | if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS && |
587 | ip->i_d.di_format != XFS_DINODE_FMT_BTREE && | |
588 | ip->i_d.di_format != XFS_DINODE_FMT_LOCAL) | |
2451337d | 589 | return -EINVAL; |
68988114 DC |
590 | |
591 | if (xfs_get_extsz_hint(ip) || | |
592 | ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){ | |
593 | prealloced = 1; | |
594 | fixlen = mp->m_super->s_maxbytes; | |
595 | } else { | |
596 | prealloced = 0; | |
597 | fixlen = XFS_ISIZE(ip); | |
598 | } | |
f86f4037 | 599 | break; |
68988114 DC |
600 | } |
601 | ||
602 | if (bmv->bmv_length == -1) { | |
603 | fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen)); | |
604 | bmv->bmv_length = | |
605 | max_t(__int64_t, fixlen - bmv->bmv_offset, 0); | |
606 | } else if (bmv->bmv_length == 0) { | |
607 | bmv->bmv_entries = 0; | |
608 | return 0; | |
609 | } else if (bmv->bmv_length < 0) { | |
2451337d | 610 | return -EINVAL; |
68988114 DC |
611 | } |
612 | ||
613 | nex = bmv->bmv_count - 1; | |
614 | if (nex <= 0) | |
2451337d | 615 | return -EINVAL; |
68988114 DC |
616 | bmvend = bmv->bmv_offset + bmv->bmv_length; |
617 | ||
618 | ||
619 | if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx)) | |
2451337d | 620 | return -ENOMEM; |
fdd3ccee DC |
621 | out = kmem_zalloc_large(bmv->bmv_count * sizeof(struct getbmapx), 0); |
622 | if (!out) | |
2451337d | 623 | return -ENOMEM; |
68988114 DC |
624 | |
625 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
f86f4037 DW |
626 | switch (whichfork) { |
627 | case XFS_DATA_FORK: | |
efa70be1 CH |
628 | if (!(iflags & BMV_IF_DELALLOC) && |
629 | (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) { | |
2451337d | 630 | error = filemap_write_and_wait(VFS_I(ip)->i_mapping); |
68988114 DC |
631 | if (error) |
632 | goto out_unlock_iolock; | |
efa70be1 CH |
633 | |
634 | /* | |
635 | * Even after flushing the inode, there can still be | |
636 | * delalloc blocks on the inode beyond EOF due to | |
637 | * speculative preallocation. These are not removed | |
638 | * until the release function is called or the inode | |
639 | * is inactivated. Hence we cannot assert here that | |
640 | * ip->i_delayed_blks == 0. | |
641 | */ | |
68988114 | 642 | } |
68988114 | 643 | |
efa70be1 | 644 | lock = xfs_ilock_data_map_shared(ip); |
f86f4037 DW |
645 | break; |
646 | case XFS_COW_FORK: | |
647 | lock = XFS_ILOCK_SHARED; | |
648 | xfs_ilock(ip, lock); | |
649 | break; | |
650 | case XFS_ATTR_FORK: | |
efa70be1 | 651 | lock = xfs_ilock_attr_map_shared(ip); |
f86f4037 | 652 | break; |
efa70be1 | 653 | } |
68988114 DC |
654 | |
655 | /* | |
656 | * Don't let nex be bigger than the number of extents | |
657 | * we can have assuming alternating holes and real extents. | |
658 | */ | |
659 | if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1) | |
660 | nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1; | |
661 | ||
662 | bmapi_flags = xfs_bmapi_aflag(whichfork); | |
663 | if (!(iflags & BMV_IF_PREALLOC)) | |
664 | bmapi_flags |= XFS_BMAPI_IGSTATE; | |
665 | ||
666 | /* | |
667 | * Allocate enough space to handle "subnex" maps at a time. | |
668 | */ | |
2451337d | 669 | error = -ENOMEM; |
68988114 DC |
670 | subnex = 16; |
671 | map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS); | |
672 | if (!map) | |
673 | goto out_unlock_ilock; | |
674 | ||
675 | bmv->bmv_entries = 0; | |
676 | ||
677 | if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 && | |
678 | (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) { | |
679 | error = 0; | |
680 | goto out_free_map; | |
681 | } | |
682 | ||
68988114 | 683 | do { |
c364b6d0 | 684 | nmap = (nex> subnex) ? subnex : nex; |
68988114 DC |
685 | error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset), |
686 | XFS_BB_TO_FSB(mp, bmv->bmv_length), | |
687 | map, &nmap, bmapi_flags); | |
688 | if (error) | |
689 | goto out_free_map; | |
690 | ASSERT(nmap <= subnex); | |
691 | ||
c364b6d0 DW |
692 | for (i = 0; i < nmap && bmv->bmv_length && |
693 | cur_ext < bmv->bmv_count - 1; i++) { | |
68988114 DC |
694 | out[cur_ext].bmv_oflags = 0; |
695 | if (map[i].br_state == XFS_EXT_UNWRITTEN) | |
696 | out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC; | |
697 | else if (map[i].br_startblock == DELAYSTARTBLOCK) | |
698 | out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC; | |
699 | out[cur_ext].bmv_offset = | |
700 | XFS_FSB_TO_BB(mp, map[i].br_startoff); | |
701 | out[cur_ext].bmv_length = | |
702 | XFS_FSB_TO_BB(mp, map[i].br_blockcount); | |
703 | out[cur_ext].bmv_unused1 = 0; | |
704 | out[cur_ext].bmv_unused2 = 0; | |
705 | ||
706 | /* | |
707 | * delayed allocation extents that start beyond EOF can | |
708 | * occur due to speculative EOF allocation when the | |
709 | * delalloc extent is larger than the largest freespace | |
710 | * extent at conversion time. These extents cannot be | |
711 | * converted by data writeback, so can exist here even | |
712 | * if we are not supposed to be finding delalloc | |
713 | * extents. | |
714 | */ | |
715 | if (map[i].br_startblock == DELAYSTARTBLOCK && | |
716 | map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip))) | |
717 | ASSERT((iflags & BMV_IF_DELALLOC) != 0); | |
718 | ||
719 | if (map[i].br_startblock == HOLESTARTBLOCK && | |
720 | whichfork == XFS_ATTR_FORK) { | |
721 | /* came to the end of attribute fork */ | |
722 | out[cur_ext].bmv_oflags |= BMV_OF_LAST; | |
723 | goto out_free_map; | |
724 | } | |
725 | ||
f86f4037 DW |
726 | /* Is this a shared block? */ |
727 | error = xfs_getbmap_adjust_shared(ip, whichfork, | |
728 | &map[i], &out[cur_ext], &inject_map); | |
729 | if (error) | |
730 | goto out_free_map; | |
731 | ||
732 | if (!xfs_getbmapx_fix_eof_hole(ip, whichfork, | |
733 | &out[cur_ext], prealloced, bmvend, | |
734 | map[i].br_startblock, | |
735 | inject_map.br_startblock != NULLFSBLOCK)) | |
68988114 DC |
736 | goto out_free_map; |
737 | ||
738 | bmv->bmv_offset = | |
739 | out[cur_ext].bmv_offset + | |
740 | out[cur_ext].bmv_length; | |
741 | bmv->bmv_length = | |
742 | max_t(__int64_t, 0, bmvend - bmv->bmv_offset); | |
743 | ||
744 | /* | |
745 | * In case we don't want to return the hole, | |
746 | * don't increase cur_ext so that we can reuse | |
747 | * it in the next loop. | |
748 | */ | |
749 | if ((iflags & BMV_IF_NO_HOLES) && | |
750 | map[i].br_startblock == HOLESTARTBLOCK) { | |
751 | memset(&out[cur_ext], 0, sizeof(out[cur_ext])); | |
752 | continue; | |
753 | } | |
754 | ||
c364b6d0 DW |
755 | /* |
756 | * In order to report shared extents accurately, | |
757 | * we report each distinct shared/unshared part | |
758 | * of a single bmbt record using multiple bmap | |
759 | * extents. To make that happen, we iterate the | |
760 | * same map array item multiple times, each | |
761 | * time trimming out the subextent that we just | |
762 | * reported. | |
763 | * | |
764 | * Because of this, we must check the out array | |
765 | * index (cur_ext) directly against bmv_count-1 | |
766 | * to avoid overflows. | |
767 | */ | |
f86f4037 DW |
768 | if (inject_map.br_startblock != NULLFSBLOCK) { |
769 | map[i] = inject_map; | |
770 | i--; | |
c364b6d0 | 771 | } |
68988114 DC |
772 | bmv->bmv_entries++; |
773 | cur_ext++; | |
774 | } | |
c364b6d0 | 775 | } while (nmap && bmv->bmv_length && cur_ext < bmv->bmv_count - 1); |
68988114 DC |
776 | |
777 | out_free_map: | |
778 | kmem_free(map); | |
779 | out_unlock_ilock: | |
01f4f327 | 780 | xfs_iunlock(ip, lock); |
68988114 DC |
781 | out_unlock_iolock: |
782 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
783 | ||
784 | for (i = 0; i < cur_ext; i++) { | |
68988114 | 785 | /* format results & advance arg */ |
1dbba086 ES |
786 | error = formatter(&arg, &out[i]); |
787 | if (error) | |
68988114 DC |
788 | break; |
789 | } | |
790 | ||
fdd3ccee | 791 | kmem_free(out); |
68988114 DC |
792 | return error; |
793 | } | |
794 | ||
795 | /* | |
796 | * dead simple method of punching delalyed allocation blocks from a range in | |
797 | * the inode. Walks a block at a time so will be slow, but is only executed in | |
ad4809bf | 798 | * rare error cases so the overhead is not critical. This will always punch out |
68988114 DC |
799 | * both the start and end blocks, even if the ranges only partially overlap |
800 | * them, so it is up to the caller to ensure that partial blocks are not | |
801 | * passed in. | |
802 | */ | |
803 | int | |
804 | xfs_bmap_punch_delalloc_range( | |
805 | struct xfs_inode *ip, | |
806 | xfs_fileoff_t start_fsb, | |
807 | xfs_fileoff_t length) | |
808 | { | |
809 | xfs_fileoff_t remaining = length; | |
810 | int error = 0; | |
811 | ||
812 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
813 | ||
814 | do { | |
815 | int done; | |
816 | xfs_bmbt_irec_t imap; | |
817 | int nimaps = 1; | |
818 | xfs_fsblock_t firstblock; | |
2c3234d1 | 819 | struct xfs_defer_ops dfops; |
68988114 DC |
820 | |
821 | /* | |
822 | * Map the range first and check that it is a delalloc extent | |
823 | * before trying to unmap the range. Otherwise we will be | |
824 | * trying to remove a real extent (which requires a | |
825 | * transaction) or a hole, which is probably a bad idea... | |
826 | */ | |
827 | error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps, | |
828 | XFS_BMAPI_ENTIRE); | |
829 | ||
830 | if (error) { | |
831 | /* something screwed, just bail */ | |
832 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
833 | xfs_alert(ip->i_mount, | |
834 | "Failed delalloc mapping lookup ino %lld fsb %lld.", | |
835 | ip->i_ino, start_fsb); | |
836 | } | |
837 | break; | |
838 | } | |
839 | if (!nimaps) { | |
840 | /* nothing there */ | |
841 | goto next_block; | |
842 | } | |
843 | if (imap.br_startblock != DELAYSTARTBLOCK) { | |
844 | /* been converted, ignore */ | |
845 | goto next_block; | |
846 | } | |
847 | WARN_ON(imap.br_blockcount == 0); | |
848 | ||
849 | /* | |
2c3234d1 | 850 | * Note: while we initialise the firstblock/dfops pair, they |
68988114 DC |
851 | * should never be used because blocks should never be |
852 | * allocated or freed for a delalloc extent and hence we need | |
853 | * don't cancel or finish them after the xfs_bunmapi() call. | |
854 | */ | |
2c3234d1 | 855 | xfs_defer_init(&dfops, &firstblock); |
68988114 | 856 | error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock, |
2c3234d1 | 857 | &dfops, &done); |
68988114 DC |
858 | if (error) |
859 | break; | |
860 | ||
2c3234d1 | 861 | ASSERT(!xfs_defer_has_unfinished_work(&dfops)); |
68988114 DC |
862 | next_block: |
863 | start_fsb++; | |
864 | remaining--; | |
865 | } while(remaining > 0); | |
866 | ||
867 | return error; | |
868 | } | |
c24b5dfa DC |
869 | |
870 | /* | |
871 | * Test whether it is appropriate to check an inode for and free post EOF | |
872 | * blocks. The 'force' parameter determines whether we should also consider | |
873 | * regular files that are marked preallocated or append-only. | |
874 | */ | |
875 | bool | |
876 | xfs_can_free_eofblocks(struct xfs_inode *ip, bool force) | |
877 | { | |
878 | /* prealloc/delalloc exists only on regular files */ | |
c19b3b05 | 879 | if (!S_ISREG(VFS_I(ip)->i_mode)) |
c24b5dfa DC |
880 | return false; |
881 | ||
882 | /* | |
883 | * Zero sized files with no cached pages and delalloc blocks will not | |
884 | * have speculative prealloc/delalloc blocks to remove. | |
885 | */ | |
886 | if (VFS_I(ip)->i_size == 0 && | |
2667c6f9 | 887 | VFS_I(ip)->i_mapping->nrpages == 0 && |
c24b5dfa DC |
888 | ip->i_delayed_blks == 0) |
889 | return false; | |
890 | ||
891 | /* If we haven't read in the extent list, then don't do it now. */ | |
892 | if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) | |
893 | return false; | |
894 | ||
895 | /* | |
896 | * Do not free real preallocated or append-only files unless the file | |
897 | * has delalloc blocks and we are forced to remove them. | |
898 | */ | |
899 | if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) | |
900 | if (!force || ip->i_delayed_blks == 0) | |
901 | return false; | |
902 | ||
903 | return true; | |
904 | } | |
905 | ||
906 | /* | |
907 | * This is called by xfs_inactive to free any blocks beyond eof | |
908 | * when the link count isn't zero and by xfs_dm_punch_hole() when | |
909 | * punching a hole to EOF. | |
910 | */ | |
911 | int | |
912 | xfs_free_eofblocks( | |
a36b9261 | 913 | struct xfs_inode *ip) |
c24b5dfa | 914 | { |
a36b9261 BF |
915 | struct xfs_trans *tp; |
916 | int error; | |
917 | xfs_fileoff_t end_fsb; | |
918 | xfs_fileoff_t last_fsb; | |
919 | xfs_filblks_t map_len; | |
920 | int nimaps; | |
921 | struct xfs_bmbt_irec imap; | |
922 | struct xfs_mount *mp = ip->i_mount; | |
923 | ||
924 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
c24b5dfa DC |
925 | |
926 | /* | |
927 | * Figure out if there are any blocks beyond the end | |
928 | * of the file. If not, then there is nothing to do. | |
929 | */ | |
930 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip)); | |
931 | last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); | |
932 | if (last_fsb <= end_fsb) | |
933 | return 0; | |
934 | map_len = last_fsb - end_fsb; | |
935 | ||
936 | nimaps = 1; | |
937 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
938 | error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0); | |
939 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
940 | ||
a36b9261 BF |
941 | /* |
942 | * If there are blocks after the end of file, truncate the file to its | |
943 | * current size to free them up. | |
944 | */ | |
c24b5dfa DC |
945 | if (!error && (nimaps != 0) && |
946 | (imap.br_startblock != HOLESTARTBLOCK || | |
947 | ip->i_delayed_blks)) { | |
948 | /* | |
949 | * Attach the dquots to the inode up front. | |
950 | */ | |
951 | error = xfs_qm_dqattach(ip, 0); | |
952 | if (error) | |
953 | return error; | |
954 | ||
e4229d6b BF |
955 | /* wait on dio to ensure i_size has settled */ |
956 | inode_dio_wait(VFS_I(ip)); | |
957 | ||
253f4911 CH |
958 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, |
959 | &tp); | |
c24b5dfa DC |
960 | if (error) { |
961 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
c24b5dfa DC |
962 | return error; |
963 | } | |
964 | ||
965 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
966 | xfs_trans_ijoin(tp, ip, 0); | |
967 | ||
968 | /* | |
969 | * Do not update the on-disk file size. If we update the | |
970 | * on-disk file size and then the system crashes before the | |
971 | * contents of the file are flushed to disk then the files | |
972 | * may be full of holes (ie NULL files bug). | |
973 | */ | |
974 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, | |
975 | XFS_ISIZE(ip)); | |
976 | if (error) { | |
977 | /* | |
978 | * If we get an error at this point we simply don't | |
979 | * bother truncating the file. | |
980 | */ | |
4906e215 | 981 | xfs_trans_cancel(tp); |
c24b5dfa | 982 | } else { |
70393313 | 983 | error = xfs_trans_commit(tp); |
c24b5dfa DC |
984 | if (!error) |
985 | xfs_inode_clear_eofblocks_tag(ip); | |
986 | } | |
987 | ||
988 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
c24b5dfa DC |
989 | } |
990 | return error; | |
991 | } | |
992 | ||
83aee9e4 | 993 | int |
c24b5dfa | 994 | xfs_alloc_file_space( |
83aee9e4 | 995 | struct xfs_inode *ip, |
c24b5dfa DC |
996 | xfs_off_t offset, |
997 | xfs_off_t len, | |
5f8aca8b | 998 | int alloc_type) |
c24b5dfa DC |
999 | { |
1000 | xfs_mount_t *mp = ip->i_mount; | |
1001 | xfs_off_t count; | |
1002 | xfs_filblks_t allocated_fsb; | |
1003 | xfs_filblks_t allocatesize_fsb; | |
1004 | xfs_extlen_t extsz, temp; | |
1005 | xfs_fileoff_t startoffset_fsb; | |
1006 | xfs_fsblock_t firstfsb; | |
1007 | int nimaps; | |
1008 | int quota_flag; | |
1009 | int rt; | |
1010 | xfs_trans_t *tp; | |
1011 | xfs_bmbt_irec_t imaps[1], *imapp; | |
2c3234d1 | 1012 | struct xfs_defer_ops dfops; |
c24b5dfa | 1013 | uint qblocks, resblks, resrtextents; |
c24b5dfa DC |
1014 | int error; |
1015 | ||
1016 | trace_xfs_alloc_file_space(ip); | |
1017 | ||
1018 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 1019 | return -EIO; |
c24b5dfa DC |
1020 | |
1021 | error = xfs_qm_dqattach(ip, 0); | |
1022 | if (error) | |
1023 | return error; | |
1024 | ||
1025 | if (len <= 0) | |
2451337d | 1026 | return -EINVAL; |
c24b5dfa DC |
1027 | |
1028 | rt = XFS_IS_REALTIME_INODE(ip); | |
1029 | extsz = xfs_get_extsz_hint(ip); | |
1030 | ||
1031 | count = len; | |
1032 | imapp = &imaps[0]; | |
1033 | nimaps = 1; | |
1034 | startoffset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1035 | allocatesize_fsb = XFS_B_TO_FSB(mp, count); | |
1036 | ||
1037 | /* | |
1038 | * Allocate file space until done or until there is an error | |
1039 | */ | |
1040 | while (allocatesize_fsb && !error) { | |
1041 | xfs_fileoff_t s, e; | |
1042 | ||
1043 | /* | |
1044 | * Determine space reservations for data/realtime. | |
1045 | */ | |
1046 | if (unlikely(extsz)) { | |
1047 | s = startoffset_fsb; | |
1048 | do_div(s, extsz); | |
1049 | s *= extsz; | |
1050 | e = startoffset_fsb + allocatesize_fsb; | |
1051 | if ((temp = do_mod(startoffset_fsb, extsz))) | |
1052 | e += temp; | |
1053 | if ((temp = do_mod(e, extsz))) | |
1054 | e += extsz - temp; | |
1055 | } else { | |
1056 | s = 0; | |
1057 | e = allocatesize_fsb; | |
1058 | } | |
1059 | ||
1060 | /* | |
1061 | * The transaction reservation is limited to a 32-bit block | |
1062 | * count, hence we need to limit the number of blocks we are | |
1063 | * trying to reserve to avoid an overflow. We can't allocate | |
1064 | * more than @nimaps extents, and an extent is limited on disk | |
1065 | * to MAXEXTLEN (21 bits), so use that to enforce the limit. | |
1066 | */ | |
1067 | resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps)); | |
1068 | if (unlikely(rt)) { | |
1069 | resrtextents = qblocks = resblks; | |
1070 | resrtextents /= mp->m_sb.sb_rextsize; | |
1071 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
1072 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
1073 | } else { | |
1074 | resrtextents = 0; | |
1075 | resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks); | |
1076 | quota_flag = XFS_QMOPT_RES_REGBLKS; | |
1077 | } | |
1078 | ||
1079 | /* | |
1080 | * Allocate and setup the transaction. | |
1081 | */ | |
253f4911 CH |
1082 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, |
1083 | resrtextents, 0, &tp); | |
1084 | ||
c24b5dfa DC |
1085 | /* |
1086 | * Check for running out of space | |
1087 | */ | |
1088 | if (error) { | |
1089 | /* | |
1090 | * Free the transaction structure. | |
1091 | */ | |
2451337d | 1092 | ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp)); |
c24b5dfa DC |
1093 | break; |
1094 | } | |
1095 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1096 | error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, | |
1097 | 0, quota_flag); | |
1098 | if (error) | |
1099 | goto error1; | |
1100 | ||
1101 | xfs_trans_ijoin(tp, ip, 0); | |
1102 | ||
2c3234d1 | 1103 | xfs_defer_init(&dfops, &firstfsb); |
c24b5dfa DC |
1104 | error = xfs_bmapi_write(tp, ip, startoffset_fsb, |
1105 | allocatesize_fsb, alloc_type, &firstfsb, | |
2c3234d1 | 1106 | resblks, imapp, &nimaps, &dfops); |
f6106efa | 1107 | if (error) |
c24b5dfa | 1108 | goto error0; |
c24b5dfa DC |
1109 | |
1110 | /* | |
1111 | * Complete the transaction | |
1112 | */ | |
2c3234d1 | 1113 | error = xfs_defer_finish(&tp, &dfops, NULL); |
f6106efa | 1114 | if (error) |
c24b5dfa | 1115 | goto error0; |
c24b5dfa | 1116 | |
70393313 | 1117 | error = xfs_trans_commit(tp); |
c24b5dfa | 1118 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
f6106efa | 1119 | if (error) |
c24b5dfa | 1120 | break; |
c24b5dfa DC |
1121 | |
1122 | allocated_fsb = imapp->br_blockcount; | |
1123 | ||
1124 | if (nimaps == 0) { | |
2451337d | 1125 | error = -ENOSPC; |
c24b5dfa DC |
1126 | break; |
1127 | } | |
1128 | ||
1129 | startoffset_fsb += allocated_fsb; | |
1130 | allocatesize_fsb -= allocated_fsb; | |
1131 | } | |
1132 | ||
1133 | return error; | |
1134 | ||
1135 | error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */ | |
2c3234d1 | 1136 | xfs_defer_cancel(&dfops); |
c24b5dfa DC |
1137 | xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag); |
1138 | ||
1139 | error1: /* Just cancel transaction */ | |
4906e215 | 1140 | xfs_trans_cancel(tp); |
c24b5dfa DC |
1141 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1142 | return error; | |
1143 | } | |
1144 | ||
bdb0d04f CH |
1145 | static int |
1146 | xfs_unmap_extent( | |
1147 | struct xfs_inode *ip, | |
1148 | xfs_fileoff_t startoffset_fsb, | |
1149 | xfs_filblks_t len_fsb, | |
1150 | int *done) | |
c24b5dfa | 1151 | { |
bdb0d04f CH |
1152 | struct xfs_mount *mp = ip->i_mount; |
1153 | struct xfs_trans *tp; | |
2c3234d1 | 1154 | struct xfs_defer_ops dfops; |
bdb0d04f CH |
1155 | xfs_fsblock_t firstfsb; |
1156 | uint resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
1157 | int error; | |
c24b5dfa | 1158 | |
bdb0d04f CH |
1159 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp); |
1160 | if (error) { | |
1161 | ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
1162 | return error; | |
1163 | } | |
c24b5dfa | 1164 | |
bdb0d04f CH |
1165 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
1166 | error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot, ip->i_gdquot, | |
1167 | ip->i_pdquot, resblks, 0, XFS_QMOPT_RES_REGBLKS); | |
1168 | if (error) | |
1169 | goto out_trans_cancel; | |
c24b5dfa | 1170 | |
bdb0d04f | 1171 | xfs_trans_ijoin(tp, ip, 0); |
4f317369 | 1172 | |
2c3234d1 | 1173 | xfs_defer_init(&dfops, &firstfsb); |
bdb0d04f | 1174 | error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, &firstfsb, |
2c3234d1 | 1175 | &dfops, done); |
bdb0d04f CH |
1176 | if (error) |
1177 | goto out_bmap_cancel; | |
4f317369 | 1178 | |
2c3234d1 | 1179 | error = xfs_defer_finish(&tp, &dfops, ip); |
bdb0d04f CH |
1180 | if (error) |
1181 | goto out_bmap_cancel; | |
4f317369 | 1182 | |
bdb0d04f CH |
1183 | error = xfs_trans_commit(tp); |
1184 | out_unlock: | |
1185 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1186 | return error; | |
4f69f578 | 1187 | |
bdb0d04f | 1188 | out_bmap_cancel: |
2c3234d1 | 1189 | xfs_defer_cancel(&dfops); |
bdb0d04f CH |
1190 | out_trans_cancel: |
1191 | xfs_trans_cancel(tp); | |
1192 | goto out_unlock; | |
1193 | } | |
4f69f578 | 1194 | |
bdb0d04f CH |
1195 | static int |
1196 | xfs_adjust_extent_unmap_boundaries( | |
1197 | struct xfs_inode *ip, | |
1198 | xfs_fileoff_t *startoffset_fsb, | |
1199 | xfs_fileoff_t *endoffset_fsb) | |
1200 | { | |
1201 | struct xfs_mount *mp = ip->i_mount; | |
1202 | struct xfs_bmbt_irec imap; | |
1203 | int nimap, error; | |
1204 | xfs_extlen_t mod = 0; | |
4f69f578 | 1205 | |
bdb0d04f CH |
1206 | nimap = 1; |
1207 | error = xfs_bmapi_read(ip, *startoffset_fsb, 1, &imap, &nimap, 0); | |
1208 | if (error) | |
1209 | return error; | |
83a0adc3 | 1210 | |
bdb0d04f CH |
1211 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { |
1212 | xfs_daddr_t block; | |
8c156125 | 1213 | |
bdb0d04f CH |
1214 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); |
1215 | block = imap.br_startblock; | |
1216 | mod = do_div(block, mp->m_sb.sb_rextsize); | |
1217 | if (mod) | |
1218 | *startoffset_fsb += mp->m_sb.sb_rextsize - mod; | |
1219 | } | |
83a0adc3 | 1220 | |
bdb0d04f CH |
1221 | nimap = 1; |
1222 | error = xfs_bmapi_read(ip, *endoffset_fsb - 1, 1, &imap, &nimap, 0); | |
1223 | if (error) | |
1224 | return error; | |
1225 | ||
1226 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
1227 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
1228 | mod++; | |
1229 | if (mod && mod != mp->m_sb.sb_rextsize) | |
1230 | *endoffset_fsb -= mod; | |
c24b5dfa | 1231 | } |
bdb0d04f CH |
1232 | |
1233 | return 0; | |
1234 | } | |
1235 | ||
1236 | static int | |
1237 | xfs_flush_unmap_range( | |
1238 | struct xfs_inode *ip, | |
1239 | xfs_off_t offset, | |
1240 | xfs_off_t len) | |
1241 | { | |
1242 | struct xfs_mount *mp = ip->i_mount; | |
1243 | struct inode *inode = VFS_I(ip); | |
1244 | xfs_off_t rounding, start, end; | |
1245 | int error; | |
1246 | ||
1247 | /* wait for the completion of any pending DIOs */ | |
1248 | inode_dio_wait(inode); | |
1249 | ||
1250 | rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_SIZE); | |
1251 | start = round_down(offset, rounding); | |
1252 | end = round_up(offset + len, rounding) - 1; | |
1253 | ||
1254 | error = filemap_write_and_wait_range(inode->i_mapping, start, end); | |
1255 | if (error) | |
1256 | return error; | |
1257 | truncate_pagecache_range(inode, start, end); | |
1258 | return 0; | |
c24b5dfa DC |
1259 | } |
1260 | ||
83aee9e4 | 1261 | int |
c24b5dfa | 1262 | xfs_free_file_space( |
83aee9e4 | 1263 | struct xfs_inode *ip, |
c24b5dfa | 1264 | xfs_off_t offset, |
5f8aca8b | 1265 | xfs_off_t len) |
c24b5dfa | 1266 | { |
bdb0d04f | 1267 | struct xfs_mount *mp = ip->i_mount; |
c24b5dfa | 1268 | xfs_fileoff_t startoffset_fsb; |
bdb0d04f | 1269 | xfs_fileoff_t endoffset_fsb; |
3c2bdc91 | 1270 | int done = 0, error; |
c24b5dfa DC |
1271 | |
1272 | trace_xfs_free_file_space(ip); | |
1273 | ||
1274 | error = xfs_qm_dqattach(ip, 0); | |
1275 | if (error) | |
1276 | return error; | |
1277 | ||
c24b5dfa | 1278 | if (len <= 0) /* if nothing being freed */ |
bdb0d04f | 1279 | return 0; |
c24b5dfa | 1280 | |
bdb0d04f | 1281 | error = xfs_flush_unmap_range(ip, offset, len); |
c24b5dfa | 1282 | if (error) |
bdb0d04f CH |
1283 | return error; |
1284 | ||
1285 | startoffset_fsb = XFS_B_TO_FSB(mp, offset); | |
1286 | endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len); | |
c24b5dfa DC |
1287 | |
1288 | /* | |
bdb0d04f CH |
1289 | * Need to zero the stuff we're not freeing, on disk. If it's a RT file |
1290 | * and we can't use unwritten extents then we actually need to ensure | |
1291 | * to zero the whole extent, otherwise we just need to take of block | |
1292 | * boundaries, and xfs_bunmapi will handle the rest. | |
c24b5dfa | 1293 | */ |
bdb0d04f CH |
1294 | if (XFS_IS_REALTIME_INODE(ip) && |
1295 | !xfs_sb_version_hasextflgbit(&mp->m_sb)) { | |
1296 | error = xfs_adjust_extent_unmap_boundaries(ip, &startoffset_fsb, | |
1297 | &endoffset_fsb); | |
c24b5dfa | 1298 | if (error) |
bdb0d04f | 1299 | return error; |
c24b5dfa DC |
1300 | } |
1301 | ||
3c2bdc91 CH |
1302 | if (endoffset_fsb > startoffset_fsb) { |
1303 | while (!done) { | |
1304 | error = xfs_unmap_extent(ip, startoffset_fsb, | |
1305 | endoffset_fsb - startoffset_fsb, &done); | |
1306 | if (error) | |
1307 | return error; | |
c24b5dfa | 1308 | } |
c24b5dfa DC |
1309 | } |
1310 | ||
3c2bdc91 CH |
1311 | /* |
1312 | * Now that we've unmap all full blocks we'll have to zero out any | |
1313 | * partial block at the beginning and/or end. xfs_zero_range is | |
3dd09d5a CO |
1314 | * smart enough to skip any holes, including those we just created, |
1315 | * but we must take care not to zero beyond EOF and enlarge i_size. | |
3c2bdc91 | 1316 | */ |
3dd09d5a CO |
1317 | |
1318 | if (offset >= XFS_ISIZE(ip)) | |
1319 | return 0; | |
1320 | ||
1321 | if (offset + len > XFS_ISIZE(ip)) | |
1322 | len = XFS_ISIZE(ip) - offset; | |
1323 | ||
3c2bdc91 | 1324 | return xfs_zero_range(ip, offset, len, NULL); |
c24b5dfa DC |
1325 | } |
1326 | ||
5d11fb4b BF |
1327 | /* |
1328 | * Preallocate and zero a range of a file. This mechanism has the allocation | |
1329 | * semantics of fallocate and in addition converts data in the range to zeroes. | |
1330 | */ | |
865e9446 | 1331 | int |
c24b5dfa DC |
1332 | xfs_zero_file_space( |
1333 | struct xfs_inode *ip, | |
1334 | xfs_off_t offset, | |
5f8aca8b | 1335 | xfs_off_t len) |
c24b5dfa DC |
1336 | { |
1337 | struct xfs_mount *mp = ip->i_mount; | |
5d11fb4b | 1338 | uint blksize; |
c24b5dfa DC |
1339 | int error; |
1340 | ||
897b73b6 DC |
1341 | trace_xfs_zero_file_space(ip); |
1342 | ||
5d11fb4b | 1343 | blksize = 1 << mp->m_sb.sb_blocklog; |
c24b5dfa DC |
1344 | |
1345 | /* | |
5d11fb4b BF |
1346 | * Punch a hole and prealloc the range. We use hole punch rather than |
1347 | * unwritten extent conversion for two reasons: | |
1348 | * | |
1349 | * 1.) Hole punch handles partial block zeroing for us. | |
1350 | * | |
1351 | * 2.) If prealloc returns ENOSPC, the file range is still zero-valued | |
1352 | * by virtue of the hole punch. | |
c24b5dfa | 1353 | */ |
5d11fb4b BF |
1354 | error = xfs_free_file_space(ip, offset, len); |
1355 | if (error) | |
1356 | goto out; | |
c24b5dfa | 1357 | |
5d11fb4b BF |
1358 | error = xfs_alloc_file_space(ip, round_down(offset, blksize), |
1359 | round_up(offset + len, blksize) - | |
1360 | round_down(offset, blksize), | |
1361 | XFS_BMAPI_PREALLOC); | |
5f8aca8b | 1362 | out: |
c24b5dfa DC |
1363 | return error; |
1364 | ||
1365 | } | |
1366 | ||
e1d8fb88 | 1367 | /* |
a904b1ca NJ |
1368 | * @next_fsb will keep track of the extent currently undergoing shift. |
1369 | * @stop_fsb will keep track of the extent at which we have to stop. | |
1370 | * If we are shifting left, we will start with block (offset + len) and | |
1371 | * shift each extent till last extent. | |
1372 | * If we are shifting right, we will start with last extent inside file space | |
1373 | * and continue until we reach the block corresponding to offset. | |
e1d8fb88 | 1374 | */ |
72c1a739 | 1375 | static int |
a904b1ca NJ |
1376 | xfs_shift_file_space( |
1377 | struct xfs_inode *ip, | |
1378 | xfs_off_t offset, | |
1379 | xfs_off_t len, | |
1380 | enum shift_direction direction) | |
e1d8fb88 NJ |
1381 | { |
1382 | int done = 0; | |
1383 | struct xfs_mount *mp = ip->i_mount; | |
1384 | struct xfs_trans *tp; | |
1385 | int error; | |
2c3234d1 | 1386 | struct xfs_defer_ops dfops; |
e1d8fb88 | 1387 | xfs_fsblock_t first_block; |
a904b1ca | 1388 | xfs_fileoff_t stop_fsb; |
2c845f5a | 1389 | xfs_fileoff_t next_fsb; |
e1d8fb88 | 1390 | xfs_fileoff_t shift_fsb; |
48af96ab | 1391 | uint resblks; |
e1d8fb88 | 1392 | |
a904b1ca | 1393 | ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT); |
e1d8fb88 | 1394 | |
a904b1ca | 1395 | if (direction == SHIFT_LEFT) { |
48af96ab BF |
1396 | /* |
1397 | * Reserve blocks to cover potential extent merges after left | |
1398 | * shift operations. | |
1399 | */ | |
1400 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
a904b1ca NJ |
1401 | next_fsb = XFS_B_TO_FSB(mp, offset + len); |
1402 | stop_fsb = XFS_B_TO_FSB(mp, VFS_I(ip)->i_size); | |
1403 | } else { | |
1404 | /* | |
1405 | * If right shift, delegate the work of initialization of | |
1406 | * next_fsb to xfs_bmap_shift_extent as it has ilock held. | |
1407 | */ | |
48af96ab | 1408 | resblks = 0; |
a904b1ca NJ |
1409 | next_fsb = NULLFSBLOCK; |
1410 | stop_fsb = XFS_B_TO_FSB(mp, offset); | |
1411 | } | |
e1d8fb88 | 1412 | |
e1d8fb88 NJ |
1413 | shift_fsb = XFS_B_TO_FSB(mp, len); |
1414 | ||
f71721d0 BF |
1415 | /* |
1416 | * Trim eofblocks to avoid shifting uninitialized post-eof preallocation | |
1417 | * into the accessible region of the file. | |
1418 | */ | |
41b9d726 | 1419 | if (xfs_can_free_eofblocks(ip, true)) { |
a36b9261 | 1420 | error = xfs_free_eofblocks(ip); |
41b9d726 BF |
1421 | if (error) |
1422 | return error; | |
1423 | } | |
1669a8ca | 1424 | |
f71721d0 BF |
1425 | /* |
1426 | * Writeback and invalidate cache for the remainder of the file as we're | |
a904b1ca | 1427 | * about to shift down every extent from offset to EOF. |
f71721d0 BF |
1428 | */ |
1429 | error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, | |
a904b1ca | 1430 | offset, -1); |
f71721d0 BF |
1431 | if (error) |
1432 | return error; | |
1433 | error = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, | |
09cbfeaf | 1434 | offset >> PAGE_SHIFT, -1); |
e1d8fb88 NJ |
1435 | if (error) |
1436 | return error; | |
1437 | ||
a904b1ca NJ |
1438 | /* |
1439 | * The extent shiting code works on extent granularity. So, if | |
1440 | * stop_fsb is not the starting block of extent, we need to split | |
1441 | * the extent at stop_fsb. | |
1442 | */ | |
1443 | if (direction == SHIFT_RIGHT) { | |
1444 | error = xfs_bmap_split_extent(ip, stop_fsb); | |
1445 | if (error) | |
1446 | return error; | |
1447 | } | |
1448 | ||
e1d8fb88 | 1449 | while (!error && !done) { |
48af96ab BF |
1450 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, |
1451 | &tp); | |
253f4911 | 1452 | if (error) |
e1d8fb88 | 1453 | break; |
e1d8fb88 NJ |
1454 | |
1455 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1456 | error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot, | |
48af96ab | 1457 | ip->i_gdquot, ip->i_pdquot, resblks, 0, |
e1d8fb88 NJ |
1458 | XFS_QMOPT_RES_REGBLKS); |
1459 | if (error) | |
d4a97a04 | 1460 | goto out_trans_cancel; |
e1d8fb88 | 1461 | |
a904b1ca | 1462 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
e1d8fb88 | 1463 | |
2c3234d1 | 1464 | xfs_defer_init(&dfops, &first_block); |
e1d8fb88 NJ |
1465 | |
1466 | /* | |
1467 | * We are using the write transaction in which max 2 bmbt | |
1468 | * updates are allowed | |
1469 | */ | |
a904b1ca | 1470 | error = xfs_bmap_shift_extents(tp, ip, &next_fsb, shift_fsb, |
2c3234d1 | 1471 | &done, stop_fsb, &first_block, &dfops, |
a904b1ca | 1472 | direction, XFS_BMAP_MAX_SHIFT_EXTENTS); |
e1d8fb88 | 1473 | if (error) |
d4a97a04 | 1474 | goto out_bmap_cancel; |
e1d8fb88 | 1475 | |
2c3234d1 | 1476 | error = xfs_defer_finish(&tp, &dfops, NULL); |
e1d8fb88 | 1477 | if (error) |
d4a97a04 | 1478 | goto out_bmap_cancel; |
e1d8fb88 | 1479 | |
70393313 | 1480 | error = xfs_trans_commit(tp); |
e1d8fb88 NJ |
1481 | } |
1482 | ||
1483 | return error; | |
1484 | ||
d4a97a04 | 1485 | out_bmap_cancel: |
2c3234d1 | 1486 | xfs_defer_cancel(&dfops); |
d4a97a04 | 1487 | out_trans_cancel: |
4906e215 | 1488 | xfs_trans_cancel(tp); |
e1d8fb88 NJ |
1489 | return error; |
1490 | } | |
1491 | ||
a904b1ca NJ |
1492 | /* |
1493 | * xfs_collapse_file_space() | |
1494 | * This routine frees disk space and shift extent for the given file. | |
1495 | * The first thing we do is to free data blocks in the specified range | |
1496 | * by calling xfs_free_file_space(). It would also sync dirty data | |
1497 | * and invalidate page cache over the region on which collapse range | |
1498 | * is working. And Shift extent records to the left to cover a hole. | |
1499 | * RETURNS: | |
1500 | * 0 on success | |
1501 | * errno on error | |
1502 | * | |
1503 | */ | |
1504 | int | |
1505 | xfs_collapse_file_space( | |
1506 | struct xfs_inode *ip, | |
1507 | xfs_off_t offset, | |
1508 | xfs_off_t len) | |
1509 | { | |
1510 | int error; | |
1511 | ||
1512 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
1513 | trace_xfs_collapse_file_space(ip); | |
1514 | ||
1515 | error = xfs_free_file_space(ip, offset, len); | |
1516 | if (error) | |
1517 | return error; | |
1518 | ||
1519 | return xfs_shift_file_space(ip, offset, len, SHIFT_LEFT); | |
1520 | } | |
1521 | ||
1522 | /* | |
1523 | * xfs_insert_file_space() | |
1524 | * This routine create hole space by shifting extents for the given file. | |
1525 | * The first thing we do is to sync dirty data and invalidate page cache | |
1526 | * over the region on which insert range is working. And split an extent | |
1527 | * to two extents at given offset by calling xfs_bmap_split_extent. | |
1528 | * And shift all extent records which are laying between [offset, | |
1529 | * last allocated extent] to the right to reserve hole range. | |
1530 | * RETURNS: | |
1531 | * 0 on success | |
1532 | * errno on error | |
1533 | */ | |
1534 | int | |
1535 | xfs_insert_file_space( | |
1536 | struct xfs_inode *ip, | |
1537 | loff_t offset, | |
1538 | loff_t len) | |
1539 | { | |
1540 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
1541 | trace_xfs_insert_file_space(ip); | |
1542 | ||
1543 | return xfs_shift_file_space(ip, offset, len, SHIFT_RIGHT); | |
1544 | } | |
1545 | ||
a133d952 DC |
1546 | /* |
1547 | * We need to check that the format of the data fork in the temporary inode is | |
1548 | * valid for the target inode before doing the swap. This is not a problem with | |
1549 | * attr1 because of the fixed fork offset, but attr2 has a dynamically sized | |
1550 | * data fork depending on the space the attribute fork is taking so we can get | |
1551 | * invalid formats on the target inode. | |
1552 | * | |
1553 | * E.g. target has space for 7 extents in extent format, temp inode only has | |
1554 | * space for 6. If we defragment down to 7 extents, then the tmp format is a | |
1555 | * btree, but when swapped it needs to be in extent format. Hence we can't just | |
1556 | * blindly swap data forks on attr2 filesystems. | |
1557 | * | |
1558 | * Note that we check the swap in both directions so that we don't end up with | |
1559 | * a corrupt temporary inode, either. | |
1560 | * | |
1561 | * Note that fixing the way xfs_fsr sets up the attribute fork in the source | |
1562 | * inode will prevent this situation from occurring, so all we do here is | |
1563 | * reject and log the attempt. basically we are putting the responsibility on | |
1564 | * userspace to get this right. | |
1565 | */ | |
1566 | static int | |
1567 | xfs_swap_extents_check_format( | |
e06259aa DW |
1568 | struct xfs_inode *ip, /* target inode */ |
1569 | struct xfs_inode *tip) /* tmp inode */ | |
a133d952 DC |
1570 | { |
1571 | ||
1572 | /* Should never get a local format */ | |
1573 | if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL || | |
1574 | tip->i_d.di_format == XFS_DINODE_FMT_LOCAL) | |
2451337d | 1575 | return -EINVAL; |
a133d952 DC |
1576 | |
1577 | /* | |
1578 | * if the target inode has less extents that then temporary inode then | |
1579 | * why did userspace call us? | |
1580 | */ | |
1581 | if (ip->i_d.di_nextents < tip->i_d.di_nextents) | |
2451337d | 1582 | return -EINVAL; |
a133d952 | 1583 | |
1f08af52 DW |
1584 | /* |
1585 | * If we have to use the (expensive) rmap swap method, we can | |
1586 | * handle any number of extents and any format. | |
1587 | */ | |
1588 | if (xfs_sb_version_hasrmapbt(&ip->i_mount->m_sb)) | |
1589 | return 0; | |
1590 | ||
a133d952 DC |
1591 | /* |
1592 | * if the target inode is in extent form and the temp inode is in btree | |
1593 | * form then we will end up with the target inode in the wrong format | |
1594 | * as we already know there are less extents in the temp inode. | |
1595 | */ | |
1596 | if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS && | |
1597 | tip->i_d.di_format == XFS_DINODE_FMT_BTREE) | |
2451337d | 1598 | return -EINVAL; |
a133d952 DC |
1599 | |
1600 | /* Check temp in extent form to max in target */ | |
1601 | if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS && | |
1602 | XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) > | |
1603 | XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)) | |
2451337d | 1604 | return -EINVAL; |
a133d952 DC |
1605 | |
1606 | /* Check target in extent form to max in temp */ | |
1607 | if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS && | |
1608 | XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) > | |
1609 | XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK)) | |
2451337d | 1610 | return -EINVAL; |
a133d952 DC |
1611 | |
1612 | /* | |
1613 | * If we are in a btree format, check that the temp root block will fit | |
1614 | * in the target and that it has enough extents to be in btree format | |
1615 | * in the target. | |
1616 | * | |
1617 | * Note that we have to be careful to allow btree->extent conversions | |
1618 | * (a common defrag case) which will occur when the temp inode is in | |
1619 | * extent format... | |
1620 | */ | |
1621 | if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) { | |
1622 | if (XFS_IFORK_BOFF(ip) && | |
1623 | XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip)) | |
2451337d | 1624 | return -EINVAL; |
a133d952 DC |
1625 | if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <= |
1626 | XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)) | |
2451337d | 1627 | return -EINVAL; |
a133d952 DC |
1628 | } |
1629 | ||
1630 | /* Reciprocal target->temp btree format checks */ | |
1631 | if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) { | |
1632 | if (XFS_IFORK_BOFF(tip) && | |
1633 | XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip)) | |
2451337d | 1634 | return -EINVAL; |
a133d952 DC |
1635 | if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <= |
1636 | XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK)) | |
2451337d | 1637 | return -EINVAL; |
a133d952 DC |
1638 | } |
1639 | ||
1640 | return 0; | |
1641 | } | |
1642 | ||
7abbb8f9 | 1643 | static int |
4ef897a2 DC |
1644 | xfs_swap_extent_flush( |
1645 | struct xfs_inode *ip) | |
1646 | { | |
1647 | int error; | |
1648 | ||
1649 | error = filemap_write_and_wait(VFS_I(ip)->i_mapping); | |
1650 | if (error) | |
1651 | return error; | |
1652 | truncate_pagecache_range(VFS_I(ip), 0, -1); | |
1653 | ||
1654 | /* Verify O_DIRECT for ftmp */ | |
1655 | if (VFS_I(ip)->i_mapping->nrpages) | |
1656 | return -EINVAL; | |
4ef897a2 DC |
1657 | return 0; |
1658 | } | |
1659 | ||
1f08af52 DW |
1660 | /* |
1661 | * Move extents from one file to another, when rmap is enabled. | |
1662 | */ | |
1663 | STATIC int | |
1664 | xfs_swap_extent_rmap( | |
1665 | struct xfs_trans **tpp, | |
1666 | struct xfs_inode *ip, | |
1667 | struct xfs_inode *tip) | |
1668 | { | |
1669 | struct xfs_bmbt_irec irec; | |
1670 | struct xfs_bmbt_irec uirec; | |
1671 | struct xfs_bmbt_irec tirec; | |
1672 | xfs_fileoff_t offset_fsb; | |
1673 | xfs_fileoff_t end_fsb; | |
1674 | xfs_filblks_t count_fsb; | |
1675 | xfs_fsblock_t firstfsb; | |
1676 | struct xfs_defer_ops dfops; | |
1677 | int error; | |
1678 | xfs_filblks_t ilen; | |
1679 | xfs_filblks_t rlen; | |
1680 | int nimaps; | |
1681 | __uint64_t tip_flags2; | |
1682 | ||
1683 | /* | |
1684 | * If the source file has shared blocks, we must flag the donor | |
1685 | * file as having shared blocks so that we get the shared-block | |
1686 | * rmap functions when we go to fix up the rmaps. The flags | |
1687 | * will be switch for reals later. | |
1688 | */ | |
1689 | tip_flags2 = tip->i_d.di_flags2; | |
1690 | if (ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK) | |
1691 | tip->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK; | |
1692 | ||
1693 | offset_fsb = 0; | |
1694 | end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip))); | |
1695 | count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb); | |
1696 | ||
1697 | while (count_fsb) { | |
1698 | /* Read extent from the donor file */ | |
1699 | nimaps = 1; | |
1700 | error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec, | |
1701 | &nimaps, 0); | |
1702 | if (error) | |
1703 | goto out; | |
1704 | ASSERT(nimaps == 1); | |
1705 | ASSERT(tirec.br_startblock != DELAYSTARTBLOCK); | |
1706 | ||
1707 | trace_xfs_swap_extent_rmap_remap(tip, &tirec); | |
1708 | ilen = tirec.br_blockcount; | |
1709 | ||
1710 | /* Unmap the old blocks in the source file. */ | |
1711 | while (tirec.br_blockcount) { | |
1712 | xfs_defer_init(&dfops, &firstfsb); | |
1713 | trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec); | |
1714 | ||
1715 | /* Read extent from the source file */ | |
1716 | nimaps = 1; | |
1717 | error = xfs_bmapi_read(ip, tirec.br_startoff, | |
1718 | tirec.br_blockcount, &irec, | |
1719 | &nimaps, 0); | |
1720 | if (error) | |
1721 | goto out_defer; | |
1722 | ASSERT(nimaps == 1); | |
1723 | ASSERT(tirec.br_startoff == irec.br_startoff); | |
1724 | trace_xfs_swap_extent_rmap_remap_piece(ip, &irec); | |
1725 | ||
1726 | /* Trim the extent. */ | |
1727 | uirec = tirec; | |
1728 | uirec.br_blockcount = rlen = min_t(xfs_filblks_t, | |
1729 | tirec.br_blockcount, | |
1730 | irec.br_blockcount); | |
1731 | trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec); | |
1732 | ||
1733 | /* Remove the mapping from the donor file. */ | |
1734 | error = xfs_bmap_unmap_extent((*tpp)->t_mountp, &dfops, | |
1735 | tip, &uirec); | |
1736 | if (error) | |
1737 | goto out_defer; | |
1738 | ||
1739 | /* Remove the mapping from the source file. */ | |
1740 | error = xfs_bmap_unmap_extent((*tpp)->t_mountp, &dfops, | |
1741 | ip, &irec); | |
1742 | if (error) | |
1743 | goto out_defer; | |
1744 | ||
1745 | /* Map the donor file's blocks into the source file. */ | |
1746 | error = xfs_bmap_map_extent((*tpp)->t_mountp, &dfops, | |
1747 | ip, &uirec); | |
1748 | if (error) | |
1749 | goto out_defer; | |
1750 | ||
1751 | /* Map the source file's blocks into the donor file. */ | |
1752 | error = xfs_bmap_map_extent((*tpp)->t_mountp, &dfops, | |
1753 | tip, &irec); | |
1754 | if (error) | |
1755 | goto out_defer; | |
1756 | ||
1757 | error = xfs_defer_finish(tpp, &dfops, ip); | |
1758 | if (error) | |
1759 | goto out_defer; | |
1760 | ||
1761 | tirec.br_startoff += rlen; | |
1762 | if (tirec.br_startblock != HOLESTARTBLOCK && | |
1763 | tirec.br_startblock != DELAYSTARTBLOCK) | |
1764 | tirec.br_startblock += rlen; | |
1765 | tirec.br_blockcount -= rlen; | |
1766 | } | |
1767 | ||
1768 | /* Roll on... */ | |
1769 | count_fsb -= ilen; | |
1770 | offset_fsb += ilen; | |
1771 | } | |
1772 | ||
1773 | tip->i_d.di_flags2 = tip_flags2; | |
1774 | return 0; | |
1775 | ||
1776 | out_defer: | |
1777 | xfs_defer_cancel(&dfops); | |
1778 | out: | |
1779 | trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_); | |
1780 | tip->i_d.di_flags2 = tip_flags2; | |
1781 | return error; | |
1782 | } | |
1783 | ||
39aff5fd DW |
1784 | /* Swap the extents of two files by swapping data forks. */ |
1785 | STATIC int | |
1786 | xfs_swap_extent_forks( | |
1787 | struct xfs_trans *tp, | |
1788 | struct xfs_inode *ip, | |
1789 | struct xfs_inode *tip, | |
1790 | int *src_log_flags, | |
1791 | int *target_log_flags) | |
a133d952 | 1792 | { |
39aff5fd | 1793 | struct xfs_ifork tempifp, *ifp, *tifp; |
e06259aa DW |
1794 | int aforkblks = 0; |
1795 | int taforkblks = 0; | |
4dfce57d | 1796 | xfs_extnum_t nextents; |
e06259aa | 1797 | __uint64_t tmp; |
39aff5fd | 1798 | int error; |
a133d952 | 1799 | |
a133d952 DC |
1800 | /* |
1801 | * Count the number of extended attribute blocks | |
1802 | */ | |
1803 | if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) && | |
1804 | (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) { | |
39aff5fd DW |
1805 | error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, |
1806 | &aforkblks); | |
a133d952 | 1807 | if (error) |
39aff5fd | 1808 | return error; |
a133d952 DC |
1809 | } |
1810 | if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) && | |
1811 | (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) { | |
1812 | error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, | |
39aff5fd | 1813 | &taforkblks); |
a133d952 | 1814 | if (error) |
39aff5fd | 1815 | return error; |
a133d952 DC |
1816 | } |
1817 | ||
21b5c978 DC |
1818 | /* |
1819 | * Before we've swapped the forks, lets set the owners of the forks | |
1820 | * appropriately. We have to do this as we are demand paging the btree | |
1821 | * buffers, and so the validation done on read will expect the owner | |
1822 | * field to be correctly set. Once we change the owners, we can swap the | |
1823 | * inode forks. | |
21b5c978 | 1824 | */ |
21b5c978 DC |
1825 | if (ip->i_d.di_version == 3 && |
1826 | ip->i_d.di_format == XFS_DINODE_FMT_BTREE) { | |
39aff5fd | 1827 | (*target_log_flags) |= XFS_ILOG_DOWNER; |
638f4416 DC |
1828 | error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, |
1829 | tip->i_ino, NULL); | |
21b5c978 | 1830 | if (error) |
39aff5fd | 1831 | return error; |
21b5c978 DC |
1832 | } |
1833 | ||
1834 | if (tip->i_d.di_version == 3 && | |
1835 | tip->i_d.di_format == XFS_DINODE_FMT_BTREE) { | |
39aff5fd | 1836 | (*src_log_flags) |= XFS_ILOG_DOWNER; |
638f4416 DC |
1837 | error = xfs_bmbt_change_owner(tp, tip, XFS_DATA_FORK, |
1838 | ip->i_ino, NULL); | |
21b5c978 | 1839 | if (error) |
39aff5fd | 1840 | return error; |
21b5c978 DC |
1841 | } |
1842 | ||
a133d952 DC |
1843 | /* |
1844 | * Swap the data forks of the inodes | |
1845 | */ | |
1846 | ifp = &ip->i_df; | |
1847 | tifp = &tip->i_df; | |
39aff5fd | 1848 | tempifp = *ifp; /* struct copy */ |
a133d952 | 1849 | *ifp = *tifp; /* struct copy */ |
39aff5fd | 1850 | *tifp = tempifp; /* struct copy */ |
a133d952 DC |
1851 | |
1852 | /* | |
1853 | * Fix the on-disk inode values | |
1854 | */ | |
1855 | tmp = (__uint64_t)ip->i_d.di_nblocks; | |
1856 | ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks; | |
1857 | tip->i_d.di_nblocks = tmp + taforkblks - aforkblks; | |
1858 | ||
1859 | tmp = (__uint64_t) ip->i_d.di_nextents; | |
1860 | ip->i_d.di_nextents = tip->i_d.di_nextents; | |
1861 | tip->i_d.di_nextents = tmp; | |
1862 | ||
1863 | tmp = (__uint64_t) ip->i_d.di_format; | |
1864 | ip->i_d.di_format = tip->i_d.di_format; | |
1865 | tip->i_d.di_format = tmp; | |
1866 | ||
1867 | /* | |
1868 | * The extents in the source inode could still contain speculative | |
1869 | * preallocation beyond EOF (e.g. the file is open but not modified | |
1870 | * while defrag is in progress). In that case, we need to copy over the | |
1871 | * number of delalloc blocks the data fork in the source inode is | |
1872 | * tracking beyond EOF so that when the fork is truncated away when the | |
1873 | * temporary inode is unlinked we don't underrun the i_delayed_blks | |
1874 | * counter on that inode. | |
1875 | */ | |
1876 | ASSERT(tip->i_delayed_blks == 0); | |
1877 | tip->i_delayed_blks = ip->i_delayed_blks; | |
1878 | ip->i_delayed_blks = 0; | |
1879 | ||
a133d952 DC |
1880 | switch (ip->i_d.di_format) { |
1881 | case XFS_DINODE_FMT_EXTENTS: | |
5d829300 ES |
1882 | /* |
1883 | * If the extents fit in the inode, fix the pointer. Otherwise | |
1884 | * it's already NULL or pointing to the extent. | |
a133d952 | 1885 | */ |
5d829300 ES |
1886 | nextents = xfs_iext_count(&ip->i_df); |
1887 | if (nextents <= XFS_INLINE_EXTS) | |
1888 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; | |
39aff5fd | 1889 | (*src_log_flags) |= XFS_ILOG_DEXT; |
a133d952 DC |
1890 | break; |
1891 | case XFS_DINODE_FMT_BTREE: | |
21b5c978 | 1892 | ASSERT(ip->i_d.di_version < 3 || |
39aff5fd DW |
1893 | (*src_log_flags & XFS_ILOG_DOWNER)); |
1894 | (*src_log_flags) |= XFS_ILOG_DBROOT; | |
a133d952 DC |
1895 | break; |
1896 | } | |
1897 | ||
a133d952 DC |
1898 | switch (tip->i_d.di_format) { |
1899 | case XFS_DINODE_FMT_EXTENTS: | |
5d829300 ES |
1900 | /* |
1901 | * If the extents fit in the inode, fix the pointer. Otherwise | |
1902 | * it's already NULL or pointing to the extent. | |
a133d952 | 1903 | */ |
5d829300 ES |
1904 | nextents = xfs_iext_count(&tip->i_df); |
1905 | if (nextents <= XFS_INLINE_EXTS) | |
1906 | tifp->if_u1.if_extents = tifp->if_u2.if_inline_ext; | |
39aff5fd | 1907 | (*target_log_flags) |= XFS_ILOG_DEXT; |
a133d952 DC |
1908 | break; |
1909 | case XFS_DINODE_FMT_BTREE: | |
39aff5fd | 1910 | (*target_log_flags) |= XFS_ILOG_DBROOT; |
21b5c978 | 1911 | ASSERT(tip->i_d.di_version < 3 || |
39aff5fd | 1912 | (*target_log_flags & XFS_ILOG_DOWNER)); |
a133d952 DC |
1913 | break; |
1914 | } | |
1915 | ||
39aff5fd DW |
1916 | return 0; |
1917 | } | |
1918 | ||
1919 | int | |
1920 | xfs_swap_extents( | |
1921 | struct xfs_inode *ip, /* target inode */ | |
1922 | struct xfs_inode *tip, /* tmp inode */ | |
1923 | struct xfs_swapext *sxp) | |
1924 | { | |
1925 | struct xfs_mount *mp = ip->i_mount; | |
1926 | struct xfs_trans *tp; | |
1927 | struct xfs_bstat *sbp = &sxp->sx_stat; | |
1928 | int src_log_flags, target_log_flags; | |
1929 | int error = 0; | |
1930 | int lock_flags; | |
1931 | struct xfs_ifork *cowfp; | |
1932 | __uint64_t f; | |
1f08af52 | 1933 | int resblks; |
39aff5fd DW |
1934 | |
1935 | /* | |
1936 | * Lock the inodes against other IO, page faults and truncate to | |
1937 | * begin with. Then we can ensure the inodes are flushed and have no | |
1938 | * page cache safely. Once we have done this we can take the ilocks and | |
1939 | * do the rest of the checks. | |
1940 | */ | |
65523218 CH |
1941 | lock_two_nondirectories(VFS_I(ip), VFS_I(tip)); |
1942 | lock_flags = XFS_MMAPLOCK_EXCL; | |
39aff5fd DW |
1943 | xfs_lock_two_inodes(ip, tip, XFS_MMAPLOCK_EXCL); |
1944 | ||
1945 | /* Verify that both files have the same format */ | |
1946 | if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) { | |
1947 | error = -EINVAL; | |
1948 | goto out_unlock; | |
1949 | } | |
1950 | ||
1951 | /* Verify both files are either real-time or non-realtime */ | |
1952 | if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) { | |
1953 | error = -EINVAL; | |
1954 | goto out_unlock; | |
1955 | } | |
1956 | ||
1957 | error = xfs_swap_extent_flush(ip); | |
1958 | if (error) | |
1959 | goto out_unlock; | |
1960 | error = xfs_swap_extent_flush(tip); | |
1961 | if (error) | |
1962 | goto out_unlock; | |
1963 | ||
1f08af52 DW |
1964 | /* |
1965 | * Extent "swapping" with rmap requires a permanent reservation and | |
1966 | * a block reservation because it's really just a remap operation | |
1967 | * performed with log redo items! | |
1968 | */ | |
1969 | if (xfs_sb_version_hasrmapbt(&mp->m_sb)) { | |
1970 | /* | |
1971 | * Conceptually this shouldn't affect the shape of either | |
1972 | * bmbt, but since we atomically move extents one by one, | |
1973 | * we reserve enough space to rebuild both trees. | |
1974 | */ | |
1975 | resblks = XFS_SWAP_RMAP_SPACE_RES(mp, | |
1976 | XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK), | |
1977 | XFS_DATA_FORK) + | |
1978 | XFS_SWAP_RMAP_SPACE_RES(mp, | |
1979 | XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK), | |
1980 | XFS_DATA_FORK); | |
1981 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, | |
1982 | 0, 0, &tp); | |
1983 | } else | |
1984 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, | |
1985 | 0, 0, &tp); | |
39aff5fd DW |
1986 | if (error) |
1987 | goto out_unlock; | |
1988 | ||
1989 | /* | |
1990 | * Lock and join the inodes to the tansaction so that transaction commit | |
1991 | * or cancel will unlock the inodes from this point onwards. | |
1992 | */ | |
1993 | xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL); | |
1994 | lock_flags |= XFS_ILOCK_EXCL; | |
1995 | xfs_trans_ijoin(tp, ip, 0); | |
1996 | xfs_trans_ijoin(tp, tip, 0); | |
1997 | ||
1998 | ||
1999 | /* Verify all data are being swapped */ | |
2000 | if (sxp->sx_offset != 0 || | |
2001 | sxp->sx_length != ip->i_d.di_size || | |
2002 | sxp->sx_length != tip->i_d.di_size) { | |
2003 | error = -EFAULT; | |
2004 | goto out_trans_cancel; | |
2005 | } | |
2006 | ||
2007 | trace_xfs_swap_extent_before(ip, 0); | |
2008 | trace_xfs_swap_extent_before(tip, 1); | |
2009 | ||
2010 | /* check inode formats now that data is flushed */ | |
2011 | error = xfs_swap_extents_check_format(ip, tip); | |
2012 | if (error) { | |
2013 | xfs_notice(mp, | |
2014 | "%s: inode 0x%llx format is incompatible for exchanging.", | |
2015 | __func__, ip->i_ino); | |
2016 | goto out_trans_cancel; | |
2017 | } | |
2018 | ||
2019 | /* | |
2020 | * Compare the current change & modify times with that | |
2021 | * passed in. If they differ, we abort this swap. | |
2022 | * This is the mechanism used to ensure the calling | |
2023 | * process that the file was not changed out from | |
2024 | * under it. | |
2025 | */ | |
2026 | if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) || | |
2027 | (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) || | |
2028 | (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) || | |
2029 | (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) { | |
2030 | error = -EBUSY; | |
2031 | goto out_trans_cancel; | |
2032 | } | |
2033 | ||
2034 | /* | |
2035 | * Note the trickiness in setting the log flags - we set the owner log | |
2036 | * flag on the opposite inode (i.e. the inode we are setting the new | |
2037 | * owner to be) because once we swap the forks and log that, log | |
2038 | * recovery is going to see the fork as owned by the swapped inode, | |
2039 | * not the pre-swapped inodes. | |
2040 | */ | |
2041 | src_log_flags = XFS_ILOG_CORE; | |
2042 | target_log_flags = XFS_ILOG_CORE; | |
2043 | ||
1f08af52 DW |
2044 | if (xfs_sb_version_hasrmapbt(&mp->m_sb)) |
2045 | error = xfs_swap_extent_rmap(&tp, ip, tip); | |
2046 | else | |
2047 | error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags, | |
2048 | &target_log_flags); | |
39aff5fd DW |
2049 | if (error) |
2050 | goto out_trans_cancel; | |
2051 | ||
f0bc4d13 DW |
2052 | /* Do we have to swap reflink flags? */ |
2053 | if ((ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK) ^ | |
2054 | (tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)) { | |
2055 | f = ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK; | |
2056 | ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK; | |
2057 | ip->i_d.di_flags2 |= tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK; | |
2058 | tip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK; | |
2059 | tip->i_d.di_flags2 |= f & XFS_DIFLAG2_REFLINK; | |
2060 | cowfp = ip->i_cowfp; | |
2061 | ip->i_cowfp = tip->i_cowfp; | |
2062 | tip->i_cowfp = cowfp; | |
83104d44 DW |
2063 | xfs_inode_set_cowblocks_tag(ip); |
2064 | xfs_inode_set_cowblocks_tag(tip); | |
f0bc4d13 DW |
2065 | } |
2066 | ||
a133d952 DC |
2067 | xfs_trans_log_inode(tp, ip, src_log_flags); |
2068 | xfs_trans_log_inode(tp, tip, target_log_flags); | |
2069 | ||
2070 | /* | |
2071 | * If this is a synchronous mount, make sure that the | |
2072 | * transaction goes to disk before returning to the user. | |
2073 | */ | |
2074 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
2075 | xfs_trans_set_sync(tp); | |
2076 | ||
70393313 | 2077 | error = xfs_trans_commit(tp); |
a133d952 DC |
2078 | |
2079 | trace_xfs_swap_extent_after(ip, 0); | |
2080 | trace_xfs_swap_extent_after(tip, 1); | |
a133d952 | 2081 | |
65523218 | 2082 | out_unlock: |
81217683 DC |
2083 | xfs_iunlock(ip, lock_flags); |
2084 | xfs_iunlock(tip, lock_flags); | |
65523218 | 2085 | unlock_two_nondirectories(VFS_I(ip), VFS_I(tip)); |
39aff5fd | 2086 | return error; |
a133d952 DC |
2087 | |
2088 | out_trans_cancel: | |
4906e215 | 2089 | xfs_trans_cancel(tp); |
65523218 | 2090 | goto out_unlock; |
a133d952 | 2091 | } |