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); | |
75 | ssize_t size = XFS_FSB_TO_B(mp, count_fsb); | |
76 | ||
77 | if (IS_DAX(VFS_I(ip))) | |
78 | return dax_clear_blocks(VFS_I(ip), block, size); | |
79 | ||
80 | /* | |
81 | * let the block layer decide on the fastest method of | |
82 | * implementing the zeroing. | |
83 | */ | |
84 | return sb_issue_zeroout(mp->m_super, block, count_fsb, GFP_NOFS); | |
85 | ||
86 | } | |
87 | ||
68988114 DC |
88 | /* |
89 | * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi | |
90 | * caller. Frees all the extents that need freeing, which must be done | |
91 | * last due to locking considerations. We never free any extents in | |
92 | * the first transaction. | |
93 | * | |
f6106efa ES |
94 | * If an inode *ip is provided, rejoin it to the transaction if |
95 | * the transaction was committed. | |
68988114 DC |
96 | */ |
97 | int /* error */ | |
98 | xfs_bmap_finish( | |
8d99fe92 BF |
99 | struct xfs_trans **tp, /* transaction pointer addr */ |
100 | struct xfs_bmap_free *flist, /* i/o: list extents to free */ | |
f6106efa | 101 | struct xfs_inode *ip) |
68988114 | 102 | { |
8d99fe92 BF |
103 | struct xfs_efd_log_item *efd; /* extent free data */ |
104 | struct xfs_efi_log_item *efi; /* extent free intention */ | |
105 | int error; /* error return value */ | |
f6106efa | 106 | int committed;/* xact committed or not */ |
8d99fe92 BF |
107 | struct xfs_bmap_free_item *free; /* free extent item */ |
108 | struct xfs_bmap_free_item *next; /* next item on free list */ | |
68988114 DC |
109 | |
110 | ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES); | |
f6106efa | 111 | if (flist->xbf_count == 0) |
68988114 | 112 | return 0; |
f6106efa | 113 | |
2e6db6c4 | 114 | efi = xfs_trans_get_efi(*tp, flist->xbf_count); |
68988114 | 115 | for (free = flist->xbf_first; free; free = free->xbfi_next) |
2e6db6c4 | 116 | xfs_trans_log_efi_extent(*tp, efi, free->xbfi_startblock, |
68988114 | 117 | free->xbfi_blockcount); |
3d3c8b52 | 118 | |
f6106efa | 119 | error = __xfs_trans_roll(tp, ip, &committed); |
8d99fe92 BF |
120 | if (error) { |
121 | /* | |
122 | * If the transaction was committed, drop the EFD reference | |
123 | * since we're bailing out of here. The other reference is | |
124 | * dropped when the EFI hits the AIL. | |
125 | * | |
126 | * If the transaction was not committed, the EFI is freed by the | |
127 | * EFI item unlock handler on abort. Also, we have a new | |
128 | * transaction so we should return committed=1 even though we're | |
129 | * returning an error. | |
130 | */ | |
f6106efa | 131 | if (committed) { |
8d99fe92 BF |
132 | xfs_efi_release(efi); |
133 | xfs_force_shutdown((*tp)->t_mountp, | |
134 | (error == -EFSCORRUPTED) ? | |
135 | SHUTDOWN_CORRUPT_INCORE : | |
136 | SHUTDOWN_META_IO_ERROR); | |
8d99fe92 | 137 | } |
68988114 | 138 | return error; |
8d99fe92 | 139 | } |
68988114 | 140 | |
6bc43af3 BF |
141 | /* |
142 | * Get an EFD and free each extent in the list, logging to the EFD in | |
143 | * the process. The remaining bmap free list is cleaned up by the caller | |
144 | * on error. | |
145 | */ | |
2e6db6c4 | 146 | efd = xfs_trans_get_efd(*tp, efi, flist->xbf_count); |
68988114 DC |
147 | for (free = flist->xbf_first; free != NULL; free = next) { |
148 | next = free->xbfi_next; | |
8d99fe92 | 149 | |
6bc43af3 BF |
150 | error = xfs_trans_free_extent(*tp, efd, free->xbfi_startblock, |
151 | free->xbfi_blockcount); | |
8d99fe92 BF |
152 | if (error) |
153 | return error; | |
154 | ||
68988114 DC |
155 | xfs_bmap_del_free(flist, NULL, free); |
156 | } | |
8d99fe92 | 157 | |
68988114 DC |
158 | return 0; |
159 | } | |
160 | ||
161 | int | |
162 | xfs_bmap_rtalloc( | |
163 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ | |
164 | { | |
165 | xfs_alloctype_t atype = 0; /* type for allocation routines */ | |
166 | int error; /* error return value */ | |
167 | xfs_mount_t *mp; /* mount point structure */ | |
168 | xfs_extlen_t prod = 0; /* product factor for allocators */ | |
169 | xfs_extlen_t ralen = 0; /* realtime allocation length */ | |
170 | xfs_extlen_t align; /* minimum allocation alignment */ | |
171 | xfs_rtblock_t rtb; | |
172 | ||
173 | mp = ap->ip->i_mount; | |
174 | align = xfs_get_extsz_hint(ap->ip); | |
175 | prod = align / mp->m_sb.sb_rextsize; | |
176 | error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, | |
177 | align, 1, ap->eof, 0, | |
178 | ap->conv, &ap->offset, &ap->length); | |
179 | if (error) | |
180 | return error; | |
181 | ASSERT(ap->length); | |
182 | ASSERT(ap->length % mp->m_sb.sb_rextsize == 0); | |
183 | ||
184 | /* | |
185 | * If the offset & length are not perfectly aligned | |
186 | * then kill prod, it will just get us in trouble. | |
187 | */ | |
188 | if (do_mod(ap->offset, align) || ap->length % align) | |
189 | prod = 1; | |
190 | /* | |
191 | * Set ralen to be the actual requested length in rtextents. | |
192 | */ | |
193 | ralen = ap->length / mp->m_sb.sb_rextsize; | |
194 | /* | |
195 | * If the old value was close enough to MAXEXTLEN that | |
196 | * we rounded up to it, cut it back so it's valid again. | |
197 | * Note that if it's a really large request (bigger than | |
198 | * MAXEXTLEN), we don't hear about that number, and can't | |
199 | * adjust the starting point to match it. | |
200 | */ | |
201 | if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN) | |
202 | ralen = MAXEXTLEN / mp->m_sb.sb_rextsize; | |
203 | ||
204 | /* | |
205 | * Lock out other modifications to the RT bitmap inode. | |
206 | */ | |
207 | xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL); | |
208 | xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL); | |
209 | ||
210 | /* | |
211 | * If it's an allocation to an empty file at offset 0, | |
212 | * pick an extent that will space things out in the rt area. | |
213 | */ | |
214 | if (ap->eof && ap->offset == 0) { | |
215 | xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */ | |
216 | ||
217 | error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx); | |
218 | if (error) | |
219 | return error; | |
220 | ap->blkno = rtx * mp->m_sb.sb_rextsize; | |
221 | } else { | |
222 | ap->blkno = 0; | |
223 | } | |
224 | ||
225 | xfs_bmap_adjacent(ap); | |
226 | ||
227 | /* | |
228 | * Realtime allocation, done through xfs_rtallocate_extent. | |
229 | */ | |
230 | atype = ap->blkno == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO; | |
231 | do_div(ap->blkno, mp->m_sb.sb_rextsize); | |
232 | rtb = ap->blkno; | |
233 | ap->length = ralen; | |
234 | if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length, | |
235 | &ralen, atype, ap->wasdel, prod, &rtb))) | |
236 | return error; | |
237 | if (rtb == NULLFSBLOCK && prod > 1 && | |
238 | (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, | |
239 | ap->length, &ralen, atype, | |
240 | ap->wasdel, 1, &rtb))) | |
241 | return error; | |
242 | ap->blkno = rtb; | |
243 | if (ap->blkno != NULLFSBLOCK) { | |
244 | ap->blkno *= mp->m_sb.sb_rextsize; | |
245 | ralen *= mp->m_sb.sb_rextsize; | |
246 | ap->length = ralen; | |
247 | ap->ip->i_d.di_nblocks += ralen; | |
248 | xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE); | |
249 | if (ap->wasdel) | |
250 | ap->ip->i_delayed_blks -= ralen; | |
251 | /* | |
252 | * Adjust the disk quota also. This was reserved | |
253 | * earlier. | |
254 | */ | |
255 | xfs_trans_mod_dquot_byino(ap->tp, ap->ip, | |
256 | ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT : | |
257 | XFS_TRANS_DQ_RTBCOUNT, (long) ralen); | |
3fbbbea3 DC |
258 | |
259 | /* Zero the extent if we were asked to do so */ | |
260 | if (ap->userdata & XFS_ALLOC_USERDATA_ZERO) { | |
261 | error = xfs_zero_extent(ap->ip, ap->blkno, ap->length); | |
262 | if (error) | |
263 | return error; | |
264 | } | |
68988114 DC |
265 | } else { |
266 | ap->length = 0; | |
267 | } | |
268 | return 0; | |
269 | } | |
270 | ||
68988114 DC |
271 | /* |
272 | * Check if the endoff is outside the last extent. If so the caller will grow | |
273 | * the allocation to a stripe unit boundary. All offsets are considered outside | |
274 | * the end of file for an empty fork, so 1 is returned in *eof in that case. | |
275 | */ | |
276 | int | |
277 | xfs_bmap_eof( | |
278 | struct xfs_inode *ip, | |
279 | xfs_fileoff_t endoff, | |
280 | int whichfork, | |
281 | int *eof) | |
282 | { | |
283 | struct xfs_bmbt_irec rec; | |
284 | int error; | |
285 | ||
286 | error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof); | |
287 | if (error || *eof) | |
288 | return error; | |
289 | ||
290 | *eof = endoff >= rec.br_startoff + rec.br_blockcount; | |
291 | return 0; | |
292 | } | |
293 | ||
294 | /* | |
295 | * Extent tree block counting routines. | |
296 | */ | |
297 | ||
298 | /* | |
299 | * Count leaf blocks given a range of extent records. | |
300 | */ | |
301 | STATIC void | |
302 | xfs_bmap_count_leaves( | |
303 | xfs_ifork_t *ifp, | |
304 | xfs_extnum_t idx, | |
305 | int numrecs, | |
306 | int *count) | |
307 | { | |
308 | int b; | |
309 | ||
310 | for (b = 0; b < numrecs; b++) { | |
311 | xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b); | |
312 | *count += xfs_bmbt_get_blockcount(frp); | |
313 | } | |
314 | } | |
315 | ||
316 | /* | |
317 | * Count leaf blocks given a range of extent records originally | |
318 | * in btree format. | |
319 | */ | |
320 | STATIC void | |
321 | xfs_bmap_disk_count_leaves( | |
322 | struct xfs_mount *mp, | |
323 | struct xfs_btree_block *block, | |
324 | int numrecs, | |
325 | int *count) | |
326 | { | |
327 | int b; | |
328 | xfs_bmbt_rec_t *frp; | |
329 | ||
330 | for (b = 1; b <= numrecs; b++) { | |
331 | frp = XFS_BMBT_REC_ADDR(mp, block, b); | |
332 | *count += xfs_bmbt_disk_get_blockcount(frp); | |
333 | } | |
334 | } | |
335 | ||
336 | /* | |
337 | * Recursively walks each level of a btree | |
8be11e92 | 338 | * to count total fsblocks in use. |
68988114 DC |
339 | */ |
340 | STATIC int /* error */ | |
341 | xfs_bmap_count_tree( | |
342 | xfs_mount_t *mp, /* file system mount point */ | |
343 | xfs_trans_t *tp, /* transaction pointer */ | |
344 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
345 | xfs_fsblock_t blockno, /* file system block number */ | |
346 | int levelin, /* level in btree */ | |
347 | int *count) /* Count of blocks */ | |
348 | { | |
349 | int error; | |
350 | xfs_buf_t *bp, *nbp; | |
351 | int level = levelin; | |
352 | __be64 *pp; | |
353 | xfs_fsblock_t bno = blockno; | |
354 | xfs_fsblock_t nextbno; | |
355 | struct xfs_btree_block *block, *nextblock; | |
356 | int numrecs; | |
357 | ||
358 | error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF, | |
359 | &xfs_bmbt_buf_ops); | |
360 | if (error) | |
361 | return error; | |
362 | *count += 1; | |
363 | block = XFS_BUF_TO_BLOCK(bp); | |
364 | ||
365 | if (--level) { | |
366 | /* Not at node above leaves, count this level of nodes */ | |
367 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); | |
368 | while (nextbno != NULLFSBLOCK) { | |
369 | error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp, | |
370 | XFS_BMAP_BTREE_REF, | |
371 | &xfs_bmbt_buf_ops); | |
372 | if (error) | |
373 | return error; | |
374 | *count += 1; | |
375 | nextblock = XFS_BUF_TO_BLOCK(nbp); | |
376 | nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib); | |
377 | xfs_trans_brelse(tp, nbp); | |
378 | } | |
379 | ||
380 | /* Dive to the next level */ | |
381 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]); | |
382 | bno = be64_to_cpu(*pp); | |
383 | if (unlikely((error = | |
384 | xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) { | |
385 | xfs_trans_brelse(tp, bp); | |
386 | XFS_ERROR_REPORT("xfs_bmap_count_tree(1)", | |
387 | XFS_ERRLEVEL_LOW, mp); | |
2451337d | 388 | return -EFSCORRUPTED; |
68988114 DC |
389 | } |
390 | xfs_trans_brelse(tp, bp); | |
391 | } else { | |
392 | /* count all level 1 nodes and their leaves */ | |
393 | for (;;) { | |
394 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); | |
395 | numrecs = be16_to_cpu(block->bb_numrecs); | |
396 | xfs_bmap_disk_count_leaves(mp, block, numrecs, count); | |
397 | xfs_trans_brelse(tp, bp); | |
398 | if (nextbno == NULLFSBLOCK) | |
399 | break; | |
400 | bno = nextbno; | |
401 | error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, | |
402 | XFS_BMAP_BTREE_REF, | |
403 | &xfs_bmbt_buf_ops); | |
404 | if (error) | |
405 | return error; | |
406 | *count += 1; | |
407 | block = XFS_BUF_TO_BLOCK(bp); | |
408 | } | |
409 | } | |
410 | return 0; | |
411 | } | |
412 | ||
413 | /* | |
414 | * Count fsblocks of the given fork. | |
415 | */ | |
416 | int /* error */ | |
417 | xfs_bmap_count_blocks( | |
418 | xfs_trans_t *tp, /* transaction pointer */ | |
419 | xfs_inode_t *ip, /* incore inode */ | |
420 | int whichfork, /* data or attr fork */ | |
421 | int *count) /* out: count of blocks */ | |
422 | { | |
423 | struct xfs_btree_block *block; /* current btree block */ | |
424 | xfs_fsblock_t bno; /* block # of "block" */ | |
425 | xfs_ifork_t *ifp; /* fork structure */ | |
426 | int level; /* btree level, for checking */ | |
427 | xfs_mount_t *mp; /* file system mount structure */ | |
428 | __be64 *pp; /* pointer to block address */ | |
429 | ||
430 | bno = NULLFSBLOCK; | |
431 | mp = ip->i_mount; | |
432 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
433 | if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) { | |
434 | xfs_bmap_count_leaves(ifp, 0, | |
435 | ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t), | |
436 | count); | |
437 | return 0; | |
438 | } | |
439 | ||
440 | /* | |
441 | * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out. | |
442 | */ | |
443 | block = ifp->if_broot; | |
444 | level = be16_to_cpu(block->bb_level); | |
445 | ASSERT(level > 0); | |
446 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes); | |
447 | bno = be64_to_cpu(*pp); | |
d5cf09ba | 448 | ASSERT(bno != NULLFSBLOCK); |
68988114 DC |
449 | ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount); |
450 | ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks); | |
451 | ||
452 | if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) { | |
453 | XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW, | |
454 | mp); | |
2451337d | 455 | return -EFSCORRUPTED; |
68988114 DC |
456 | } |
457 | ||
458 | return 0; | |
459 | } | |
460 | ||
461 | /* | |
462 | * returns 1 for success, 0 if we failed to map the extent. | |
463 | */ | |
464 | STATIC int | |
465 | xfs_getbmapx_fix_eof_hole( | |
466 | xfs_inode_t *ip, /* xfs incore inode pointer */ | |
467 | struct getbmapx *out, /* output structure */ | |
468 | int prealloced, /* this is a file with | |
469 | * preallocated data space */ | |
470 | __int64_t end, /* last block requested */ | |
471 | xfs_fsblock_t startblock) | |
472 | { | |
473 | __int64_t fixlen; | |
474 | xfs_mount_t *mp; /* file system mount point */ | |
475 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
476 | xfs_extnum_t lastx; /* last extent pointer */ | |
477 | xfs_fileoff_t fileblock; | |
478 | ||
479 | if (startblock == HOLESTARTBLOCK) { | |
480 | mp = ip->i_mount; | |
481 | out->bmv_block = -1; | |
482 | fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip))); | |
483 | fixlen -= out->bmv_offset; | |
484 | if (prealloced && out->bmv_offset + out->bmv_length == end) { | |
485 | /* Came to hole at EOF. Trim it. */ | |
486 | if (fixlen <= 0) | |
487 | return 0; | |
488 | out->bmv_length = fixlen; | |
489 | } | |
490 | } else { | |
491 | if (startblock == DELAYSTARTBLOCK) | |
492 | out->bmv_block = -2; | |
493 | else | |
494 | out->bmv_block = xfs_fsb_to_db(ip, startblock); | |
495 | fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset); | |
496 | ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK); | |
497 | if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) && | |
498 | (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1)) | |
499 | out->bmv_oflags |= BMV_OF_LAST; | |
500 | } | |
501 | ||
502 | return 1; | |
503 | } | |
504 | ||
505 | /* | |
506 | * Get inode's extents as described in bmv, and format for output. | |
507 | * Calls formatter to fill the user's buffer until all extents | |
508 | * are mapped, until the passed-in bmv->bmv_count slots have | |
509 | * been filled, or until the formatter short-circuits the loop, | |
510 | * if it is tracking filled-in extents on its own. | |
511 | */ | |
512 | int /* error code */ | |
513 | xfs_getbmap( | |
514 | xfs_inode_t *ip, | |
515 | struct getbmapx *bmv, /* user bmap structure */ | |
516 | xfs_bmap_format_t formatter, /* format to user */ | |
517 | void *arg) /* formatter arg */ | |
518 | { | |
519 | __int64_t bmvend; /* last block requested */ | |
520 | int error = 0; /* return value */ | |
521 | __int64_t fixlen; /* length for -1 case */ | |
522 | int i; /* extent number */ | |
523 | int lock; /* lock state */ | |
524 | xfs_bmbt_irec_t *map; /* buffer for user's data */ | |
525 | xfs_mount_t *mp; /* file system mount point */ | |
526 | int nex; /* # of user extents can do */ | |
527 | int nexleft; /* # of user extents left */ | |
528 | int subnex; /* # of bmapi's can do */ | |
529 | int nmap; /* number of map entries */ | |
530 | struct getbmapx *out; /* output structure */ | |
531 | int whichfork; /* data or attr fork */ | |
532 | int prealloced; /* this is a file with | |
533 | * preallocated data space */ | |
534 | int iflags; /* interface flags */ | |
535 | int bmapi_flags; /* flags for xfs_bmapi */ | |
536 | int cur_ext = 0; | |
537 | ||
538 | mp = ip->i_mount; | |
539 | iflags = bmv->bmv_iflags; | |
540 | whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK; | |
541 | ||
542 | if (whichfork == XFS_ATTR_FORK) { | |
543 | if (XFS_IFORK_Q(ip)) { | |
544 | if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS && | |
545 | ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE && | |
546 | ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL) | |
2451337d | 547 | return -EINVAL; |
68988114 DC |
548 | } else if (unlikely( |
549 | ip->i_d.di_aformat != 0 && | |
550 | ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) { | |
551 | XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW, | |
552 | ip->i_mount); | |
2451337d | 553 | return -EFSCORRUPTED; |
68988114 DC |
554 | } |
555 | ||
556 | prealloced = 0; | |
557 | fixlen = 1LL << 32; | |
558 | } else { | |
559 | if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS && | |
560 | ip->i_d.di_format != XFS_DINODE_FMT_BTREE && | |
561 | ip->i_d.di_format != XFS_DINODE_FMT_LOCAL) | |
2451337d | 562 | return -EINVAL; |
68988114 DC |
563 | |
564 | if (xfs_get_extsz_hint(ip) || | |
565 | ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){ | |
566 | prealloced = 1; | |
567 | fixlen = mp->m_super->s_maxbytes; | |
568 | } else { | |
569 | prealloced = 0; | |
570 | fixlen = XFS_ISIZE(ip); | |
571 | } | |
572 | } | |
573 | ||
574 | if (bmv->bmv_length == -1) { | |
575 | fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen)); | |
576 | bmv->bmv_length = | |
577 | max_t(__int64_t, fixlen - bmv->bmv_offset, 0); | |
578 | } else if (bmv->bmv_length == 0) { | |
579 | bmv->bmv_entries = 0; | |
580 | return 0; | |
581 | } else if (bmv->bmv_length < 0) { | |
2451337d | 582 | return -EINVAL; |
68988114 DC |
583 | } |
584 | ||
585 | nex = bmv->bmv_count - 1; | |
586 | if (nex <= 0) | |
2451337d | 587 | return -EINVAL; |
68988114 DC |
588 | bmvend = bmv->bmv_offset + bmv->bmv_length; |
589 | ||
590 | ||
591 | if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx)) | |
2451337d | 592 | return -ENOMEM; |
fdd3ccee DC |
593 | out = kmem_zalloc_large(bmv->bmv_count * sizeof(struct getbmapx), 0); |
594 | if (!out) | |
2451337d | 595 | return -ENOMEM; |
68988114 DC |
596 | |
597 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
efa70be1 CH |
598 | if (whichfork == XFS_DATA_FORK) { |
599 | if (!(iflags & BMV_IF_DELALLOC) && | |
600 | (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) { | |
2451337d | 601 | error = filemap_write_and_wait(VFS_I(ip)->i_mapping); |
68988114 DC |
602 | if (error) |
603 | goto out_unlock_iolock; | |
efa70be1 CH |
604 | |
605 | /* | |
606 | * Even after flushing the inode, there can still be | |
607 | * delalloc blocks on the inode beyond EOF due to | |
608 | * speculative preallocation. These are not removed | |
609 | * until the release function is called or the inode | |
610 | * is inactivated. Hence we cannot assert here that | |
611 | * ip->i_delayed_blks == 0. | |
612 | */ | |
68988114 | 613 | } |
68988114 | 614 | |
efa70be1 CH |
615 | lock = xfs_ilock_data_map_shared(ip); |
616 | } else { | |
617 | lock = xfs_ilock_attr_map_shared(ip); | |
618 | } | |
68988114 DC |
619 | |
620 | /* | |
621 | * Don't let nex be bigger than the number of extents | |
622 | * we can have assuming alternating holes and real extents. | |
623 | */ | |
624 | if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1) | |
625 | nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1; | |
626 | ||
627 | bmapi_flags = xfs_bmapi_aflag(whichfork); | |
628 | if (!(iflags & BMV_IF_PREALLOC)) | |
629 | bmapi_flags |= XFS_BMAPI_IGSTATE; | |
630 | ||
631 | /* | |
632 | * Allocate enough space to handle "subnex" maps at a time. | |
633 | */ | |
2451337d | 634 | error = -ENOMEM; |
68988114 DC |
635 | subnex = 16; |
636 | map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS); | |
637 | if (!map) | |
638 | goto out_unlock_ilock; | |
639 | ||
640 | bmv->bmv_entries = 0; | |
641 | ||
642 | if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 && | |
643 | (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) { | |
644 | error = 0; | |
645 | goto out_free_map; | |
646 | } | |
647 | ||
648 | nexleft = nex; | |
649 | ||
650 | do { | |
651 | nmap = (nexleft > subnex) ? subnex : nexleft; | |
652 | error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset), | |
653 | XFS_BB_TO_FSB(mp, bmv->bmv_length), | |
654 | map, &nmap, bmapi_flags); | |
655 | if (error) | |
656 | goto out_free_map; | |
657 | ASSERT(nmap <= subnex); | |
658 | ||
659 | for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) { | |
660 | out[cur_ext].bmv_oflags = 0; | |
661 | if (map[i].br_state == XFS_EXT_UNWRITTEN) | |
662 | out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC; | |
663 | else if (map[i].br_startblock == DELAYSTARTBLOCK) | |
664 | out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC; | |
665 | out[cur_ext].bmv_offset = | |
666 | XFS_FSB_TO_BB(mp, map[i].br_startoff); | |
667 | out[cur_ext].bmv_length = | |
668 | XFS_FSB_TO_BB(mp, map[i].br_blockcount); | |
669 | out[cur_ext].bmv_unused1 = 0; | |
670 | out[cur_ext].bmv_unused2 = 0; | |
671 | ||
672 | /* | |
673 | * delayed allocation extents that start beyond EOF can | |
674 | * occur due to speculative EOF allocation when the | |
675 | * delalloc extent is larger than the largest freespace | |
676 | * extent at conversion time. These extents cannot be | |
677 | * converted by data writeback, so can exist here even | |
678 | * if we are not supposed to be finding delalloc | |
679 | * extents. | |
680 | */ | |
681 | if (map[i].br_startblock == DELAYSTARTBLOCK && | |
682 | map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip))) | |
683 | ASSERT((iflags & BMV_IF_DELALLOC) != 0); | |
684 | ||
685 | if (map[i].br_startblock == HOLESTARTBLOCK && | |
686 | whichfork == XFS_ATTR_FORK) { | |
687 | /* came to the end of attribute fork */ | |
688 | out[cur_ext].bmv_oflags |= BMV_OF_LAST; | |
689 | goto out_free_map; | |
690 | } | |
691 | ||
692 | if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext], | |
693 | prealloced, bmvend, | |
694 | map[i].br_startblock)) | |
695 | goto out_free_map; | |
696 | ||
697 | bmv->bmv_offset = | |
698 | out[cur_ext].bmv_offset + | |
699 | out[cur_ext].bmv_length; | |
700 | bmv->bmv_length = | |
701 | max_t(__int64_t, 0, bmvend - bmv->bmv_offset); | |
702 | ||
703 | /* | |
704 | * In case we don't want to return the hole, | |
705 | * don't increase cur_ext so that we can reuse | |
706 | * it in the next loop. | |
707 | */ | |
708 | if ((iflags & BMV_IF_NO_HOLES) && | |
709 | map[i].br_startblock == HOLESTARTBLOCK) { | |
710 | memset(&out[cur_ext], 0, sizeof(out[cur_ext])); | |
711 | continue; | |
712 | } | |
713 | ||
714 | nexleft--; | |
715 | bmv->bmv_entries++; | |
716 | cur_ext++; | |
717 | } | |
718 | } while (nmap && nexleft && bmv->bmv_length); | |
719 | ||
720 | out_free_map: | |
721 | kmem_free(map); | |
722 | out_unlock_ilock: | |
01f4f327 | 723 | xfs_iunlock(ip, lock); |
68988114 DC |
724 | out_unlock_iolock: |
725 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
726 | ||
727 | for (i = 0; i < cur_ext; i++) { | |
728 | int full = 0; /* user array is full */ | |
729 | ||
730 | /* format results & advance arg */ | |
731 | error = formatter(&arg, &out[i], &full); | |
732 | if (error || full) | |
733 | break; | |
734 | } | |
735 | ||
fdd3ccee | 736 | kmem_free(out); |
68988114 DC |
737 | return error; |
738 | } | |
739 | ||
740 | /* | |
741 | * dead simple method of punching delalyed allocation blocks from a range in | |
742 | * the inode. Walks a block at a time so will be slow, but is only executed in | |
ad4809bf | 743 | * rare error cases so the overhead is not critical. This will always punch out |
68988114 DC |
744 | * both the start and end blocks, even if the ranges only partially overlap |
745 | * them, so it is up to the caller to ensure that partial blocks are not | |
746 | * passed in. | |
747 | */ | |
748 | int | |
749 | xfs_bmap_punch_delalloc_range( | |
750 | struct xfs_inode *ip, | |
751 | xfs_fileoff_t start_fsb, | |
752 | xfs_fileoff_t length) | |
753 | { | |
754 | xfs_fileoff_t remaining = length; | |
755 | int error = 0; | |
756 | ||
757 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
758 | ||
759 | do { | |
760 | int done; | |
761 | xfs_bmbt_irec_t imap; | |
762 | int nimaps = 1; | |
763 | xfs_fsblock_t firstblock; | |
764 | xfs_bmap_free_t flist; | |
765 | ||
766 | /* | |
767 | * Map the range first and check that it is a delalloc extent | |
768 | * before trying to unmap the range. Otherwise we will be | |
769 | * trying to remove a real extent (which requires a | |
770 | * transaction) or a hole, which is probably a bad idea... | |
771 | */ | |
772 | error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps, | |
773 | XFS_BMAPI_ENTIRE); | |
774 | ||
775 | if (error) { | |
776 | /* something screwed, just bail */ | |
777 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
778 | xfs_alert(ip->i_mount, | |
779 | "Failed delalloc mapping lookup ino %lld fsb %lld.", | |
780 | ip->i_ino, start_fsb); | |
781 | } | |
782 | break; | |
783 | } | |
784 | if (!nimaps) { | |
785 | /* nothing there */ | |
786 | goto next_block; | |
787 | } | |
788 | if (imap.br_startblock != DELAYSTARTBLOCK) { | |
789 | /* been converted, ignore */ | |
790 | goto next_block; | |
791 | } | |
792 | WARN_ON(imap.br_blockcount == 0); | |
793 | ||
794 | /* | |
795 | * Note: while we initialise the firstblock/flist pair, they | |
796 | * should never be used because blocks should never be | |
797 | * allocated or freed for a delalloc extent and hence we need | |
798 | * don't cancel or finish them after the xfs_bunmapi() call. | |
799 | */ | |
800 | xfs_bmap_init(&flist, &firstblock); | |
801 | error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock, | |
802 | &flist, &done); | |
803 | if (error) | |
804 | break; | |
805 | ||
806 | ASSERT(!flist.xbf_count && !flist.xbf_first); | |
807 | next_block: | |
808 | start_fsb++; | |
809 | remaining--; | |
810 | } while(remaining > 0); | |
811 | ||
812 | return error; | |
813 | } | |
c24b5dfa DC |
814 | |
815 | /* | |
816 | * Test whether it is appropriate to check an inode for and free post EOF | |
817 | * blocks. The 'force' parameter determines whether we should also consider | |
818 | * regular files that are marked preallocated or append-only. | |
819 | */ | |
820 | bool | |
821 | xfs_can_free_eofblocks(struct xfs_inode *ip, bool force) | |
822 | { | |
823 | /* prealloc/delalloc exists only on regular files */ | |
824 | if (!S_ISREG(ip->i_d.di_mode)) | |
825 | return false; | |
826 | ||
827 | /* | |
828 | * Zero sized files with no cached pages and delalloc blocks will not | |
829 | * have speculative prealloc/delalloc blocks to remove. | |
830 | */ | |
831 | if (VFS_I(ip)->i_size == 0 && | |
2667c6f9 | 832 | VFS_I(ip)->i_mapping->nrpages == 0 && |
c24b5dfa DC |
833 | ip->i_delayed_blks == 0) |
834 | return false; | |
835 | ||
836 | /* If we haven't read in the extent list, then don't do it now. */ | |
837 | if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) | |
838 | return false; | |
839 | ||
840 | /* | |
841 | * Do not free real preallocated or append-only files unless the file | |
842 | * has delalloc blocks and we are forced to remove them. | |
843 | */ | |
844 | if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) | |
845 | if (!force || ip->i_delayed_blks == 0) | |
846 | return false; | |
847 | ||
848 | return true; | |
849 | } | |
850 | ||
851 | /* | |
852 | * This is called by xfs_inactive to free any blocks beyond eof | |
853 | * when the link count isn't zero and by xfs_dm_punch_hole() when | |
854 | * punching a hole to EOF. | |
855 | */ | |
856 | int | |
857 | xfs_free_eofblocks( | |
858 | xfs_mount_t *mp, | |
859 | xfs_inode_t *ip, | |
860 | bool need_iolock) | |
861 | { | |
862 | xfs_trans_t *tp; | |
863 | int error; | |
864 | xfs_fileoff_t end_fsb; | |
865 | xfs_fileoff_t last_fsb; | |
866 | xfs_filblks_t map_len; | |
867 | int nimaps; | |
868 | xfs_bmbt_irec_t imap; | |
869 | ||
870 | /* | |
871 | * Figure out if there are any blocks beyond the end | |
872 | * of the file. If not, then there is nothing to do. | |
873 | */ | |
874 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip)); | |
875 | last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); | |
876 | if (last_fsb <= end_fsb) | |
877 | return 0; | |
878 | map_len = last_fsb - end_fsb; | |
879 | ||
880 | nimaps = 1; | |
881 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
882 | error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0); | |
883 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
884 | ||
885 | if (!error && (nimaps != 0) && | |
886 | (imap.br_startblock != HOLESTARTBLOCK || | |
887 | ip->i_delayed_blks)) { | |
888 | /* | |
889 | * Attach the dquots to the inode up front. | |
890 | */ | |
891 | error = xfs_qm_dqattach(ip, 0); | |
892 | if (error) | |
893 | return error; | |
894 | ||
895 | /* | |
896 | * There are blocks after the end of file. | |
897 | * Free them up now by truncating the file to | |
898 | * its current size. | |
899 | */ | |
900 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
901 | ||
902 | if (need_iolock) { | |
903 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { | |
4906e215 | 904 | xfs_trans_cancel(tp); |
2451337d | 905 | return -EAGAIN; |
c24b5dfa DC |
906 | } |
907 | } | |
908 | ||
3d3c8b52 | 909 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0); |
c24b5dfa DC |
910 | if (error) { |
911 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
4906e215 | 912 | xfs_trans_cancel(tp); |
c24b5dfa DC |
913 | if (need_iolock) |
914 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
915 | return error; | |
916 | } | |
917 | ||
918 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
919 | xfs_trans_ijoin(tp, ip, 0); | |
920 | ||
921 | /* | |
922 | * Do not update the on-disk file size. If we update the | |
923 | * on-disk file size and then the system crashes before the | |
924 | * contents of the file are flushed to disk then the files | |
925 | * may be full of holes (ie NULL files bug). | |
926 | */ | |
927 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, | |
928 | XFS_ISIZE(ip)); | |
929 | if (error) { | |
930 | /* | |
931 | * If we get an error at this point we simply don't | |
932 | * bother truncating the file. | |
933 | */ | |
4906e215 | 934 | xfs_trans_cancel(tp); |
c24b5dfa | 935 | } else { |
70393313 | 936 | error = xfs_trans_commit(tp); |
c24b5dfa DC |
937 | if (!error) |
938 | xfs_inode_clear_eofblocks_tag(ip); | |
939 | } | |
940 | ||
941 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
942 | if (need_iolock) | |
943 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
944 | } | |
945 | return error; | |
946 | } | |
947 | ||
83aee9e4 | 948 | int |
c24b5dfa | 949 | xfs_alloc_file_space( |
83aee9e4 | 950 | struct xfs_inode *ip, |
c24b5dfa DC |
951 | xfs_off_t offset, |
952 | xfs_off_t len, | |
5f8aca8b | 953 | int alloc_type) |
c24b5dfa DC |
954 | { |
955 | xfs_mount_t *mp = ip->i_mount; | |
956 | xfs_off_t count; | |
957 | xfs_filblks_t allocated_fsb; | |
958 | xfs_filblks_t allocatesize_fsb; | |
959 | xfs_extlen_t extsz, temp; | |
960 | xfs_fileoff_t startoffset_fsb; | |
961 | xfs_fsblock_t firstfsb; | |
962 | int nimaps; | |
963 | int quota_flag; | |
964 | int rt; | |
965 | xfs_trans_t *tp; | |
966 | xfs_bmbt_irec_t imaps[1], *imapp; | |
967 | xfs_bmap_free_t free_list; | |
968 | uint qblocks, resblks, resrtextents; | |
c24b5dfa DC |
969 | int error; |
970 | ||
971 | trace_xfs_alloc_file_space(ip); | |
972 | ||
973 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 974 | return -EIO; |
c24b5dfa DC |
975 | |
976 | error = xfs_qm_dqattach(ip, 0); | |
977 | if (error) | |
978 | return error; | |
979 | ||
980 | if (len <= 0) | |
2451337d | 981 | return -EINVAL; |
c24b5dfa DC |
982 | |
983 | rt = XFS_IS_REALTIME_INODE(ip); | |
984 | extsz = xfs_get_extsz_hint(ip); | |
985 | ||
986 | count = len; | |
987 | imapp = &imaps[0]; | |
988 | nimaps = 1; | |
989 | startoffset_fsb = XFS_B_TO_FSBT(mp, offset); | |
990 | allocatesize_fsb = XFS_B_TO_FSB(mp, count); | |
991 | ||
992 | /* | |
993 | * Allocate file space until done or until there is an error | |
994 | */ | |
995 | while (allocatesize_fsb && !error) { | |
996 | xfs_fileoff_t s, e; | |
997 | ||
998 | /* | |
999 | * Determine space reservations for data/realtime. | |
1000 | */ | |
1001 | if (unlikely(extsz)) { | |
1002 | s = startoffset_fsb; | |
1003 | do_div(s, extsz); | |
1004 | s *= extsz; | |
1005 | e = startoffset_fsb + allocatesize_fsb; | |
1006 | if ((temp = do_mod(startoffset_fsb, extsz))) | |
1007 | e += temp; | |
1008 | if ((temp = do_mod(e, extsz))) | |
1009 | e += extsz - temp; | |
1010 | } else { | |
1011 | s = 0; | |
1012 | e = allocatesize_fsb; | |
1013 | } | |
1014 | ||
1015 | /* | |
1016 | * The transaction reservation is limited to a 32-bit block | |
1017 | * count, hence we need to limit the number of blocks we are | |
1018 | * trying to reserve to avoid an overflow. We can't allocate | |
1019 | * more than @nimaps extents, and an extent is limited on disk | |
1020 | * to MAXEXTLEN (21 bits), so use that to enforce the limit. | |
1021 | */ | |
1022 | resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps)); | |
1023 | if (unlikely(rt)) { | |
1024 | resrtextents = qblocks = resblks; | |
1025 | resrtextents /= mp->m_sb.sb_rextsize; | |
1026 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
1027 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
1028 | } else { | |
1029 | resrtextents = 0; | |
1030 | resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks); | |
1031 | quota_flag = XFS_QMOPT_RES_REGBLKS; | |
1032 | } | |
1033 | ||
1034 | /* | |
1035 | * Allocate and setup the transaction. | |
1036 | */ | |
1037 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
3d3c8b52 JL |
1038 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, |
1039 | resblks, resrtextents); | |
c24b5dfa DC |
1040 | /* |
1041 | * Check for running out of space | |
1042 | */ | |
1043 | if (error) { | |
1044 | /* | |
1045 | * Free the transaction structure. | |
1046 | */ | |
2451337d | 1047 | ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp)); |
4906e215 | 1048 | xfs_trans_cancel(tp); |
c24b5dfa DC |
1049 | break; |
1050 | } | |
1051 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1052 | error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, | |
1053 | 0, quota_flag); | |
1054 | if (error) | |
1055 | goto error1; | |
1056 | ||
1057 | xfs_trans_ijoin(tp, ip, 0); | |
1058 | ||
1059 | xfs_bmap_init(&free_list, &firstfsb); | |
1060 | error = xfs_bmapi_write(tp, ip, startoffset_fsb, | |
1061 | allocatesize_fsb, alloc_type, &firstfsb, | |
dbd5c8c9 | 1062 | resblks, imapp, &nimaps, &free_list); |
f6106efa | 1063 | if (error) |
c24b5dfa | 1064 | goto error0; |
c24b5dfa DC |
1065 | |
1066 | /* | |
1067 | * Complete the transaction | |
1068 | */ | |
f6106efa ES |
1069 | error = xfs_bmap_finish(&tp, &free_list, NULL); |
1070 | if (error) | |
c24b5dfa | 1071 | goto error0; |
c24b5dfa | 1072 | |
70393313 | 1073 | error = xfs_trans_commit(tp); |
c24b5dfa | 1074 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
f6106efa | 1075 | if (error) |
c24b5dfa | 1076 | break; |
c24b5dfa DC |
1077 | |
1078 | allocated_fsb = imapp->br_blockcount; | |
1079 | ||
1080 | if (nimaps == 0) { | |
2451337d | 1081 | error = -ENOSPC; |
c24b5dfa DC |
1082 | break; |
1083 | } | |
1084 | ||
1085 | startoffset_fsb += allocated_fsb; | |
1086 | allocatesize_fsb -= allocated_fsb; | |
1087 | } | |
1088 | ||
1089 | return error; | |
1090 | ||
1091 | error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */ | |
1092 | xfs_bmap_cancel(&free_list); | |
1093 | xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag); | |
1094 | ||
1095 | error1: /* Just cancel transaction */ | |
4906e215 | 1096 | xfs_trans_cancel(tp); |
c24b5dfa DC |
1097 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1098 | return error; | |
1099 | } | |
1100 | ||
1101 | /* | |
1102 | * Zero file bytes between startoff and endoff inclusive. | |
1103 | * The iolock is held exclusive and no blocks are buffered. | |
1104 | * | |
1105 | * This function is used by xfs_free_file_space() to zero | |
1106 | * partial blocks when the range to free is not block aligned. | |
1107 | * When unreserving space with boundaries that are not block | |
1108 | * aligned we round up the start and round down the end | |
1109 | * boundaries and then use this function to zero the parts of | |
1110 | * the blocks that got dropped during the rounding. | |
1111 | */ | |
1112 | STATIC int | |
1113 | xfs_zero_remaining_bytes( | |
1114 | xfs_inode_t *ip, | |
1115 | xfs_off_t startoff, | |
1116 | xfs_off_t endoff) | |
1117 | { | |
1118 | xfs_bmbt_irec_t imap; | |
1119 | xfs_fileoff_t offset_fsb; | |
1120 | xfs_off_t lastoffset; | |
1121 | xfs_off_t offset; | |
1122 | xfs_buf_t *bp; | |
1123 | xfs_mount_t *mp = ip->i_mount; | |
1124 | int nimap; | |
1125 | int error = 0; | |
1126 | ||
1127 | /* | |
1128 | * Avoid doing I/O beyond eof - it's not necessary | |
1129 | * since nothing can read beyond eof. The space will | |
1130 | * be zeroed when the file is extended anyway. | |
1131 | */ | |
1132 | if (startoff >= XFS_ISIZE(ip)) | |
1133 | return 0; | |
1134 | ||
1135 | if (endoff > XFS_ISIZE(ip)) | |
1136 | endoff = XFS_ISIZE(ip); | |
1137 | ||
c24b5dfa | 1138 | for (offset = startoff; offset <= endoff; offset = lastoffset + 1) { |
4f317369 CH |
1139 | uint lock_mode; |
1140 | ||
c24b5dfa DC |
1141 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
1142 | nimap = 1; | |
4f317369 CH |
1143 | |
1144 | lock_mode = xfs_ilock_data_map_shared(ip); | |
c24b5dfa | 1145 | error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0); |
4f317369 CH |
1146 | xfs_iunlock(ip, lock_mode); |
1147 | ||
c24b5dfa DC |
1148 | if (error || nimap < 1) |
1149 | break; | |
1150 | ASSERT(imap.br_blockcount >= 1); | |
1151 | ASSERT(imap.br_startoff == offset_fsb); | |
4f69f578 DC |
1152 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); |
1153 | ||
1154 | if (imap.br_startblock == HOLESTARTBLOCK || | |
1155 | imap.br_state == XFS_EXT_UNWRITTEN) { | |
1156 | /* skip the entire extent */ | |
1157 | lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + | |
1158 | imap.br_blockcount) - 1; | |
1159 | continue; | |
1160 | } | |
1161 | ||
c24b5dfa DC |
1162 | lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1; |
1163 | if (lastoffset > endoff) | |
1164 | lastoffset = endoff; | |
4f69f578 DC |
1165 | |
1166 | /* DAX can just zero the backing device directly */ | |
1167 | if (IS_DAX(VFS_I(ip))) { | |
1168 | error = dax_zero_page_range(VFS_I(ip), offset, | |
1169 | lastoffset - offset + 1, | |
1170 | xfs_get_blocks_direct); | |
1171 | if (error) | |
1172 | return error; | |
c24b5dfa | 1173 | continue; |
4f69f578 | 1174 | } |
83a0adc3 | 1175 | |
8c156125 CH |
1176 | error = xfs_buf_read_uncached(XFS_IS_REALTIME_INODE(ip) ? |
1177 | mp->m_rtdev_targp : mp->m_ddev_targp, | |
1178 | xfs_fsb_to_db(ip, imap.br_startblock), | |
1179 | BTOBB(mp->m_sb.sb_blocksize), | |
1180 | 0, &bp, NULL); | |
1181 | if (error) | |
1182 | return error; | |
1183 | ||
c24b5dfa | 1184 | memset(bp->b_addr + |
8c156125 CH |
1185 | (offset - XFS_FSB_TO_B(mp, imap.br_startoff)), |
1186 | 0, lastoffset - offset + 1); | |
83a0adc3 | 1187 | |
8c156125 CH |
1188 | error = xfs_bwrite(bp); |
1189 | xfs_buf_relse(bp); | |
1190 | if (error) | |
1191 | return error; | |
c24b5dfa | 1192 | } |
c24b5dfa DC |
1193 | return error; |
1194 | } | |
1195 | ||
83aee9e4 | 1196 | int |
c24b5dfa | 1197 | xfs_free_file_space( |
83aee9e4 | 1198 | struct xfs_inode *ip, |
c24b5dfa | 1199 | xfs_off_t offset, |
5f8aca8b | 1200 | xfs_off_t len) |
c24b5dfa | 1201 | { |
c24b5dfa DC |
1202 | int done; |
1203 | xfs_fileoff_t endoffset_fsb; | |
1204 | int error; | |
1205 | xfs_fsblock_t firstfsb; | |
1206 | xfs_bmap_free_t free_list; | |
1207 | xfs_bmbt_irec_t imap; | |
1208 | xfs_off_t ioffset; | |
8b5279e3 | 1209 | xfs_off_t iendoffset; |
c24b5dfa DC |
1210 | xfs_extlen_t mod=0; |
1211 | xfs_mount_t *mp; | |
1212 | int nimap; | |
1213 | uint resblks; | |
1214 | xfs_off_t rounding; | |
1215 | int rt; | |
1216 | xfs_fileoff_t startoffset_fsb; | |
1217 | xfs_trans_t *tp; | |
c24b5dfa DC |
1218 | |
1219 | mp = ip->i_mount; | |
1220 | ||
1221 | trace_xfs_free_file_space(ip); | |
1222 | ||
1223 | error = xfs_qm_dqattach(ip, 0); | |
1224 | if (error) | |
1225 | return error; | |
1226 | ||
1227 | error = 0; | |
1228 | if (len <= 0) /* if nothing being freed */ | |
1229 | return error; | |
1230 | rt = XFS_IS_REALTIME_INODE(ip); | |
1231 | startoffset_fsb = XFS_B_TO_FSB(mp, offset); | |
1232 | endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len); | |
1233 | ||
5f8aca8b CH |
1234 | /* wait for the completion of any pending DIOs */ |
1235 | inode_dio_wait(VFS_I(ip)); | |
c24b5dfa DC |
1236 | |
1237 | rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE); | |
8b5279e3 BF |
1238 | ioffset = round_down(offset, rounding); |
1239 | iendoffset = round_up(offset + len, rounding) - 1; | |
1240 | error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, ioffset, | |
1241 | iendoffset); | |
c24b5dfa | 1242 | if (error) |
5f8aca8b | 1243 | goto out; |
8b5279e3 | 1244 | truncate_pagecache_range(VFS_I(ip), ioffset, iendoffset); |
c24b5dfa DC |
1245 | |
1246 | /* | |
1247 | * Need to zero the stuff we're not freeing, on disk. | |
1248 | * If it's a realtime file & can't use unwritten extents then we | |
1249 | * actually need to zero the extent edges. Otherwise xfs_bunmapi | |
1250 | * will take care of it for us. | |
1251 | */ | |
1252 | if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) { | |
1253 | nimap = 1; | |
1254 | error = xfs_bmapi_read(ip, startoffset_fsb, 1, | |
1255 | &imap, &nimap, 0); | |
1256 | if (error) | |
5f8aca8b | 1257 | goto out; |
c24b5dfa DC |
1258 | ASSERT(nimap == 0 || nimap == 1); |
1259 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
1260 | xfs_daddr_t block; | |
1261 | ||
1262 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
1263 | block = imap.br_startblock; | |
1264 | mod = do_div(block, mp->m_sb.sb_rextsize); | |
1265 | if (mod) | |
1266 | startoffset_fsb += mp->m_sb.sb_rextsize - mod; | |
1267 | } | |
1268 | nimap = 1; | |
1269 | error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1, | |
1270 | &imap, &nimap, 0); | |
1271 | if (error) | |
5f8aca8b | 1272 | goto out; |
c24b5dfa DC |
1273 | ASSERT(nimap == 0 || nimap == 1); |
1274 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
1275 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
1276 | mod++; | |
1277 | if (mod && (mod != mp->m_sb.sb_rextsize)) | |
1278 | endoffset_fsb -= mod; | |
1279 | } | |
1280 | } | |
1281 | if ((done = (endoffset_fsb <= startoffset_fsb))) | |
1282 | /* | |
1283 | * One contiguous piece to clear | |
1284 | */ | |
1285 | error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1); | |
1286 | else { | |
1287 | /* | |
1288 | * Some full blocks, possibly two pieces to clear | |
1289 | */ | |
1290 | if (offset < XFS_FSB_TO_B(mp, startoffset_fsb)) | |
1291 | error = xfs_zero_remaining_bytes(ip, offset, | |
1292 | XFS_FSB_TO_B(mp, startoffset_fsb) - 1); | |
1293 | if (!error && | |
1294 | XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len) | |
1295 | error = xfs_zero_remaining_bytes(ip, | |
1296 | XFS_FSB_TO_B(mp, endoffset_fsb), | |
1297 | offset + len - 1); | |
1298 | } | |
1299 | ||
1300 | /* | |
1301 | * free file space until done or until there is an error | |
1302 | */ | |
1303 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
1304 | while (!error && !done) { | |
1305 | ||
1306 | /* | |
1307 | * allocate and setup the transaction. Allow this | |
1308 | * transaction to dip into the reserve blocks to ensure | |
1309 | * the freeing of the space succeeds at ENOSPC. | |
1310 | */ | |
1311 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
3d3c8b52 | 1312 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, resblks, 0); |
c24b5dfa DC |
1313 | |
1314 | /* | |
1315 | * check for running out of space | |
1316 | */ | |
1317 | if (error) { | |
1318 | /* | |
1319 | * Free the transaction structure. | |
1320 | */ | |
2451337d | 1321 | ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp)); |
4906e215 | 1322 | xfs_trans_cancel(tp); |
c24b5dfa DC |
1323 | break; |
1324 | } | |
1325 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1326 | error = xfs_trans_reserve_quota(tp, mp, | |
1327 | ip->i_udquot, ip->i_gdquot, ip->i_pdquot, | |
1328 | resblks, 0, XFS_QMOPT_RES_REGBLKS); | |
1329 | if (error) | |
1330 | goto error1; | |
1331 | ||
1332 | xfs_trans_ijoin(tp, ip, 0); | |
1333 | ||
1334 | /* | |
1335 | * issue the bunmapi() call to free the blocks | |
1336 | */ | |
1337 | xfs_bmap_init(&free_list, &firstfsb); | |
1338 | error = xfs_bunmapi(tp, ip, startoffset_fsb, | |
1339 | endoffset_fsb - startoffset_fsb, | |
1340 | 0, 2, &firstfsb, &free_list, &done); | |
f6106efa | 1341 | if (error) |
c24b5dfa | 1342 | goto error0; |
c24b5dfa DC |
1343 | |
1344 | /* | |
1345 | * complete the transaction | |
1346 | */ | |
f6106efa ES |
1347 | error = xfs_bmap_finish(&tp, &free_list, NULL); |
1348 | if (error) | |
c24b5dfa | 1349 | goto error0; |
c24b5dfa | 1350 | |
70393313 | 1351 | error = xfs_trans_commit(tp); |
c24b5dfa DC |
1352 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1353 | } | |
1354 | ||
5f8aca8b | 1355 | out: |
c24b5dfa DC |
1356 | return error; |
1357 | ||
1358 | error0: | |
1359 | xfs_bmap_cancel(&free_list); | |
1360 | error1: | |
4906e215 | 1361 | xfs_trans_cancel(tp); |
5f8aca8b CH |
1362 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1363 | goto out; | |
c24b5dfa DC |
1364 | } |
1365 | ||
5d11fb4b BF |
1366 | /* |
1367 | * Preallocate and zero a range of a file. This mechanism has the allocation | |
1368 | * semantics of fallocate and in addition converts data in the range to zeroes. | |
1369 | */ | |
865e9446 | 1370 | int |
c24b5dfa DC |
1371 | xfs_zero_file_space( |
1372 | struct xfs_inode *ip, | |
1373 | xfs_off_t offset, | |
5f8aca8b | 1374 | xfs_off_t len) |
c24b5dfa DC |
1375 | { |
1376 | struct xfs_mount *mp = ip->i_mount; | |
5d11fb4b | 1377 | uint blksize; |
c24b5dfa DC |
1378 | int error; |
1379 | ||
897b73b6 DC |
1380 | trace_xfs_zero_file_space(ip); |
1381 | ||
5d11fb4b | 1382 | blksize = 1 << mp->m_sb.sb_blocklog; |
c24b5dfa DC |
1383 | |
1384 | /* | |
5d11fb4b BF |
1385 | * Punch a hole and prealloc the range. We use hole punch rather than |
1386 | * unwritten extent conversion for two reasons: | |
1387 | * | |
1388 | * 1.) Hole punch handles partial block zeroing for us. | |
1389 | * | |
1390 | * 2.) If prealloc returns ENOSPC, the file range is still zero-valued | |
1391 | * by virtue of the hole punch. | |
c24b5dfa | 1392 | */ |
5d11fb4b BF |
1393 | error = xfs_free_file_space(ip, offset, len); |
1394 | if (error) | |
1395 | goto out; | |
c24b5dfa | 1396 | |
5d11fb4b BF |
1397 | error = xfs_alloc_file_space(ip, round_down(offset, blksize), |
1398 | round_up(offset + len, blksize) - | |
1399 | round_down(offset, blksize), | |
1400 | XFS_BMAPI_PREALLOC); | |
5f8aca8b | 1401 | out: |
c24b5dfa DC |
1402 | return error; |
1403 | ||
1404 | } | |
1405 | ||
e1d8fb88 | 1406 | /* |
a904b1ca NJ |
1407 | * @next_fsb will keep track of the extent currently undergoing shift. |
1408 | * @stop_fsb will keep track of the extent at which we have to stop. | |
1409 | * If we are shifting left, we will start with block (offset + len) and | |
1410 | * shift each extent till last extent. | |
1411 | * If we are shifting right, we will start with last extent inside file space | |
1412 | * and continue until we reach the block corresponding to offset. | |
e1d8fb88 | 1413 | */ |
72c1a739 | 1414 | static int |
a904b1ca NJ |
1415 | xfs_shift_file_space( |
1416 | struct xfs_inode *ip, | |
1417 | xfs_off_t offset, | |
1418 | xfs_off_t len, | |
1419 | enum shift_direction direction) | |
e1d8fb88 NJ |
1420 | { |
1421 | int done = 0; | |
1422 | struct xfs_mount *mp = ip->i_mount; | |
1423 | struct xfs_trans *tp; | |
1424 | int error; | |
e1d8fb88 NJ |
1425 | struct xfs_bmap_free free_list; |
1426 | xfs_fsblock_t first_block; | |
a904b1ca | 1427 | xfs_fileoff_t stop_fsb; |
2c845f5a | 1428 | xfs_fileoff_t next_fsb; |
e1d8fb88 NJ |
1429 | xfs_fileoff_t shift_fsb; |
1430 | ||
a904b1ca | 1431 | ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT); |
e1d8fb88 | 1432 | |
a904b1ca NJ |
1433 | if (direction == SHIFT_LEFT) { |
1434 | next_fsb = XFS_B_TO_FSB(mp, offset + len); | |
1435 | stop_fsb = XFS_B_TO_FSB(mp, VFS_I(ip)->i_size); | |
1436 | } else { | |
1437 | /* | |
1438 | * If right shift, delegate the work of initialization of | |
1439 | * next_fsb to xfs_bmap_shift_extent as it has ilock held. | |
1440 | */ | |
1441 | next_fsb = NULLFSBLOCK; | |
1442 | stop_fsb = XFS_B_TO_FSB(mp, offset); | |
1443 | } | |
e1d8fb88 | 1444 | |
e1d8fb88 NJ |
1445 | shift_fsb = XFS_B_TO_FSB(mp, len); |
1446 | ||
f71721d0 BF |
1447 | /* |
1448 | * Trim eofblocks to avoid shifting uninitialized post-eof preallocation | |
1449 | * into the accessible region of the file. | |
1450 | */ | |
41b9d726 BF |
1451 | if (xfs_can_free_eofblocks(ip, true)) { |
1452 | error = xfs_free_eofblocks(mp, ip, false); | |
1453 | if (error) | |
1454 | return error; | |
1455 | } | |
1669a8ca | 1456 | |
f71721d0 BF |
1457 | /* |
1458 | * Writeback and invalidate cache for the remainder of the file as we're | |
a904b1ca | 1459 | * about to shift down every extent from offset to EOF. |
f71721d0 BF |
1460 | */ |
1461 | error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, | |
a904b1ca | 1462 | offset, -1); |
f71721d0 BF |
1463 | if (error) |
1464 | return error; | |
1465 | error = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, | |
a904b1ca | 1466 | offset >> PAGE_CACHE_SHIFT, -1); |
e1d8fb88 NJ |
1467 | if (error) |
1468 | return error; | |
1469 | ||
a904b1ca NJ |
1470 | /* |
1471 | * The extent shiting code works on extent granularity. So, if | |
1472 | * stop_fsb is not the starting block of extent, we need to split | |
1473 | * the extent at stop_fsb. | |
1474 | */ | |
1475 | if (direction == SHIFT_RIGHT) { | |
1476 | error = xfs_bmap_split_extent(ip, stop_fsb); | |
1477 | if (error) | |
1478 | return error; | |
1479 | } | |
1480 | ||
e1d8fb88 NJ |
1481 | while (!error && !done) { |
1482 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
e1d8fb88 NJ |
1483 | /* |
1484 | * We would need to reserve permanent block for transaction. | |
1485 | * This will come into picture when after shifting extent into | |
1486 | * hole we found that adjacent extents can be merged which | |
1487 | * may lead to freeing of a block during record update. | |
1488 | */ | |
1489 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, | |
1490 | XFS_DIOSTRAT_SPACE_RES(mp, 0), 0); | |
1491 | if (error) { | |
4906e215 | 1492 | xfs_trans_cancel(tp); |
e1d8fb88 NJ |
1493 | break; |
1494 | } | |
1495 | ||
1496 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1497 | error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot, | |
1498 | ip->i_gdquot, ip->i_pdquot, | |
1499 | XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, | |
1500 | XFS_QMOPT_RES_REGBLKS); | |
1501 | if (error) | |
d4a97a04 | 1502 | goto out_trans_cancel; |
e1d8fb88 | 1503 | |
a904b1ca | 1504 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
e1d8fb88 NJ |
1505 | |
1506 | xfs_bmap_init(&free_list, &first_block); | |
1507 | ||
1508 | /* | |
1509 | * We are using the write transaction in which max 2 bmbt | |
1510 | * updates are allowed | |
1511 | */ | |
a904b1ca NJ |
1512 | error = xfs_bmap_shift_extents(tp, ip, &next_fsb, shift_fsb, |
1513 | &done, stop_fsb, &first_block, &free_list, | |
1514 | direction, XFS_BMAP_MAX_SHIFT_EXTENTS); | |
e1d8fb88 | 1515 | if (error) |
d4a97a04 | 1516 | goto out_bmap_cancel; |
e1d8fb88 | 1517 | |
f6106efa | 1518 | error = xfs_bmap_finish(&tp, &free_list, NULL); |
e1d8fb88 | 1519 | if (error) |
d4a97a04 | 1520 | goto out_bmap_cancel; |
e1d8fb88 | 1521 | |
70393313 | 1522 | error = xfs_trans_commit(tp); |
e1d8fb88 NJ |
1523 | } |
1524 | ||
1525 | return error; | |
1526 | ||
d4a97a04 BF |
1527 | out_bmap_cancel: |
1528 | xfs_bmap_cancel(&free_list); | |
1529 | out_trans_cancel: | |
4906e215 | 1530 | xfs_trans_cancel(tp); |
e1d8fb88 NJ |
1531 | return error; |
1532 | } | |
1533 | ||
a904b1ca NJ |
1534 | /* |
1535 | * xfs_collapse_file_space() | |
1536 | * This routine frees disk space and shift extent for the given file. | |
1537 | * The first thing we do is to free data blocks in the specified range | |
1538 | * by calling xfs_free_file_space(). It would also sync dirty data | |
1539 | * and invalidate page cache over the region on which collapse range | |
1540 | * is working. And Shift extent records to the left to cover a hole. | |
1541 | * RETURNS: | |
1542 | * 0 on success | |
1543 | * errno on error | |
1544 | * | |
1545 | */ | |
1546 | int | |
1547 | xfs_collapse_file_space( | |
1548 | struct xfs_inode *ip, | |
1549 | xfs_off_t offset, | |
1550 | xfs_off_t len) | |
1551 | { | |
1552 | int error; | |
1553 | ||
1554 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
1555 | trace_xfs_collapse_file_space(ip); | |
1556 | ||
1557 | error = xfs_free_file_space(ip, offset, len); | |
1558 | if (error) | |
1559 | return error; | |
1560 | ||
1561 | return xfs_shift_file_space(ip, offset, len, SHIFT_LEFT); | |
1562 | } | |
1563 | ||
1564 | /* | |
1565 | * xfs_insert_file_space() | |
1566 | * This routine create hole space by shifting extents for the given file. | |
1567 | * The first thing we do is to sync dirty data and invalidate page cache | |
1568 | * over the region on which insert range is working. And split an extent | |
1569 | * to two extents at given offset by calling xfs_bmap_split_extent. | |
1570 | * And shift all extent records which are laying between [offset, | |
1571 | * last allocated extent] to the right to reserve hole range. | |
1572 | * RETURNS: | |
1573 | * 0 on success | |
1574 | * errno on error | |
1575 | */ | |
1576 | int | |
1577 | xfs_insert_file_space( | |
1578 | struct xfs_inode *ip, | |
1579 | loff_t offset, | |
1580 | loff_t len) | |
1581 | { | |
1582 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
1583 | trace_xfs_insert_file_space(ip); | |
1584 | ||
1585 | return xfs_shift_file_space(ip, offset, len, SHIFT_RIGHT); | |
1586 | } | |
1587 | ||
a133d952 DC |
1588 | /* |
1589 | * We need to check that the format of the data fork in the temporary inode is | |
1590 | * valid for the target inode before doing the swap. This is not a problem with | |
1591 | * attr1 because of the fixed fork offset, but attr2 has a dynamically sized | |
1592 | * data fork depending on the space the attribute fork is taking so we can get | |
1593 | * invalid formats on the target inode. | |
1594 | * | |
1595 | * E.g. target has space for 7 extents in extent format, temp inode only has | |
1596 | * space for 6. If we defragment down to 7 extents, then the tmp format is a | |
1597 | * btree, but when swapped it needs to be in extent format. Hence we can't just | |
1598 | * blindly swap data forks on attr2 filesystems. | |
1599 | * | |
1600 | * Note that we check the swap in both directions so that we don't end up with | |
1601 | * a corrupt temporary inode, either. | |
1602 | * | |
1603 | * Note that fixing the way xfs_fsr sets up the attribute fork in the source | |
1604 | * inode will prevent this situation from occurring, so all we do here is | |
1605 | * reject and log the attempt. basically we are putting the responsibility on | |
1606 | * userspace to get this right. | |
1607 | */ | |
1608 | static int | |
1609 | xfs_swap_extents_check_format( | |
1610 | xfs_inode_t *ip, /* target inode */ | |
1611 | xfs_inode_t *tip) /* tmp inode */ | |
1612 | { | |
1613 | ||
1614 | /* Should never get a local format */ | |
1615 | if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL || | |
1616 | tip->i_d.di_format == XFS_DINODE_FMT_LOCAL) | |
2451337d | 1617 | return -EINVAL; |
a133d952 DC |
1618 | |
1619 | /* | |
1620 | * if the target inode has less extents that then temporary inode then | |
1621 | * why did userspace call us? | |
1622 | */ | |
1623 | if (ip->i_d.di_nextents < tip->i_d.di_nextents) | |
2451337d | 1624 | return -EINVAL; |
a133d952 DC |
1625 | |
1626 | /* | |
1627 | * if the target inode is in extent form and the temp inode is in btree | |
1628 | * form then we will end up with the target inode in the wrong format | |
1629 | * as we already know there are less extents in the temp inode. | |
1630 | */ | |
1631 | if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS && | |
1632 | tip->i_d.di_format == XFS_DINODE_FMT_BTREE) | |
2451337d | 1633 | return -EINVAL; |
a133d952 DC |
1634 | |
1635 | /* Check temp in extent form to max in target */ | |
1636 | if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS && | |
1637 | XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) > | |
1638 | XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)) | |
2451337d | 1639 | return -EINVAL; |
a133d952 DC |
1640 | |
1641 | /* Check target in extent form to max in temp */ | |
1642 | if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS && | |
1643 | XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) > | |
1644 | XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK)) | |
2451337d | 1645 | return -EINVAL; |
a133d952 DC |
1646 | |
1647 | /* | |
1648 | * If we are in a btree format, check that the temp root block will fit | |
1649 | * in the target and that it has enough extents to be in btree format | |
1650 | * in the target. | |
1651 | * | |
1652 | * Note that we have to be careful to allow btree->extent conversions | |
1653 | * (a common defrag case) which will occur when the temp inode is in | |
1654 | * extent format... | |
1655 | */ | |
1656 | if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) { | |
1657 | if (XFS_IFORK_BOFF(ip) && | |
1658 | XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip)) | |
2451337d | 1659 | return -EINVAL; |
a133d952 DC |
1660 | if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <= |
1661 | XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)) | |
2451337d | 1662 | return -EINVAL; |
a133d952 DC |
1663 | } |
1664 | ||
1665 | /* Reciprocal target->temp btree format checks */ | |
1666 | if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) { | |
1667 | if (XFS_IFORK_BOFF(tip) && | |
1668 | XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip)) | |
2451337d | 1669 | return -EINVAL; |
a133d952 DC |
1670 | if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <= |
1671 | XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK)) | |
2451337d | 1672 | return -EINVAL; |
a133d952 DC |
1673 | } |
1674 | ||
1675 | return 0; | |
1676 | } | |
1677 | ||
7abbb8f9 | 1678 | static int |
4ef897a2 DC |
1679 | xfs_swap_extent_flush( |
1680 | struct xfs_inode *ip) | |
1681 | { | |
1682 | int error; | |
1683 | ||
1684 | error = filemap_write_and_wait(VFS_I(ip)->i_mapping); | |
1685 | if (error) | |
1686 | return error; | |
1687 | truncate_pagecache_range(VFS_I(ip), 0, -1); | |
1688 | ||
1689 | /* Verify O_DIRECT for ftmp */ | |
1690 | if (VFS_I(ip)->i_mapping->nrpages) | |
1691 | return -EINVAL; | |
4ef897a2 DC |
1692 | return 0; |
1693 | } | |
1694 | ||
a133d952 DC |
1695 | int |
1696 | xfs_swap_extents( | |
1697 | xfs_inode_t *ip, /* target inode */ | |
1698 | xfs_inode_t *tip, /* tmp inode */ | |
1699 | xfs_swapext_t *sxp) | |
1700 | { | |
1701 | xfs_mount_t *mp = ip->i_mount; | |
1702 | xfs_trans_t *tp; | |
1703 | xfs_bstat_t *sbp = &sxp->sx_stat; | |
1704 | xfs_ifork_t *tempifp, *ifp, *tifp; | |
1705 | int src_log_flags, target_log_flags; | |
1706 | int error = 0; | |
1707 | int aforkblks = 0; | |
1708 | int taforkblks = 0; | |
1709 | __uint64_t tmp; | |
81217683 | 1710 | int lock_flags; |
a133d952 | 1711 | |
a133d952 DC |
1712 | tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL); |
1713 | if (!tempifp) { | |
2451337d | 1714 | error = -ENOMEM; |
a133d952 DC |
1715 | goto out; |
1716 | } | |
1717 | ||
1718 | /* | |
723cac48 DC |
1719 | * Lock the inodes against other IO, page faults and truncate to |
1720 | * begin with. Then we can ensure the inodes are flushed and have no | |
1721 | * page cache safely. Once we have done this we can take the ilocks and | |
1722 | * do the rest of the checks. | |
a133d952 | 1723 | */ |
723cac48 | 1724 | lock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; |
a133d952 | 1725 | xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL); |
723cac48 | 1726 | xfs_lock_two_inodes(ip, tip, XFS_MMAPLOCK_EXCL); |
a133d952 DC |
1727 | |
1728 | /* Verify that both files have the same format */ | |
1729 | if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) { | |
2451337d | 1730 | error = -EINVAL; |
a133d952 DC |
1731 | goto out_unlock; |
1732 | } | |
1733 | ||
1734 | /* Verify both files are either real-time or non-realtime */ | |
1735 | if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) { | |
2451337d | 1736 | error = -EINVAL; |
a133d952 DC |
1737 | goto out_unlock; |
1738 | } | |
1739 | ||
4ef897a2 DC |
1740 | error = xfs_swap_extent_flush(ip); |
1741 | if (error) | |
1742 | goto out_unlock; | |
1743 | error = xfs_swap_extent_flush(tip); | |
a133d952 DC |
1744 | if (error) |
1745 | goto out_unlock; | |
a133d952 | 1746 | |
4ef897a2 DC |
1747 | tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT); |
1748 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0); | |
1749 | if (error) { | |
4906e215 | 1750 | xfs_trans_cancel(tp); |
a133d952 DC |
1751 | goto out_unlock; |
1752 | } | |
723cac48 DC |
1753 | |
1754 | /* | |
1755 | * Lock and join the inodes to the tansaction so that transaction commit | |
1756 | * or cancel will unlock the inodes from this point onwards. | |
1757 | */ | |
4ef897a2 DC |
1758 | xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL); |
1759 | lock_flags |= XFS_ILOCK_EXCL; | |
723cac48 DC |
1760 | xfs_trans_ijoin(tp, ip, lock_flags); |
1761 | xfs_trans_ijoin(tp, tip, lock_flags); | |
1762 | ||
a133d952 DC |
1763 | |
1764 | /* Verify all data are being swapped */ | |
1765 | if (sxp->sx_offset != 0 || | |
1766 | sxp->sx_length != ip->i_d.di_size || | |
1767 | sxp->sx_length != tip->i_d.di_size) { | |
2451337d | 1768 | error = -EFAULT; |
4ef897a2 | 1769 | goto out_trans_cancel; |
a133d952 DC |
1770 | } |
1771 | ||
1772 | trace_xfs_swap_extent_before(ip, 0); | |
1773 | trace_xfs_swap_extent_before(tip, 1); | |
1774 | ||
1775 | /* check inode formats now that data is flushed */ | |
1776 | error = xfs_swap_extents_check_format(ip, tip); | |
1777 | if (error) { | |
1778 | xfs_notice(mp, | |
1779 | "%s: inode 0x%llx format is incompatible for exchanging.", | |
1780 | __func__, ip->i_ino); | |
4ef897a2 | 1781 | goto out_trans_cancel; |
a133d952 DC |
1782 | } |
1783 | ||
1784 | /* | |
1785 | * Compare the current change & modify times with that | |
1786 | * passed in. If they differ, we abort this swap. | |
1787 | * This is the mechanism used to ensure the calling | |
1788 | * process that the file was not changed out from | |
1789 | * under it. | |
1790 | */ | |
1791 | if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) || | |
1792 | (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) || | |
1793 | (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) || | |
1794 | (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) { | |
2451337d | 1795 | error = -EBUSY; |
81217683 | 1796 | goto out_trans_cancel; |
a133d952 | 1797 | } |
a133d952 DC |
1798 | /* |
1799 | * Count the number of extended attribute blocks | |
1800 | */ | |
1801 | if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) && | |
1802 | (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) { | |
1803 | error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks); | |
1804 | if (error) | |
1805 | goto out_trans_cancel; | |
1806 | } | |
1807 | if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) && | |
1808 | (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) { | |
1809 | error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, | |
1810 | &taforkblks); | |
1811 | if (error) | |
1812 | goto out_trans_cancel; | |
1813 | } | |
1814 | ||
21b5c978 DC |
1815 | /* |
1816 | * Before we've swapped the forks, lets set the owners of the forks | |
1817 | * appropriately. We have to do this as we are demand paging the btree | |
1818 | * buffers, and so the validation done on read will expect the owner | |
1819 | * field to be correctly set. Once we change the owners, we can swap the | |
1820 | * inode forks. | |
1821 | * | |
1822 | * Note the trickiness in setting the log flags - we set the owner log | |
1823 | * flag on the opposite inode (i.e. the inode we are setting the new | |
1824 | * owner to be) because once we swap the forks and log that, log | |
1825 | * recovery is going to see the fork as owned by the swapped inode, | |
1826 | * not the pre-swapped inodes. | |
1827 | */ | |
1828 | src_log_flags = XFS_ILOG_CORE; | |
1829 | target_log_flags = XFS_ILOG_CORE; | |
1830 | if (ip->i_d.di_version == 3 && | |
1831 | ip->i_d.di_format == XFS_DINODE_FMT_BTREE) { | |
638f4416 DC |
1832 | target_log_flags |= XFS_ILOG_DOWNER; |
1833 | error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, | |
1834 | tip->i_ino, NULL); | |
21b5c978 DC |
1835 | if (error) |
1836 | goto out_trans_cancel; | |
1837 | } | |
1838 | ||
1839 | if (tip->i_d.di_version == 3 && | |
1840 | tip->i_d.di_format == XFS_DINODE_FMT_BTREE) { | |
638f4416 DC |
1841 | src_log_flags |= XFS_ILOG_DOWNER; |
1842 | error = xfs_bmbt_change_owner(tp, tip, XFS_DATA_FORK, | |
1843 | ip->i_ino, NULL); | |
21b5c978 DC |
1844 | if (error) |
1845 | goto out_trans_cancel; | |
1846 | } | |
1847 | ||
a133d952 DC |
1848 | /* |
1849 | * Swap the data forks of the inodes | |
1850 | */ | |
1851 | ifp = &ip->i_df; | |
1852 | tifp = &tip->i_df; | |
1853 | *tempifp = *ifp; /* struct copy */ | |
1854 | *ifp = *tifp; /* struct copy */ | |
1855 | *tifp = *tempifp; /* struct copy */ | |
1856 | ||
1857 | /* | |
1858 | * Fix the on-disk inode values | |
1859 | */ | |
1860 | tmp = (__uint64_t)ip->i_d.di_nblocks; | |
1861 | ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks; | |
1862 | tip->i_d.di_nblocks = tmp + taforkblks - aforkblks; | |
1863 | ||
1864 | tmp = (__uint64_t) ip->i_d.di_nextents; | |
1865 | ip->i_d.di_nextents = tip->i_d.di_nextents; | |
1866 | tip->i_d.di_nextents = tmp; | |
1867 | ||
1868 | tmp = (__uint64_t) ip->i_d.di_format; | |
1869 | ip->i_d.di_format = tip->i_d.di_format; | |
1870 | tip->i_d.di_format = tmp; | |
1871 | ||
1872 | /* | |
1873 | * The extents in the source inode could still contain speculative | |
1874 | * preallocation beyond EOF (e.g. the file is open but not modified | |
1875 | * while defrag is in progress). In that case, we need to copy over the | |
1876 | * number of delalloc blocks the data fork in the source inode is | |
1877 | * tracking beyond EOF so that when the fork is truncated away when the | |
1878 | * temporary inode is unlinked we don't underrun the i_delayed_blks | |
1879 | * counter on that inode. | |
1880 | */ | |
1881 | ASSERT(tip->i_delayed_blks == 0); | |
1882 | tip->i_delayed_blks = ip->i_delayed_blks; | |
1883 | ip->i_delayed_blks = 0; | |
1884 | ||
a133d952 DC |
1885 | switch (ip->i_d.di_format) { |
1886 | case XFS_DINODE_FMT_EXTENTS: | |
1887 | /* If the extents fit in the inode, fix the | |
1888 | * pointer. Otherwise it's already NULL or | |
1889 | * pointing to the extent. | |
1890 | */ | |
1891 | if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) { | |
1892 | ifp->if_u1.if_extents = | |
1893 | ifp->if_u2.if_inline_ext; | |
1894 | } | |
1895 | src_log_flags |= XFS_ILOG_DEXT; | |
1896 | break; | |
1897 | case XFS_DINODE_FMT_BTREE: | |
21b5c978 | 1898 | ASSERT(ip->i_d.di_version < 3 || |
638f4416 | 1899 | (src_log_flags & XFS_ILOG_DOWNER)); |
a133d952 DC |
1900 | src_log_flags |= XFS_ILOG_DBROOT; |
1901 | break; | |
1902 | } | |
1903 | ||
a133d952 DC |
1904 | switch (tip->i_d.di_format) { |
1905 | case XFS_DINODE_FMT_EXTENTS: | |
1906 | /* If the extents fit in the inode, fix the | |
1907 | * pointer. Otherwise it's already NULL or | |
1908 | * pointing to the extent. | |
1909 | */ | |
1910 | if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) { | |
1911 | tifp->if_u1.if_extents = | |
1912 | tifp->if_u2.if_inline_ext; | |
1913 | } | |
1914 | target_log_flags |= XFS_ILOG_DEXT; | |
1915 | break; | |
1916 | case XFS_DINODE_FMT_BTREE: | |
1917 | target_log_flags |= XFS_ILOG_DBROOT; | |
21b5c978 | 1918 | ASSERT(tip->i_d.di_version < 3 || |
638f4416 | 1919 | (target_log_flags & XFS_ILOG_DOWNER)); |
a133d952 DC |
1920 | break; |
1921 | } | |
1922 | ||
a133d952 DC |
1923 | xfs_trans_log_inode(tp, ip, src_log_flags); |
1924 | xfs_trans_log_inode(tp, tip, target_log_flags); | |
1925 | ||
1926 | /* | |
1927 | * If this is a synchronous mount, make sure that the | |
1928 | * transaction goes to disk before returning to the user. | |
1929 | */ | |
1930 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
1931 | xfs_trans_set_sync(tp); | |
1932 | ||
70393313 | 1933 | error = xfs_trans_commit(tp); |
a133d952 DC |
1934 | |
1935 | trace_xfs_swap_extent_after(ip, 0); | |
1936 | trace_xfs_swap_extent_after(tip, 1); | |
1937 | out: | |
1938 | kmem_free(tempifp); | |
1939 | return error; | |
1940 | ||
1941 | out_unlock: | |
81217683 DC |
1942 | xfs_iunlock(ip, lock_flags); |
1943 | xfs_iunlock(tip, lock_flags); | |
a133d952 DC |
1944 | goto out; |
1945 | ||
1946 | out_trans_cancel: | |
4906e215 | 1947 | xfs_trans_cancel(tp); |
723cac48 | 1948 | goto out; |
a133d952 | 1949 | } |