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 DC |
26 | #include "xfs_sb.h" |
27 | #include "xfs_ag.h" | |
28 | #include "xfs_mount.h" | |
57062787 | 29 | #include "xfs_da_format.h" |
68988114 DC |
30 | #include "xfs_inode.h" |
31 | #include "xfs_btree.h" | |
239880ef | 32 | #include "xfs_trans.h" |
68988114 DC |
33 | #include "xfs_extfree_item.h" |
34 | #include "xfs_alloc.h" | |
35 | #include "xfs_bmap.h" | |
36 | #include "xfs_bmap_util.h" | |
a4fbe6ab | 37 | #include "xfs_bmap_btree.h" |
68988114 DC |
38 | #include "xfs_rtalloc.h" |
39 | #include "xfs_error.h" | |
40 | #include "xfs_quota.h" | |
41 | #include "xfs_trans_space.h" | |
42 | #include "xfs_trace.h" | |
c24b5dfa | 43 | #include "xfs_icache.h" |
239880ef | 44 | #include "xfs_log.h" |
a4fbe6ab | 45 | #include "xfs_dinode.h" |
68988114 DC |
46 | |
47 | /* Kernel only BMAP related definitions and functions */ | |
48 | ||
49 | /* | |
50 | * Convert the given file system block to a disk block. We have to treat it | |
51 | * differently based on whether the file is a real time file or not, because the | |
52 | * bmap code does. | |
53 | */ | |
54 | xfs_daddr_t | |
55 | xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb) | |
56 | { | |
57 | return (XFS_IS_REALTIME_INODE(ip) ? \ | |
58 | (xfs_daddr_t)XFS_FSB_TO_BB((ip)->i_mount, (fsb)) : \ | |
59 | XFS_FSB_TO_DADDR((ip)->i_mount, (fsb))); | |
60 | } | |
61 | ||
62 | /* | |
63 | * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi | |
64 | * caller. Frees all the extents that need freeing, which must be done | |
65 | * last due to locking considerations. We never free any extents in | |
66 | * the first transaction. | |
67 | * | |
68 | * Return 1 if the given transaction was committed and a new one | |
69 | * started, and 0 otherwise in the committed parameter. | |
70 | */ | |
71 | int /* error */ | |
72 | xfs_bmap_finish( | |
73 | xfs_trans_t **tp, /* transaction pointer addr */ | |
74 | xfs_bmap_free_t *flist, /* i/o: list extents to free */ | |
75 | int *committed) /* xact committed or not */ | |
76 | { | |
77 | xfs_efd_log_item_t *efd; /* extent free data */ | |
78 | xfs_efi_log_item_t *efi; /* extent free intention */ | |
79 | int error; /* error return value */ | |
80 | xfs_bmap_free_item_t *free; /* free extent item */ | |
3d3c8b52 | 81 | struct xfs_trans_res tres; /* new log reservation */ |
68988114 DC |
82 | xfs_mount_t *mp; /* filesystem mount structure */ |
83 | xfs_bmap_free_item_t *next; /* next item on free list */ | |
84 | xfs_trans_t *ntp; /* new transaction pointer */ | |
85 | ||
86 | ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES); | |
87 | if (flist->xbf_count == 0) { | |
88 | *committed = 0; | |
89 | return 0; | |
90 | } | |
91 | ntp = *tp; | |
92 | efi = xfs_trans_get_efi(ntp, flist->xbf_count); | |
93 | for (free = flist->xbf_first; free; free = free->xbfi_next) | |
94 | xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock, | |
95 | free->xbfi_blockcount); | |
3d3c8b52 JL |
96 | |
97 | tres.tr_logres = ntp->t_log_res; | |
98 | tres.tr_logcount = ntp->t_log_count; | |
99 | tres.tr_logflags = XFS_TRANS_PERM_LOG_RES; | |
68988114 DC |
100 | ntp = xfs_trans_dup(*tp); |
101 | error = xfs_trans_commit(*tp, 0); | |
102 | *tp = ntp; | |
103 | *committed = 1; | |
104 | /* | |
105 | * We have a new transaction, so we should return committed=1, | |
106 | * even though we're returning an error. | |
107 | */ | |
108 | if (error) | |
109 | return error; | |
110 | ||
111 | /* | |
112 | * transaction commit worked ok so we can drop the extra ticket | |
113 | * reference that we gained in xfs_trans_dup() | |
114 | */ | |
115 | xfs_log_ticket_put(ntp->t_ticket); | |
116 | ||
3d3c8b52 JL |
117 | error = xfs_trans_reserve(ntp, &tres, 0, 0); |
118 | if (error) | |
68988114 DC |
119 | return error; |
120 | efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count); | |
121 | for (free = flist->xbf_first; free != NULL; free = next) { | |
122 | next = free->xbfi_next; | |
123 | if ((error = xfs_free_extent(ntp, free->xbfi_startblock, | |
124 | free->xbfi_blockcount))) { | |
125 | /* | |
126 | * The bmap free list will be cleaned up at a | |
127 | * higher level. The EFI will be canceled when | |
128 | * this transaction is aborted. | |
129 | * Need to force shutdown here to make sure it | |
130 | * happens, since this transaction may not be | |
131 | * dirty yet. | |
132 | */ | |
133 | mp = ntp->t_mountp; | |
134 | if (!XFS_FORCED_SHUTDOWN(mp)) | |
135 | xfs_force_shutdown(mp, | |
136 | (error == EFSCORRUPTED) ? | |
137 | SHUTDOWN_CORRUPT_INCORE : | |
138 | SHUTDOWN_META_IO_ERROR); | |
139 | return error; | |
140 | } | |
141 | xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock, | |
142 | free->xbfi_blockcount); | |
143 | xfs_bmap_del_free(flist, NULL, free); | |
144 | } | |
145 | return 0; | |
146 | } | |
147 | ||
148 | int | |
149 | xfs_bmap_rtalloc( | |
150 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ | |
151 | { | |
152 | xfs_alloctype_t atype = 0; /* type for allocation routines */ | |
153 | int error; /* error return value */ | |
154 | xfs_mount_t *mp; /* mount point structure */ | |
155 | xfs_extlen_t prod = 0; /* product factor for allocators */ | |
156 | xfs_extlen_t ralen = 0; /* realtime allocation length */ | |
157 | xfs_extlen_t align; /* minimum allocation alignment */ | |
158 | xfs_rtblock_t rtb; | |
159 | ||
160 | mp = ap->ip->i_mount; | |
161 | align = xfs_get_extsz_hint(ap->ip); | |
162 | prod = align / mp->m_sb.sb_rextsize; | |
163 | error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, | |
164 | align, 1, ap->eof, 0, | |
165 | ap->conv, &ap->offset, &ap->length); | |
166 | if (error) | |
167 | return error; | |
168 | ASSERT(ap->length); | |
169 | ASSERT(ap->length % mp->m_sb.sb_rextsize == 0); | |
170 | ||
171 | /* | |
172 | * If the offset & length are not perfectly aligned | |
173 | * then kill prod, it will just get us in trouble. | |
174 | */ | |
175 | if (do_mod(ap->offset, align) || ap->length % align) | |
176 | prod = 1; | |
177 | /* | |
178 | * Set ralen to be the actual requested length in rtextents. | |
179 | */ | |
180 | ralen = ap->length / mp->m_sb.sb_rextsize; | |
181 | /* | |
182 | * If the old value was close enough to MAXEXTLEN that | |
183 | * we rounded up to it, cut it back so it's valid again. | |
184 | * Note that if it's a really large request (bigger than | |
185 | * MAXEXTLEN), we don't hear about that number, and can't | |
186 | * adjust the starting point to match it. | |
187 | */ | |
188 | if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN) | |
189 | ralen = MAXEXTLEN / mp->m_sb.sb_rextsize; | |
190 | ||
191 | /* | |
192 | * Lock out other modifications to the RT bitmap inode. | |
193 | */ | |
194 | xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL); | |
195 | xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL); | |
196 | ||
197 | /* | |
198 | * If it's an allocation to an empty file at offset 0, | |
199 | * pick an extent that will space things out in the rt area. | |
200 | */ | |
201 | if (ap->eof && ap->offset == 0) { | |
202 | xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */ | |
203 | ||
204 | error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx); | |
205 | if (error) | |
206 | return error; | |
207 | ap->blkno = rtx * mp->m_sb.sb_rextsize; | |
208 | } else { | |
209 | ap->blkno = 0; | |
210 | } | |
211 | ||
212 | xfs_bmap_adjacent(ap); | |
213 | ||
214 | /* | |
215 | * Realtime allocation, done through xfs_rtallocate_extent. | |
216 | */ | |
217 | atype = ap->blkno == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO; | |
218 | do_div(ap->blkno, mp->m_sb.sb_rextsize); | |
219 | rtb = ap->blkno; | |
220 | ap->length = ralen; | |
221 | if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length, | |
222 | &ralen, atype, ap->wasdel, prod, &rtb))) | |
223 | return error; | |
224 | if (rtb == NULLFSBLOCK && prod > 1 && | |
225 | (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, | |
226 | ap->length, &ralen, atype, | |
227 | ap->wasdel, 1, &rtb))) | |
228 | return error; | |
229 | ap->blkno = rtb; | |
230 | if (ap->blkno != NULLFSBLOCK) { | |
231 | ap->blkno *= mp->m_sb.sb_rextsize; | |
232 | ralen *= mp->m_sb.sb_rextsize; | |
233 | ap->length = ralen; | |
234 | ap->ip->i_d.di_nblocks += ralen; | |
235 | xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE); | |
236 | if (ap->wasdel) | |
237 | ap->ip->i_delayed_blks -= ralen; | |
238 | /* | |
239 | * Adjust the disk quota also. This was reserved | |
240 | * earlier. | |
241 | */ | |
242 | xfs_trans_mod_dquot_byino(ap->tp, ap->ip, | |
243 | ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT : | |
244 | XFS_TRANS_DQ_RTBCOUNT, (long) ralen); | |
245 | } else { | |
246 | ap->length = 0; | |
247 | } | |
248 | return 0; | |
249 | } | |
250 | ||
251 | /* | |
252 | * Stack switching interfaces for allocation | |
253 | */ | |
254 | static void | |
255 | xfs_bmapi_allocate_worker( | |
256 | struct work_struct *work) | |
257 | { | |
258 | struct xfs_bmalloca *args = container_of(work, | |
259 | struct xfs_bmalloca, work); | |
260 | unsigned long pflags; | |
1f6d6482 | 261 | unsigned long new_pflags = PF_FSTRANS; |
68988114 | 262 | |
1f6d6482 DC |
263 | /* |
264 | * we are in a transaction context here, but may also be doing work | |
265 | * in kswapd context, and hence we may need to inherit that state | |
266 | * temporarily to ensure that we don't block waiting for memory reclaim | |
267 | * in any way. | |
268 | */ | |
269 | if (args->kswapd) | |
270 | new_pflags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD; | |
271 | ||
272 | current_set_flags_nested(&pflags, new_pflags); | |
68988114 DC |
273 | |
274 | args->result = __xfs_bmapi_allocate(args); | |
275 | complete(args->done); | |
276 | ||
1f6d6482 | 277 | current_restore_flags_nested(&pflags, new_pflags); |
68988114 DC |
278 | } |
279 | ||
280 | /* | |
281 | * Some allocation requests often come in with little stack to work on. Push | |
282 | * them off to a worker thread so there is lots of stack to use. Otherwise just | |
283 | * call directly to avoid the context switch overhead here. | |
284 | */ | |
285 | int | |
286 | xfs_bmapi_allocate( | |
287 | struct xfs_bmalloca *args) | |
288 | { | |
289 | DECLARE_COMPLETION_ONSTACK(done); | |
290 | ||
291 | if (!args->stack_switch) | |
292 | return __xfs_bmapi_allocate(args); | |
293 | ||
294 | ||
295 | args->done = &done; | |
1f6d6482 | 296 | args->kswapd = current_is_kswapd(); |
68988114 DC |
297 | INIT_WORK_ONSTACK(&args->work, xfs_bmapi_allocate_worker); |
298 | queue_work(xfs_alloc_wq, &args->work); | |
299 | wait_for_completion(&done); | |
6f96b306 | 300 | destroy_work_on_stack(&args->work); |
68988114 DC |
301 | return args->result; |
302 | } | |
303 | ||
304 | /* | |
305 | * Check if the endoff is outside the last extent. If so the caller will grow | |
306 | * the allocation to a stripe unit boundary. All offsets are considered outside | |
307 | * the end of file for an empty fork, so 1 is returned in *eof in that case. | |
308 | */ | |
309 | int | |
310 | xfs_bmap_eof( | |
311 | struct xfs_inode *ip, | |
312 | xfs_fileoff_t endoff, | |
313 | int whichfork, | |
314 | int *eof) | |
315 | { | |
316 | struct xfs_bmbt_irec rec; | |
317 | int error; | |
318 | ||
319 | error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof); | |
320 | if (error || *eof) | |
321 | return error; | |
322 | ||
323 | *eof = endoff >= rec.br_startoff + rec.br_blockcount; | |
324 | return 0; | |
325 | } | |
326 | ||
327 | /* | |
328 | * Extent tree block counting routines. | |
329 | */ | |
330 | ||
331 | /* | |
332 | * Count leaf blocks given a range of extent records. | |
333 | */ | |
334 | STATIC void | |
335 | xfs_bmap_count_leaves( | |
336 | xfs_ifork_t *ifp, | |
337 | xfs_extnum_t idx, | |
338 | int numrecs, | |
339 | int *count) | |
340 | { | |
341 | int b; | |
342 | ||
343 | for (b = 0; b < numrecs; b++) { | |
344 | xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b); | |
345 | *count += xfs_bmbt_get_blockcount(frp); | |
346 | } | |
347 | } | |
348 | ||
349 | /* | |
350 | * Count leaf blocks given a range of extent records originally | |
351 | * in btree format. | |
352 | */ | |
353 | STATIC void | |
354 | xfs_bmap_disk_count_leaves( | |
355 | struct xfs_mount *mp, | |
356 | struct xfs_btree_block *block, | |
357 | int numrecs, | |
358 | int *count) | |
359 | { | |
360 | int b; | |
361 | xfs_bmbt_rec_t *frp; | |
362 | ||
363 | for (b = 1; b <= numrecs; b++) { | |
364 | frp = XFS_BMBT_REC_ADDR(mp, block, b); | |
365 | *count += xfs_bmbt_disk_get_blockcount(frp); | |
366 | } | |
367 | } | |
368 | ||
369 | /* | |
370 | * Recursively walks each level of a btree | |
8be11e92 | 371 | * to count total fsblocks in use. |
68988114 DC |
372 | */ |
373 | STATIC int /* error */ | |
374 | xfs_bmap_count_tree( | |
375 | xfs_mount_t *mp, /* file system mount point */ | |
376 | xfs_trans_t *tp, /* transaction pointer */ | |
377 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
378 | xfs_fsblock_t blockno, /* file system block number */ | |
379 | int levelin, /* level in btree */ | |
380 | int *count) /* Count of blocks */ | |
381 | { | |
382 | int error; | |
383 | xfs_buf_t *bp, *nbp; | |
384 | int level = levelin; | |
385 | __be64 *pp; | |
386 | xfs_fsblock_t bno = blockno; | |
387 | xfs_fsblock_t nextbno; | |
388 | struct xfs_btree_block *block, *nextblock; | |
389 | int numrecs; | |
390 | ||
391 | error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF, | |
392 | &xfs_bmbt_buf_ops); | |
393 | if (error) | |
394 | return error; | |
395 | *count += 1; | |
396 | block = XFS_BUF_TO_BLOCK(bp); | |
397 | ||
398 | if (--level) { | |
399 | /* Not at node above leaves, count this level of nodes */ | |
400 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); | |
401 | while (nextbno != NULLFSBLOCK) { | |
402 | error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp, | |
403 | XFS_BMAP_BTREE_REF, | |
404 | &xfs_bmbt_buf_ops); | |
405 | if (error) | |
406 | return error; | |
407 | *count += 1; | |
408 | nextblock = XFS_BUF_TO_BLOCK(nbp); | |
409 | nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib); | |
410 | xfs_trans_brelse(tp, nbp); | |
411 | } | |
412 | ||
413 | /* Dive to the next level */ | |
414 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]); | |
415 | bno = be64_to_cpu(*pp); | |
416 | if (unlikely((error = | |
417 | xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) { | |
418 | xfs_trans_brelse(tp, bp); | |
419 | XFS_ERROR_REPORT("xfs_bmap_count_tree(1)", | |
420 | XFS_ERRLEVEL_LOW, mp); | |
b474c7ae | 421 | return EFSCORRUPTED; |
68988114 DC |
422 | } |
423 | xfs_trans_brelse(tp, bp); | |
424 | } else { | |
425 | /* count all level 1 nodes and their leaves */ | |
426 | for (;;) { | |
427 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); | |
428 | numrecs = be16_to_cpu(block->bb_numrecs); | |
429 | xfs_bmap_disk_count_leaves(mp, block, numrecs, count); | |
430 | xfs_trans_brelse(tp, bp); | |
431 | if (nextbno == NULLFSBLOCK) | |
432 | break; | |
433 | bno = nextbno; | |
434 | error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, | |
435 | XFS_BMAP_BTREE_REF, | |
436 | &xfs_bmbt_buf_ops); | |
437 | if (error) | |
438 | return error; | |
439 | *count += 1; | |
440 | block = XFS_BUF_TO_BLOCK(bp); | |
441 | } | |
442 | } | |
443 | return 0; | |
444 | } | |
445 | ||
446 | /* | |
447 | * Count fsblocks of the given fork. | |
448 | */ | |
449 | int /* error */ | |
450 | xfs_bmap_count_blocks( | |
451 | xfs_trans_t *tp, /* transaction pointer */ | |
452 | xfs_inode_t *ip, /* incore inode */ | |
453 | int whichfork, /* data or attr fork */ | |
454 | int *count) /* out: count of blocks */ | |
455 | { | |
456 | struct xfs_btree_block *block; /* current btree block */ | |
457 | xfs_fsblock_t bno; /* block # of "block" */ | |
458 | xfs_ifork_t *ifp; /* fork structure */ | |
459 | int level; /* btree level, for checking */ | |
460 | xfs_mount_t *mp; /* file system mount structure */ | |
461 | __be64 *pp; /* pointer to block address */ | |
462 | ||
463 | bno = NULLFSBLOCK; | |
464 | mp = ip->i_mount; | |
465 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
466 | if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) { | |
467 | xfs_bmap_count_leaves(ifp, 0, | |
468 | ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t), | |
469 | count); | |
470 | return 0; | |
471 | } | |
472 | ||
473 | /* | |
474 | * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out. | |
475 | */ | |
476 | block = ifp->if_broot; | |
477 | level = be16_to_cpu(block->bb_level); | |
478 | ASSERT(level > 0); | |
479 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes); | |
480 | bno = be64_to_cpu(*pp); | |
481 | ASSERT(bno != NULLDFSBNO); | |
482 | ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount); | |
483 | ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks); | |
484 | ||
485 | if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) { | |
486 | XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW, | |
487 | mp); | |
b474c7ae | 488 | return EFSCORRUPTED; |
68988114 DC |
489 | } |
490 | ||
491 | return 0; | |
492 | } | |
493 | ||
494 | /* | |
495 | * returns 1 for success, 0 if we failed to map the extent. | |
496 | */ | |
497 | STATIC int | |
498 | xfs_getbmapx_fix_eof_hole( | |
499 | xfs_inode_t *ip, /* xfs incore inode pointer */ | |
500 | struct getbmapx *out, /* output structure */ | |
501 | int prealloced, /* this is a file with | |
502 | * preallocated data space */ | |
503 | __int64_t end, /* last block requested */ | |
504 | xfs_fsblock_t startblock) | |
505 | { | |
506 | __int64_t fixlen; | |
507 | xfs_mount_t *mp; /* file system mount point */ | |
508 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
509 | xfs_extnum_t lastx; /* last extent pointer */ | |
510 | xfs_fileoff_t fileblock; | |
511 | ||
512 | if (startblock == HOLESTARTBLOCK) { | |
513 | mp = ip->i_mount; | |
514 | out->bmv_block = -1; | |
515 | fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip))); | |
516 | fixlen -= out->bmv_offset; | |
517 | if (prealloced && out->bmv_offset + out->bmv_length == end) { | |
518 | /* Came to hole at EOF. Trim it. */ | |
519 | if (fixlen <= 0) | |
520 | return 0; | |
521 | out->bmv_length = fixlen; | |
522 | } | |
523 | } else { | |
524 | if (startblock == DELAYSTARTBLOCK) | |
525 | out->bmv_block = -2; | |
526 | else | |
527 | out->bmv_block = xfs_fsb_to_db(ip, startblock); | |
528 | fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset); | |
529 | ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK); | |
530 | if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) && | |
531 | (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1)) | |
532 | out->bmv_oflags |= BMV_OF_LAST; | |
533 | } | |
534 | ||
535 | return 1; | |
536 | } | |
537 | ||
538 | /* | |
539 | * Get inode's extents as described in bmv, and format for output. | |
540 | * Calls formatter to fill the user's buffer until all extents | |
541 | * are mapped, until the passed-in bmv->bmv_count slots have | |
542 | * been filled, or until the formatter short-circuits the loop, | |
543 | * if it is tracking filled-in extents on its own. | |
544 | */ | |
545 | int /* error code */ | |
546 | xfs_getbmap( | |
547 | xfs_inode_t *ip, | |
548 | struct getbmapx *bmv, /* user bmap structure */ | |
549 | xfs_bmap_format_t formatter, /* format to user */ | |
550 | void *arg) /* formatter arg */ | |
551 | { | |
552 | __int64_t bmvend; /* last block requested */ | |
553 | int error = 0; /* return value */ | |
554 | __int64_t fixlen; /* length for -1 case */ | |
555 | int i; /* extent number */ | |
556 | int lock; /* lock state */ | |
557 | xfs_bmbt_irec_t *map; /* buffer for user's data */ | |
558 | xfs_mount_t *mp; /* file system mount point */ | |
559 | int nex; /* # of user extents can do */ | |
560 | int nexleft; /* # of user extents left */ | |
561 | int subnex; /* # of bmapi's can do */ | |
562 | int nmap; /* number of map entries */ | |
563 | struct getbmapx *out; /* output structure */ | |
564 | int whichfork; /* data or attr fork */ | |
565 | int prealloced; /* this is a file with | |
566 | * preallocated data space */ | |
567 | int iflags; /* interface flags */ | |
568 | int bmapi_flags; /* flags for xfs_bmapi */ | |
569 | int cur_ext = 0; | |
570 | ||
571 | mp = ip->i_mount; | |
572 | iflags = bmv->bmv_iflags; | |
573 | whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK; | |
574 | ||
575 | if (whichfork == XFS_ATTR_FORK) { | |
576 | if (XFS_IFORK_Q(ip)) { | |
577 | if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS && | |
578 | ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE && | |
579 | ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL) | |
b474c7ae | 580 | return EINVAL; |
68988114 DC |
581 | } else if (unlikely( |
582 | ip->i_d.di_aformat != 0 && | |
583 | ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) { | |
584 | XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW, | |
585 | ip->i_mount); | |
b474c7ae | 586 | return EFSCORRUPTED; |
68988114 DC |
587 | } |
588 | ||
589 | prealloced = 0; | |
590 | fixlen = 1LL << 32; | |
591 | } else { | |
592 | if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS && | |
593 | ip->i_d.di_format != XFS_DINODE_FMT_BTREE && | |
594 | ip->i_d.di_format != XFS_DINODE_FMT_LOCAL) | |
b474c7ae | 595 | return EINVAL; |
68988114 DC |
596 | |
597 | if (xfs_get_extsz_hint(ip) || | |
598 | ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){ | |
599 | prealloced = 1; | |
600 | fixlen = mp->m_super->s_maxbytes; | |
601 | } else { | |
602 | prealloced = 0; | |
603 | fixlen = XFS_ISIZE(ip); | |
604 | } | |
605 | } | |
606 | ||
607 | if (bmv->bmv_length == -1) { | |
608 | fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen)); | |
609 | bmv->bmv_length = | |
610 | max_t(__int64_t, fixlen - bmv->bmv_offset, 0); | |
611 | } else if (bmv->bmv_length == 0) { | |
612 | bmv->bmv_entries = 0; | |
613 | return 0; | |
614 | } else if (bmv->bmv_length < 0) { | |
b474c7ae | 615 | return EINVAL; |
68988114 DC |
616 | } |
617 | ||
618 | nex = bmv->bmv_count - 1; | |
619 | if (nex <= 0) | |
b474c7ae | 620 | return EINVAL; |
68988114 DC |
621 | bmvend = bmv->bmv_offset + bmv->bmv_length; |
622 | ||
623 | ||
624 | if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx)) | |
b474c7ae | 625 | return ENOMEM; |
fdd3ccee DC |
626 | out = kmem_zalloc_large(bmv->bmv_count * sizeof(struct getbmapx), 0); |
627 | if (!out) | |
b474c7ae | 628 | return ENOMEM; |
68988114 DC |
629 | |
630 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
efa70be1 CH |
631 | if (whichfork == XFS_DATA_FORK) { |
632 | if (!(iflags & BMV_IF_DELALLOC) && | |
633 | (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) { | |
68988114 DC |
634 | error = -filemap_write_and_wait(VFS_I(ip)->i_mapping); |
635 | if (error) | |
636 | goto out_unlock_iolock; | |
efa70be1 CH |
637 | |
638 | /* | |
639 | * Even after flushing the inode, there can still be | |
640 | * delalloc blocks on the inode beyond EOF due to | |
641 | * speculative preallocation. These are not removed | |
642 | * until the release function is called or the inode | |
643 | * is inactivated. Hence we cannot assert here that | |
644 | * ip->i_delayed_blks == 0. | |
645 | */ | |
68988114 | 646 | } |
68988114 | 647 | |
efa70be1 CH |
648 | lock = xfs_ilock_data_map_shared(ip); |
649 | } else { | |
650 | lock = xfs_ilock_attr_map_shared(ip); | |
651 | } | |
68988114 DC |
652 | |
653 | /* | |
654 | * Don't let nex be bigger than the number of extents | |
655 | * we can have assuming alternating holes and real extents. | |
656 | */ | |
657 | if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1) | |
658 | nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1; | |
659 | ||
660 | bmapi_flags = xfs_bmapi_aflag(whichfork); | |
661 | if (!(iflags & BMV_IF_PREALLOC)) | |
662 | bmapi_flags |= XFS_BMAPI_IGSTATE; | |
663 | ||
664 | /* | |
665 | * Allocate enough space to handle "subnex" maps at a time. | |
666 | */ | |
667 | error = ENOMEM; | |
668 | subnex = 16; | |
669 | map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS); | |
670 | if (!map) | |
671 | goto out_unlock_ilock; | |
672 | ||
673 | bmv->bmv_entries = 0; | |
674 | ||
675 | if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 && | |
676 | (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) { | |
677 | error = 0; | |
678 | goto out_free_map; | |
679 | } | |
680 | ||
681 | nexleft = nex; | |
682 | ||
683 | do { | |
684 | nmap = (nexleft > subnex) ? subnex : nexleft; | |
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 | ||
692 | for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) { | |
693 | out[cur_ext].bmv_oflags = 0; | |
694 | if (map[i].br_state == XFS_EXT_UNWRITTEN) | |
695 | out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC; | |
696 | else if (map[i].br_startblock == DELAYSTARTBLOCK) | |
697 | out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC; | |
698 | out[cur_ext].bmv_offset = | |
699 | XFS_FSB_TO_BB(mp, map[i].br_startoff); | |
700 | out[cur_ext].bmv_length = | |
701 | XFS_FSB_TO_BB(mp, map[i].br_blockcount); | |
702 | out[cur_ext].bmv_unused1 = 0; | |
703 | out[cur_ext].bmv_unused2 = 0; | |
704 | ||
705 | /* | |
706 | * delayed allocation extents that start beyond EOF can | |
707 | * occur due to speculative EOF allocation when the | |
708 | * delalloc extent is larger than the largest freespace | |
709 | * extent at conversion time. These extents cannot be | |
710 | * converted by data writeback, so can exist here even | |
711 | * if we are not supposed to be finding delalloc | |
712 | * extents. | |
713 | */ | |
714 | if (map[i].br_startblock == DELAYSTARTBLOCK && | |
715 | map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip))) | |
716 | ASSERT((iflags & BMV_IF_DELALLOC) != 0); | |
717 | ||
718 | if (map[i].br_startblock == HOLESTARTBLOCK && | |
719 | whichfork == XFS_ATTR_FORK) { | |
720 | /* came to the end of attribute fork */ | |
721 | out[cur_ext].bmv_oflags |= BMV_OF_LAST; | |
722 | goto out_free_map; | |
723 | } | |
724 | ||
725 | if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext], | |
726 | prealloced, bmvend, | |
727 | map[i].br_startblock)) | |
728 | goto out_free_map; | |
729 | ||
730 | bmv->bmv_offset = | |
731 | out[cur_ext].bmv_offset + | |
732 | out[cur_ext].bmv_length; | |
733 | bmv->bmv_length = | |
734 | max_t(__int64_t, 0, bmvend - bmv->bmv_offset); | |
735 | ||
736 | /* | |
737 | * In case we don't want to return the hole, | |
738 | * don't increase cur_ext so that we can reuse | |
739 | * it in the next loop. | |
740 | */ | |
741 | if ((iflags & BMV_IF_NO_HOLES) && | |
742 | map[i].br_startblock == HOLESTARTBLOCK) { | |
743 | memset(&out[cur_ext], 0, sizeof(out[cur_ext])); | |
744 | continue; | |
745 | } | |
746 | ||
747 | nexleft--; | |
748 | bmv->bmv_entries++; | |
749 | cur_ext++; | |
750 | } | |
751 | } while (nmap && nexleft && bmv->bmv_length); | |
752 | ||
753 | out_free_map: | |
754 | kmem_free(map); | |
755 | out_unlock_ilock: | |
01f4f327 | 756 | xfs_iunlock(ip, lock); |
68988114 DC |
757 | out_unlock_iolock: |
758 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
759 | ||
760 | for (i = 0; i < cur_ext; i++) { | |
761 | int full = 0; /* user array is full */ | |
762 | ||
763 | /* format results & advance arg */ | |
764 | error = formatter(&arg, &out[i], &full); | |
765 | if (error || full) | |
766 | break; | |
767 | } | |
768 | ||
fdd3ccee | 769 | kmem_free(out); |
68988114 DC |
770 | return error; |
771 | } | |
772 | ||
773 | /* | |
774 | * dead simple method of punching delalyed allocation blocks from a range in | |
775 | * the inode. Walks a block at a time so will be slow, but is only executed in | |
ad4809bf | 776 | * rare error cases so the overhead is not critical. This will always punch out |
68988114 DC |
777 | * both the start and end blocks, even if the ranges only partially overlap |
778 | * them, so it is up to the caller to ensure that partial blocks are not | |
779 | * passed in. | |
780 | */ | |
781 | int | |
782 | xfs_bmap_punch_delalloc_range( | |
783 | struct xfs_inode *ip, | |
784 | xfs_fileoff_t start_fsb, | |
785 | xfs_fileoff_t length) | |
786 | { | |
787 | xfs_fileoff_t remaining = length; | |
788 | int error = 0; | |
789 | ||
790 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
791 | ||
792 | do { | |
793 | int done; | |
794 | xfs_bmbt_irec_t imap; | |
795 | int nimaps = 1; | |
796 | xfs_fsblock_t firstblock; | |
797 | xfs_bmap_free_t flist; | |
798 | ||
799 | /* | |
800 | * Map the range first and check that it is a delalloc extent | |
801 | * before trying to unmap the range. Otherwise we will be | |
802 | * trying to remove a real extent (which requires a | |
803 | * transaction) or a hole, which is probably a bad idea... | |
804 | */ | |
805 | error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps, | |
806 | XFS_BMAPI_ENTIRE); | |
807 | ||
808 | if (error) { | |
809 | /* something screwed, just bail */ | |
810 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
811 | xfs_alert(ip->i_mount, | |
812 | "Failed delalloc mapping lookup ino %lld fsb %lld.", | |
813 | ip->i_ino, start_fsb); | |
814 | } | |
815 | break; | |
816 | } | |
817 | if (!nimaps) { | |
818 | /* nothing there */ | |
819 | goto next_block; | |
820 | } | |
821 | if (imap.br_startblock != DELAYSTARTBLOCK) { | |
822 | /* been converted, ignore */ | |
823 | goto next_block; | |
824 | } | |
825 | WARN_ON(imap.br_blockcount == 0); | |
826 | ||
827 | /* | |
828 | * Note: while we initialise the firstblock/flist pair, they | |
829 | * should never be used because blocks should never be | |
830 | * allocated or freed for a delalloc extent and hence we need | |
831 | * don't cancel or finish them after the xfs_bunmapi() call. | |
832 | */ | |
833 | xfs_bmap_init(&flist, &firstblock); | |
834 | error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock, | |
835 | &flist, &done); | |
836 | if (error) | |
837 | break; | |
838 | ||
839 | ASSERT(!flist.xbf_count && !flist.xbf_first); | |
840 | next_block: | |
841 | start_fsb++; | |
842 | remaining--; | |
843 | } while(remaining > 0); | |
844 | ||
845 | return error; | |
846 | } | |
c24b5dfa DC |
847 | |
848 | /* | |
849 | * Test whether it is appropriate to check an inode for and free post EOF | |
850 | * blocks. The 'force' parameter determines whether we should also consider | |
851 | * regular files that are marked preallocated or append-only. | |
852 | */ | |
853 | bool | |
854 | xfs_can_free_eofblocks(struct xfs_inode *ip, bool force) | |
855 | { | |
856 | /* prealloc/delalloc exists only on regular files */ | |
857 | if (!S_ISREG(ip->i_d.di_mode)) | |
858 | return false; | |
859 | ||
860 | /* | |
861 | * Zero sized files with no cached pages and delalloc blocks will not | |
862 | * have speculative prealloc/delalloc blocks to remove. | |
863 | */ | |
864 | if (VFS_I(ip)->i_size == 0 && | |
865 | VN_CACHED(VFS_I(ip)) == 0 && | |
866 | ip->i_delayed_blks == 0) | |
867 | return false; | |
868 | ||
869 | /* If we haven't read in the extent list, then don't do it now. */ | |
870 | if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) | |
871 | return false; | |
872 | ||
873 | /* | |
874 | * Do not free real preallocated or append-only files unless the file | |
875 | * has delalloc blocks and we are forced to remove them. | |
876 | */ | |
877 | if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) | |
878 | if (!force || ip->i_delayed_blks == 0) | |
879 | return false; | |
880 | ||
881 | return true; | |
882 | } | |
883 | ||
884 | /* | |
885 | * This is called by xfs_inactive to free any blocks beyond eof | |
886 | * when the link count isn't zero and by xfs_dm_punch_hole() when | |
887 | * punching a hole to EOF. | |
888 | */ | |
889 | int | |
890 | xfs_free_eofblocks( | |
891 | xfs_mount_t *mp, | |
892 | xfs_inode_t *ip, | |
893 | bool need_iolock) | |
894 | { | |
895 | xfs_trans_t *tp; | |
896 | int error; | |
897 | xfs_fileoff_t end_fsb; | |
898 | xfs_fileoff_t last_fsb; | |
899 | xfs_filblks_t map_len; | |
900 | int nimaps; | |
901 | xfs_bmbt_irec_t imap; | |
902 | ||
903 | /* | |
904 | * Figure out if there are any blocks beyond the end | |
905 | * of the file. If not, then there is nothing to do. | |
906 | */ | |
907 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip)); | |
908 | last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); | |
909 | if (last_fsb <= end_fsb) | |
910 | return 0; | |
911 | map_len = last_fsb - end_fsb; | |
912 | ||
913 | nimaps = 1; | |
914 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
915 | error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0); | |
916 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
917 | ||
918 | if (!error && (nimaps != 0) && | |
919 | (imap.br_startblock != HOLESTARTBLOCK || | |
920 | ip->i_delayed_blks)) { | |
921 | /* | |
922 | * Attach the dquots to the inode up front. | |
923 | */ | |
924 | error = xfs_qm_dqattach(ip, 0); | |
925 | if (error) | |
926 | return error; | |
927 | ||
928 | /* | |
929 | * There are blocks after the end of file. | |
930 | * Free them up now by truncating the file to | |
931 | * its current size. | |
932 | */ | |
933 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
934 | ||
935 | if (need_iolock) { | |
936 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { | |
937 | xfs_trans_cancel(tp, 0); | |
938 | return EAGAIN; | |
939 | } | |
940 | } | |
941 | ||
3d3c8b52 | 942 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0); |
c24b5dfa DC |
943 | if (error) { |
944 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
945 | xfs_trans_cancel(tp, 0); | |
946 | if (need_iolock) | |
947 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
948 | return error; | |
949 | } | |
950 | ||
951 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
952 | xfs_trans_ijoin(tp, ip, 0); | |
953 | ||
954 | /* | |
955 | * Do not update the on-disk file size. If we update the | |
956 | * on-disk file size and then the system crashes before the | |
957 | * contents of the file are flushed to disk then the files | |
958 | * may be full of holes (ie NULL files bug). | |
959 | */ | |
960 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, | |
961 | XFS_ISIZE(ip)); | |
962 | if (error) { | |
963 | /* | |
964 | * If we get an error at this point we simply don't | |
965 | * bother truncating the file. | |
966 | */ | |
967 | xfs_trans_cancel(tp, | |
968 | (XFS_TRANS_RELEASE_LOG_RES | | |
969 | XFS_TRANS_ABORT)); | |
970 | } else { | |
971 | error = xfs_trans_commit(tp, | |
972 | XFS_TRANS_RELEASE_LOG_RES); | |
973 | if (!error) | |
974 | xfs_inode_clear_eofblocks_tag(ip); | |
975 | } | |
976 | ||
977 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
978 | if (need_iolock) | |
979 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
980 | } | |
981 | return error; | |
982 | } | |
983 | ||
83aee9e4 | 984 | int |
c24b5dfa | 985 | xfs_alloc_file_space( |
83aee9e4 | 986 | struct xfs_inode *ip, |
c24b5dfa DC |
987 | xfs_off_t offset, |
988 | xfs_off_t len, | |
5f8aca8b | 989 | int alloc_type) |
c24b5dfa DC |
990 | { |
991 | xfs_mount_t *mp = ip->i_mount; | |
992 | xfs_off_t count; | |
993 | xfs_filblks_t allocated_fsb; | |
994 | xfs_filblks_t allocatesize_fsb; | |
995 | xfs_extlen_t extsz, temp; | |
996 | xfs_fileoff_t startoffset_fsb; | |
997 | xfs_fsblock_t firstfsb; | |
998 | int nimaps; | |
999 | int quota_flag; | |
1000 | int rt; | |
1001 | xfs_trans_t *tp; | |
1002 | xfs_bmbt_irec_t imaps[1], *imapp; | |
1003 | xfs_bmap_free_t free_list; | |
1004 | uint qblocks, resblks, resrtextents; | |
1005 | int committed; | |
1006 | int error; | |
1007 | ||
1008 | trace_xfs_alloc_file_space(ip); | |
1009 | ||
1010 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 1011 | return EIO; |
c24b5dfa DC |
1012 | |
1013 | error = xfs_qm_dqattach(ip, 0); | |
1014 | if (error) | |
1015 | return error; | |
1016 | ||
1017 | if (len <= 0) | |
b474c7ae | 1018 | return EINVAL; |
c24b5dfa DC |
1019 | |
1020 | rt = XFS_IS_REALTIME_INODE(ip); | |
1021 | extsz = xfs_get_extsz_hint(ip); | |
1022 | ||
1023 | count = len; | |
1024 | imapp = &imaps[0]; | |
1025 | nimaps = 1; | |
1026 | startoffset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1027 | allocatesize_fsb = XFS_B_TO_FSB(mp, count); | |
1028 | ||
1029 | /* | |
1030 | * Allocate file space until done or until there is an error | |
1031 | */ | |
1032 | while (allocatesize_fsb && !error) { | |
1033 | xfs_fileoff_t s, e; | |
1034 | ||
1035 | /* | |
1036 | * Determine space reservations for data/realtime. | |
1037 | */ | |
1038 | if (unlikely(extsz)) { | |
1039 | s = startoffset_fsb; | |
1040 | do_div(s, extsz); | |
1041 | s *= extsz; | |
1042 | e = startoffset_fsb + allocatesize_fsb; | |
1043 | if ((temp = do_mod(startoffset_fsb, extsz))) | |
1044 | e += temp; | |
1045 | if ((temp = do_mod(e, extsz))) | |
1046 | e += extsz - temp; | |
1047 | } else { | |
1048 | s = 0; | |
1049 | e = allocatesize_fsb; | |
1050 | } | |
1051 | ||
1052 | /* | |
1053 | * The transaction reservation is limited to a 32-bit block | |
1054 | * count, hence we need to limit the number of blocks we are | |
1055 | * trying to reserve to avoid an overflow. We can't allocate | |
1056 | * more than @nimaps extents, and an extent is limited on disk | |
1057 | * to MAXEXTLEN (21 bits), so use that to enforce the limit. | |
1058 | */ | |
1059 | resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps)); | |
1060 | if (unlikely(rt)) { | |
1061 | resrtextents = qblocks = resblks; | |
1062 | resrtextents /= mp->m_sb.sb_rextsize; | |
1063 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
1064 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
1065 | } else { | |
1066 | resrtextents = 0; | |
1067 | resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks); | |
1068 | quota_flag = XFS_QMOPT_RES_REGBLKS; | |
1069 | } | |
1070 | ||
1071 | /* | |
1072 | * Allocate and setup the transaction. | |
1073 | */ | |
1074 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
3d3c8b52 JL |
1075 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, |
1076 | resblks, resrtextents); | |
c24b5dfa DC |
1077 | /* |
1078 | * Check for running out of space | |
1079 | */ | |
1080 | if (error) { | |
1081 | /* | |
1082 | * Free the transaction structure. | |
1083 | */ | |
1084 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
1085 | xfs_trans_cancel(tp, 0); | |
1086 | break; | |
1087 | } | |
1088 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1089 | error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, | |
1090 | 0, quota_flag); | |
1091 | if (error) | |
1092 | goto error1; | |
1093 | ||
1094 | xfs_trans_ijoin(tp, ip, 0); | |
1095 | ||
1096 | xfs_bmap_init(&free_list, &firstfsb); | |
1097 | error = xfs_bmapi_write(tp, ip, startoffset_fsb, | |
1098 | allocatesize_fsb, alloc_type, &firstfsb, | |
1099 | 0, imapp, &nimaps, &free_list); | |
1100 | if (error) { | |
1101 | goto error0; | |
1102 | } | |
1103 | ||
1104 | /* | |
1105 | * Complete the transaction | |
1106 | */ | |
1107 | error = xfs_bmap_finish(&tp, &free_list, &committed); | |
1108 | if (error) { | |
1109 | goto error0; | |
1110 | } | |
1111 | ||
1112 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); | |
1113 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1114 | if (error) { | |
1115 | break; | |
1116 | } | |
1117 | ||
1118 | allocated_fsb = imapp->br_blockcount; | |
1119 | ||
1120 | if (nimaps == 0) { | |
b474c7ae | 1121 | error = ENOSPC; |
c24b5dfa DC |
1122 | break; |
1123 | } | |
1124 | ||
1125 | startoffset_fsb += allocated_fsb; | |
1126 | allocatesize_fsb -= allocated_fsb; | |
1127 | } | |
1128 | ||
1129 | return error; | |
1130 | ||
1131 | error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */ | |
1132 | xfs_bmap_cancel(&free_list); | |
1133 | xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag); | |
1134 | ||
1135 | error1: /* Just cancel transaction */ | |
1136 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1137 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1138 | return error; | |
1139 | } | |
1140 | ||
1141 | /* | |
1142 | * Zero file bytes between startoff and endoff inclusive. | |
1143 | * The iolock is held exclusive and no blocks are buffered. | |
1144 | * | |
1145 | * This function is used by xfs_free_file_space() to zero | |
1146 | * partial blocks when the range to free is not block aligned. | |
1147 | * When unreserving space with boundaries that are not block | |
1148 | * aligned we round up the start and round down the end | |
1149 | * boundaries and then use this function to zero the parts of | |
1150 | * the blocks that got dropped during the rounding. | |
1151 | */ | |
1152 | STATIC int | |
1153 | xfs_zero_remaining_bytes( | |
1154 | xfs_inode_t *ip, | |
1155 | xfs_off_t startoff, | |
1156 | xfs_off_t endoff) | |
1157 | { | |
1158 | xfs_bmbt_irec_t imap; | |
1159 | xfs_fileoff_t offset_fsb; | |
1160 | xfs_off_t lastoffset; | |
1161 | xfs_off_t offset; | |
1162 | xfs_buf_t *bp; | |
1163 | xfs_mount_t *mp = ip->i_mount; | |
1164 | int nimap; | |
1165 | int error = 0; | |
1166 | ||
1167 | /* | |
1168 | * Avoid doing I/O beyond eof - it's not necessary | |
1169 | * since nothing can read beyond eof. The space will | |
1170 | * be zeroed when the file is extended anyway. | |
1171 | */ | |
1172 | if (startoff >= XFS_ISIZE(ip)) | |
1173 | return 0; | |
1174 | ||
1175 | if (endoff > XFS_ISIZE(ip)) | |
1176 | endoff = XFS_ISIZE(ip); | |
1177 | ||
1178 | bp = xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip) ? | |
1179 | mp->m_rtdev_targp : mp->m_ddev_targp, | |
1180 | BTOBB(mp->m_sb.sb_blocksize), 0); | |
1181 | if (!bp) | |
b474c7ae | 1182 | return ENOMEM; |
c24b5dfa DC |
1183 | |
1184 | xfs_buf_unlock(bp); | |
1185 | ||
1186 | for (offset = startoff; offset <= endoff; offset = lastoffset + 1) { | |
4f317369 CH |
1187 | uint lock_mode; |
1188 | ||
c24b5dfa DC |
1189 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
1190 | nimap = 1; | |
4f317369 CH |
1191 | |
1192 | lock_mode = xfs_ilock_data_map_shared(ip); | |
c24b5dfa | 1193 | error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0); |
4f317369 CH |
1194 | xfs_iunlock(ip, lock_mode); |
1195 | ||
c24b5dfa DC |
1196 | if (error || nimap < 1) |
1197 | break; | |
1198 | ASSERT(imap.br_blockcount >= 1); | |
1199 | ASSERT(imap.br_startoff == offset_fsb); | |
1200 | lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1; | |
1201 | if (lastoffset > endoff) | |
1202 | lastoffset = endoff; | |
1203 | if (imap.br_startblock == HOLESTARTBLOCK) | |
1204 | continue; | |
1205 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
1206 | if (imap.br_state == XFS_EXT_UNWRITTEN) | |
1207 | continue; | |
1208 | XFS_BUF_UNDONE(bp); | |
1209 | XFS_BUF_UNWRITE(bp); | |
1210 | XFS_BUF_READ(bp); | |
1211 | XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock)); | |
83a0adc3 CH |
1212 | |
1213 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
b474c7ae | 1214 | error = EIO; |
83a0adc3 CH |
1215 | break; |
1216 | } | |
1217 | xfs_buf_iorequest(bp); | |
c24b5dfa DC |
1218 | error = xfs_buf_iowait(bp); |
1219 | if (error) { | |
1220 | xfs_buf_ioerror_alert(bp, | |
1221 | "xfs_zero_remaining_bytes(read)"); | |
1222 | break; | |
1223 | } | |
1224 | memset(bp->b_addr + | |
1225 | (offset - XFS_FSB_TO_B(mp, imap.br_startoff)), | |
1226 | 0, lastoffset - offset + 1); | |
1227 | XFS_BUF_UNDONE(bp); | |
1228 | XFS_BUF_UNREAD(bp); | |
1229 | XFS_BUF_WRITE(bp); | |
83a0adc3 CH |
1230 | |
1231 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
b474c7ae | 1232 | error = EIO; |
83a0adc3 CH |
1233 | break; |
1234 | } | |
1235 | xfs_buf_iorequest(bp); | |
c24b5dfa DC |
1236 | error = xfs_buf_iowait(bp); |
1237 | if (error) { | |
1238 | xfs_buf_ioerror_alert(bp, | |
1239 | "xfs_zero_remaining_bytes(write)"); | |
1240 | break; | |
1241 | } | |
1242 | } | |
1243 | xfs_buf_free(bp); | |
1244 | return error; | |
1245 | } | |
1246 | ||
83aee9e4 | 1247 | int |
c24b5dfa | 1248 | xfs_free_file_space( |
83aee9e4 | 1249 | struct xfs_inode *ip, |
c24b5dfa | 1250 | xfs_off_t offset, |
5f8aca8b | 1251 | xfs_off_t len) |
c24b5dfa DC |
1252 | { |
1253 | int committed; | |
1254 | int done; | |
1255 | xfs_fileoff_t endoffset_fsb; | |
1256 | int error; | |
1257 | xfs_fsblock_t firstfsb; | |
1258 | xfs_bmap_free_t free_list; | |
1259 | xfs_bmbt_irec_t imap; | |
1260 | xfs_off_t ioffset; | |
1261 | xfs_extlen_t mod=0; | |
1262 | xfs_mount_t *mp; | |
1263 | int nimap; | |
1264 | uint resblks; | |
1265 | xfs_off_t rounding; | |
1266 | int rt; | |
1267 | xfs_fileoff_t startoffset_fsb; | |
1268 | xfs_trans_t *tp; | |
c24b5dfa DC |
1269 | |
1270 | mp = ip->i_mount; | |
1271 | ||
1272 | trace_xfs_free_file_space(ip); | |
1273 | ||
1274 | error = xfs_qm_dqattach(ip, 0); | |
1275 | if (error) | |
1276 | return error; | |
1277 | ||
1278 | error = 0; | |
1279 | if (len <= 0) /* if nothing being freed */ | |
1280 | return error; | |
1281 | rt = XFS_IS_REALTIME_INODE(ip); | |
1282 | startoffset_fsb = XFS_B_TO_FSB(mp, offset); | |
1283 | endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len); | |
1284 | ||
5f8aca8b CH |
1285 | /* wait for the completion of any pending DIOs */ |
1286 | inode_dio_wait(VFS_I(ip)); | |
c24b5dfa DC |
1287 | |
1288 | rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE); | |
1289 | ioffset = offset & ~(rounding - 1); | |
1290 | error = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping, | |
1291 | ioffset, -1); | |
1292 | if (error) | |
5f8aca8b | 1293 | goto out; |
c24b5dfa DC |
1294 | truncate_pagecache_range(VFS_I(ip), ioffset, -1); |
1295 | ||
1296 | /* | |
1297 | * Need to zero the stuff we're not freeing, on disk. | |
1298 | * If it's a realtime file & can't use unwritten extents then we | |
1299 | * actually need to zero the extent edges. Otherwise xfs_bunmapi | |
1300 | * will take care of it for us. | |
1301 | */ | |
1302 | if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) { | |
1303 | nimap = 1; | |
1304 | error = xfs_bmapi_read(ip, startoffset_fsb, 1, | |
1305 | &imap, &nimap, 0); | |
1306 | if (error) | |
5f8aca8b | 1307 | goto out; |
c24b5dfa DC |
1308 | ASSERT(nimap == 0 || nimap == 1); |
1309 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
1310 | xfs_daddr_t block; | |
1311 | ||
1312 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
1313 | block = imap.br_startblock; | |
1314 | mod = do_div(block, mp->m_sb.sb_rextsize); | |
1315 | if (mod) | |
1316 | startoffset_fsb += mp->m_sb.sb_rextsize - mod; | |
1317 | } | |
1318 | nimap = 1; | |
1319 | error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1, | |
1320 | &imap, &nimap, 0); | |
1321 | if (error) | |
5f8aca8b | 1322 | goto out; |
c24b5dfa DC |
1323 | ASSERT(nimap == 0 || nimap == 1); |
1324 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
1325 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
1326 | mod++; | |
1327 | if (mod && (mod != mp->m_sb.sb_rextsize)) | |
1328 | endoffset_fsb -= mod; | |
1329 | } | |
1330 | } | |
1331 | if ((done = (endoffset_fsb <= startoffset_fsb))) | |
1332 | /* | |
1333 | * One contiguous piece to clear | |
1334 | */ | |
1335 | error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1); | |
1336 | else { | |
1337 | /* | |
1338 | * Some full blocks, possibly two pieces to clear | |
1339 | */ | |
1340 | if (offset < XFS_FSB_TO_B(mp, startoffset_fsb)) | |
1341 | error = xfs_zero_remaining_bytes(ip, offset, | |
1342 | XFS_FSB_TO_B(mp, startoffset_fsb) - 1); | |
1343 | if (!error && | |
1344 | XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len) | |
1345 | error = xfs_zero_remaining_bytes(ip, | |
1346 | XFS_FSB_TO_B(mp, endoffset_fsb), | |
1347 | offset + len - 1); | |
1348 | } | |
1349 | ||
1350 | /* | |
1351 | * free file space until done or until there is an error | |
1352 | */ | |
1353 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
1354 | while (!error && !done) { | |
1355 | ||
1356 | /* | |
1357 | * allocate and setup the transaction. Allow this | |
1358 | * transaction to dip into the reserve blocks to ensure | |
1359 | * the freeing of the space succeeds at ENOSPC. | |
1360 | */ | |
1361 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
3d3c8b52 | 1362 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, resblks, 0); |
c24b5dfa DC |
1363 | |
1364 | /* | |
1365 | * check for running out of space | |
1366 | */ | |
1367 | if (error) { | |
1368 | /* | |
1369 | * Free the transaction structure. | |
1370 | */ | |
1371 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
1372 | xfs_trans_cancel(tp, 0); | |
1373 | break; | |
1374 | } | |
1375 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1376 | error = xfs_trans_reserve_quota(tp, mp, | |
1377 | ip->i_udquot, ip->i_gdquot, ip->i_pdquot, | |
1378 | resblks, 0, XFS_QMOPT_RES_REGBLKS); | |
1379 | if (error) | |
1380 | goto error1; | |
1381 | ||
1382 | xfs_trans_ijoin(tp, ip, 0); | |
1383 | ||
1384 | /* | |
1385 | * issue the bunmapi() call to free the blocks | |
1386 | */ | |
1387 | xfs_bmap_init(&free_list, &firstfsb); | |
1388 | error = xfs_bunmapi(tp, ip, startoffset_fsb, | |
1389 | endoffset_fsb - startoffset_fsb, | |
1390 | 0, 2, &firstfsb, &free_list, &done); | |
1391 | if (error) { | |
1392 | goto error0; | |
1393 | } | |
1394 | ||
1395 | /* | |
1396 | * complete the transaction | |
1397 | */ | |
1398 | error = xfs_bmap_finish(&tp, &free_list, &committed); | |
1399 | if (error) { | |
1400 | goto error0; | |
1401 | } | |
1402 | ||
1403 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); | |
1404 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1405 | } | |
1406 | ||
5f8aca8b | 1407 | out: |
c24b5dfa DC |
1408 | return error; |
1409 | ||
1410 | error0: | |
1411 | xfs_bmap_cancel(&free_list); | |
1412 | error1: | |
1413 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
5f8aca8b CH |
1414 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1415 | goto out; | |
c24b5dfa DC |
1416 | } |
1417 | ||
1418 | ||
865e9446 | 1419 | int |
c24b5dfa DC |
1420 | xfs_zero_file_space( |
1421 | struct xfs_inode *ip, | |
1422 | xfs_off_t offset, | |
5f8aca8b | 1423 | xfs_off_t len) |
c24b5dfa DC |
1424 | { |
1425 | struct xfs_mount *mp = ip->i_mount; | |
1426 | uint granularity; | |
1427 | xfs_off_t start_boundary; | |
1428 | xfs_off_t end_boundary; | |
1429 | int error; | |
1430 | ||
897b73b6 DC |
1431 | trace_xfs_zero_file_space(ip); |
1432 | ||
c24b5dfa DC |
1433 | granularity = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE); |
1434 | ||
1435 | /* | |
1436 | * Round the range of extents we are going to convert inwards. If the | |
1437 | * offset is aligned, then it doesn't get changed so we zero from the | |
1438 | * start of the block offset points to. | |
1439 | */ | |
1440 | start_boundary = round_up(offset, granularity); | |
1441 | end_boundary = round_down(offset + len, granularity); | |
1442 | ||
1443 | ASSERT(start_boundary >= offset); | |
1444 | ASSERT(end_boundary <= offset + len); | |
1445 | ||
c24b5dfa | 1446 | if (start_boundary < end_boundary - 1) { |
897b73b6 DC |
1447 | /* |
1448 | * punch out delayed allocation blocks and the page cache over | |
1449 | * the conversion range | |
1450 | */ | |
1451 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1452 | error = xfs_bmap_punch_delalloc_range(ip, | |
1453 | XFS_B_TO_FSBT(mp, start_boundary), | |
1454 | XFS_B_TO_FSB(mp, end_boundary - start_boundary)); | |
1455 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
c24b5dfa DC |
1456 | truncate_pagecache_range(VFS_I(ip), start_boundary, |
1457 | end_boundary - 1); | |
897b73b6 | 1458 | |
c24b5dfa DC |
1459 | /* convert the blocks */ |
1460 | error = xfs_alloc_file_space(ip, start_boundary, | |
1461 | end_boundary - start_boundary - 1, | |
5f8aca8b | 1462 | XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT); |
c24b5dfa | 1463 | if (error) |
5f8aca8b | 1464 | goto out; |
c24b5dfa DC |
1465 | |
1466 | /* We've handled the interior of the range, now for the edges */ | |
5f8aca8b | 1467 | if (start_boundary != offset) { |
c24b5dfa | 1468 | error = xfs_iozero(ip, offset, start_boundary - offset); |
5f8aca8b CH |
1469 | if (error) |
1470 | goto out; | |
1471 | } | |
c24b5dfa DC |
1472 | |
1473 | if (end_boundary != offset + len) | |
1474 | error = xfs_iozero(ip, end_boundary, | |
1475 | offset + len - end_boundary); | |
1476 | ||
1477 | } else { | |
1478 | /* | |
1479 | * It's either a sub-granularity range or the range spanned lies | |
1480 | * partially across two adjacent blocks. | |
1481 | */ | |
1482 | error = xfs_iozero(ip, offset, len); | |
1483 | } | |
1484 | ||
5f8aca8b | 1485 | out: |
c24b5dfa DC |
1486 | return error; |
1487 | ||
1488 | } | |
1489 | ||
e1d8fb88 NJ |
1490 | /* |
1491 | * xfs_collapse_file_space() | |
1492 | * This routine frees disk space and shift extent for the given file. | |
1493 | * The first thing we do is to free data blocks in the specified range | |
1494 | * by calling xfs_free_file_space(). It would also sync dirty data | |
1495 | * and invalidate page cache over the region on which collapse range | |
1496 | * is working. And Shift extent records to the left to cover a hole. | |
1497 | * RETURNS: | |
1498 | * 0 on success | |
1499 | * errno on error | |
1500 | * | |
1501 | */ | |
1502 | int | |
1503 | xfs_collapse_file_space( | |
1504 | struct xfs_inode *ip, | |
1505 | xfs_off_t offset, | |
1506 | xfs_off_t len) | |
1507 | { | |
1508 | int done = 0; | |
1509 | struct xfs_mount *mp = ip->i_mount; | |
1510 | struct xfs_trans *tp; | |
1511 | int error; | |
1512 | xfs_extnum_t current_ext = 0; | |
1513 | struct xfs_bmap_free free_list; | |
1514 | xfs_fsblock_t first_block; | |
1515 | int committed; | |
1516 | xfs_fileoff_t start_fsb; | |
1517 | xfs_fileoff_t shift_fsb; | |
1518 | ||
1519 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
1520 | ||
1521 | trace_xfs_collapse_file_space(ip); | |
1522 | ||
1523 | start_fsb = XFS_B_TO_FSB(mp, offset + len); | |
1524 | shift_fsb = XFS_B_TO_FSB(mp, len); | |
1525 | ||
1526 | error = xfs_free_file_space(ip, offset, len); | |
1527 | if (error) | |
1528 | return error; | |
1529 | ||
1530 | while (!error && !done) { | |
1531 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
e1d8fb88 NJ |
1532 | /* |
1533 | * We would need to reserve permanent block for transaction. | |
1534 | * This will come into picture when after shifting extent into | |
1535 | * hole we found that adjacent extents can be merged which | |
1536 | * may lead to freeing of a block during record update. | |
1537 | */ | |
1538 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, | |
1539 | XFS_DIOSTRAT_SPACE_RES(mp, 0), 0); | |
1540 | if (error) { | |
e1d8fb88 NJ |
1541 | xfs_trans_cancel(tp, 0); |
1542 | break; | |
1543 | } | |
1544 | ||
1545 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1546 | error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot, | |
1547 | ip->i_gdquot, ip->i_pdquot, | |
1548 | XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, | |
1549 | XFS_QMOPT_RES_REGBLKS); | |
1550 | if (error) | |
1551 | goto out; | |
1552 | ||
1553 | xfs_trans_ijoin(tp, ip, 0); | |
1554 | ||
1555 | xfs_bmap_init(&free_list, &first_block); | |
1556 | ||
1557 | /* | |
1558 | * We are using the write transaction in which max 2 bmbt | |
1559 | * updates are allowed | |
1560 | */ | |
1561 | error = xfs_bmap_shift_extents(tp, ip, &done, start_fsb, | |
1562 | shift_fsb, ¤t_ext, | |
1563 | &first_block, &free_list, | |
1564 | XFS_BMAP_MAX_SHIFT_EXTENTS); | |
1565 | if (error) | |
1566 | goto out; | |
1567 | ||
1568 | error = xfs_bmap_finish(&tp, &free_list, &committed); | |
1569 | if (error) | |
1570 | goto out; | |
1571 | ||
1572 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); | |
1573 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1574 | } | |
1575 | ||
1576 | return error; | |
1577 | ||
1578 | out: | |
1579 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1580 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1581 | return error; | |
1582 | } | |
1583 | ||
a133d952 DC |
1584 | /* |
1585 | * We need to check that the format of the data fork in the temporary inode is | |
1586 | * valid for the target inode before doing the swap. This is not a problem with | |
1587 | * attr1 because of the fixed fork offset, but attr2 has a dynamically sized | |
1588 | * data fork depending on the space the attribute fork is taking so we can get | |
1589 | * invalid formats on the target inode. | |
1590 | * | |
1591 | * E.g. target has space for 7 extents in extent format, temp inode only has | |
1592 | * space for 6. If we defragment down to 7 extents, then the tmp format is a | |
1593 | * btree, but when swapped it needs to be in extent format. Hence we can't just | |
1594 | * blindly swap data forks on attr2 filesystems. | |
1595 | * | |
1596 | * Note that we check the swap in both directions so that we don't end up with | |
1597 | * a corrupt temporary inode, either. | |
1598 | * | |
1599 | * Note that fixing the way xfs_fsr sets up the attribute fork in the source | |
1600 | * inode will prevent this situation from occurring, so all we do here is | |
1601 | * reject and log the attempt. basically we are putting the responsibility on | |
1602 | * userspace to get this right. | |
1603 | */ | |
1604 | static int | |
1605 | xfs_swap_extents_check_format( | |
1606 | xfs_inode_t *ip, /* target inode */ | |
1607 | xfs_inode_t *tip) /* tmp inode */ | |
1608 | { | |
1609 | ||
1610 | /* Should never get a local format */ | |
1611 | if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL || | |
1612 | tip->i_d.di_format == XFS_DINODE_FMT_LOCAL) | |
1613 | return EINVAL; | |
1614 | ||
1615 | /* | |
1616 | * if the target inode has less extents that then temporary inode then | |
1617 | * why did userspace call us? | |
1618 | */ | |
1619 | if (ip->i_d.di_nextents < tip->i_d.di_nextents) | |
1620 | return EINVAL; | |
1621 | ||
1622 | /* | |
1623 | * if the target inode is in extent form and the temp inode is in btree | |
1624 | * form then we will end up with the target inode in the wrong format | |
1625 | * as we already know there are less extents in the temp inode. | |
1626 | */ | |
1627 | if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS && | |
1628 | tip->i_d.di_format == XFS_DINODE_FMT_BTREE) | |
1629 | return EINVAL; | |
1630 | ||
1631 | /* Check temp in extent form to max in target */ | |
1632 | if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS && | |
1633 | XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) > | |
1634 | XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)) | |
1635 | return EINVAL; | |
1636 | ||
1637 | /* Check target in extent form to max in temp */ | |
1638 | if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS && | |
1639 | XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) > | |
1640 | XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK)) | |
1641 | return EINVAL; | |
1642 | ||
1643 | /* | |
1644 | * If we are in a btree format, check that the temp root block will fit | |
1645 | * in the target and that it has enough extents to be in btree format | |
1646 | * in the target. | |
1647 | * | |
1648 | * Note that we have to be careful to allow btree->extent conversions | |
1649 | * (a common defrag case) which will occur when the temp inode is in | |
1650 | * extent format... | |
1651 | */ | |
1652 | if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) { | |
1653 | if (XFS_IFORK_BOFF(ip) && | |
1654 | XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip)) | |
1655 | return EINVAL; | |
1656 | if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <= | |
1657 | XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)) | |
1658 | return EINVAL; | |
1659 | } | |
1660 | ||
1661 | /* Reciprocal target->temp btree format checks */ | |
1662 | if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) { | |
1663 | if (XFS_IFORK_BOFF(tip) && | |
1664 | XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip)) | |
1665 | return EINVAL; | |
1666 | if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <= | |
1667 | XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK)) | |
1668 | return EINVAL; | |
1669 | } | |
1670 | ||
1671 | return 0; | |
1672 | } | |
1673 | ||
1674 | int | |
1675 | xfs_swap_extents( | |
1676 | xfs_inode_t *ip, /* target inode */ | |
1677 | xfs_inode_t *tip, /* tmp inode */ | |
1678 | xfs_swapext_t *sxp) | |
1679 | { | |
1680 | xfs_mount_t *mp = ip->i_mount; | |
1681 | xfs_trans_t *tp; | |
1682 | xfs_bstat_t *sbp = &sxp->sx_stat; | |
1683 | xfs_ifork_t *tempifp, *ifp, *tifp; | |
1684 | int src_log_flags, target_log_flags; | |
1685 | int error = 0; | |
1686 | int aforkblks = 0; | |
1687 | int taforkblks = 0; | |
1688 | __uint64_t tmp; | |
1689 | ||
a133d952 DC |
1690 | tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL); |
1691 | if (!tempifp) { | |
b474c7ae | 1692 | error = ENOMEM; |
a133d952 DC |
1693 | goto out; |
1694 | } | |
1695 | ||
1696 | /* | |
1697 | * we have to do two separate lock calls here to keep lockdep | |
1698 | * happy. If we try to get all the locks in one call, lock will | |
1699 | * report false positives when we drop the ILOCK and regain them | |
1700 | * below. | |
1701 | */ | |
1702 | xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL); | |
1703 | xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL); | |
1704 | ||
1705 | /* Verify that both files have the same format */ | |
1706 | if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) { | |
b474c7ae | 1707 | error = EINVAL; |
a133d952 DC |
1708 | goto out_unlock; |
1709 | } | |
1710 | ||
1711 | /* Verify both files are either real-time or non-realtime */ | |
1712 | if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) { | |
b474c7ae | 1713 | error = EINVAL; |
a133d952 DC |
1714 | goto out_unlock; |
1715 | } | |
1716 | ||
1717 | error = -filemap_write_and_wait(VFS_I(tip)->i_mapping); | |
1718 | if (error) | |
1719 | goto out_unlock; | |
1720 | truncate_pagecache_range(VFS_I(tip), 0, -1); | |
1721 | ||
1722 | /* Verify O_DIRECT for ftmp */ | |
1723 | if (VN_CACHED(VFS_I(tip)) != 0) { | |
b474c7ae | 1724 | error = EINVAL; |
a133d952 DC |
1725 | goto out_unlock; |
1726 | } | |
1727 | ||
1728 | /* Verify all data are being swapped */ | |
1729 | if (sxp->sx_offset != 0 || | |
1730 | sxp->sx_length != ip->i_d.di_size || | |
1731 | sxp->sx_length != tip->i_d.di_size) { | |
b474c7ae | 1732 | error = EFAULT; |
a133d952 DC |
1733 | goto out_unlock; |
1734 | } | |
1735 | ||
1736 | trace_xfs_swap_extent_before(ip, 0); | |
1737 | trace_xfs_swap_extent_before(tip, 1); | |
1738 | ||
1739 | /* check inode formats now that data is flushed */ | |
1740 | error = xfs_swap_extents_check_format(ip, tip); | |
1741 | if (error) { | |
1742 | xfs_notice(mp, | |
1743 | "%s: inode 0x%llx format is incompatible for exchanging.", | |
1744 | __func__, ip->i_ino); | |
1745 | goto out_unlock; | |
1746 | } | |
1747 | ||
1748 | /* | |
1749 | * Compare the current change & modify times with that | |
1750 | * passed in. If they differ, we abort this swap. | |
1751 | * This is the mechanism used to ensure the calling | |
1752 | * process that the file was not changed out from | |
1753 | * under it. | |
1754 | */ | |
1755 | if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) || | |
1756 | (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) || | |
1757 | (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) || | |
1758 | (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) { | |
b474c7ae | 1759 | error = EBUSY; |
a133d952 DC |
1760 | goto out_unlock; |
1761 | } | |
1762 | ||
1763 | /* We need to fail if the file is memory mapped. Once we have tossed | |
1764 | * all existing pages, the page fault will have no option | |
1765 | * but to go to the filesystem for pages. By making the page fault call | |
1766 | * vop_read (or write in the case of autogrow) they block on the iolock | |
1767 | * until we have switched the extents. | |
1768 | */ | |
1769 | if (VN_MAPPED(VFS_I(ip))) { | |
b474c7ae | 1770 | error = EBUSY; |
a133d952 DC |
1771 | goto out_unlock; |
1772 | } | |
1773 | ||
1774 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1775 | xfs_iunlock(tip, XFS_ILOCK_EXCL); | |
1776 | ||
1777 | /* | |
1778 | * There is a race condition here since we gave up the | |
1779 | * ilock. However, the data fork will not change since | |
1780 | * we have the iolock (locked for truncation too) so we | |
1781 | * are safe. We don't really care if non-io related | |
1782 | * fields change. | |
1783 | */ | |
1784 | truncate_pagecache_range(VFS_I(ip), 0, -1); | |
1785 | ||
1786 | tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT); | |
3d3c8b52 JL |
1787 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0); |
1788 | if (error) { | |
a133d952 DC |
1789 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); |
1790 | xfs_iunlock(tip, XFS_IOLOCK_EXCL); | |
1791 | xfs_trans_cancel(tp, 0); | |
1792 | goto out; | |
1793 | } | |
1794 | xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL); | |
1795 | ||
1796 | /* | |
1797 | * Count the number of extended attribute blocks | |
1798 | */ | |
1799 | if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) && | |
1800 | (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) { | |
1801 | error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks); | |
1802 | if (error) | |
1803 | goto out_trans_cancel; | |
1804 | } | |
1805 | if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) && | |
1806 | (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) { | |
1807 | error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, | |
1808 | &taforkblks); | |
1809 | if (error) | |
1810 | goto out_trans_cancel; | |
1811 | } | |
1812 | ||
21b5c978 DC |
1813 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); |
1814 | xfs_trans_ijoin(tp, tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1815 | ||
1816 | /* | |
1817 | * Before we've swapped the forks, lets set the owners of the forks | |
1818 | * appropriately. We have to do this as we are demand paging the btree | |
1819 | * buffers, and so the validation done on read will expect the owner | |
1820 | * field to be correctly set. Once we change the owners, we can swap the | |
1821 | * inode forks. | |
1822 | * | |
1823 | * Note the trickiness in setting the log flags - we set the owner log | |
1824 | * flag on the opposite inode (i.e. the inode we are setting the new | |
1825 | * owner to be) because once we swap the forks and log that, log | |
1826 | * recovery is going to see the fork as owned by the swapped inode, | |
1827 | * not the pre-swapped inodes. | |
1828 | */ | |
1829 | src_log_flags = XFS_ILOG_CORE; | |
1830 | target_log_flags = XFS_ILOG_CORE; | |
1831 | if (ip->i_d.di_version == 3 && | |
1832 | ip->i_d.di_format == XFS_DINODE_FMT_BTREE) { | |
638f4416 DC |
1833 | target_log_flags |= XFS_ILOG_DOWNER; |
1834 | error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, | |
1835 | tip->i_ino, NULL); | |
21b5c978 DC |
1836 | if (error) |
1837 | goto out_trans_cancel; | |
1838 | } | |
1839 | ||
1840 | if (tip->i_d.di_version == 3 && | |
1841 | tip->i_d.di_format == XFS_DINODE_FMT_BTREE) { | |
638f4416 DC |
1842 | src_log_flags |= XFS_ILOG_DOWNER; |
1843 | error = xfs_bmbt_change_owner(tp, tip, XFS_DATA_FORK, | |
1844 | ip->i_ino, NULL); | |
21b5c978 DC |
1845 | if (error) |
1846 | goto out_trans_cancel; | |
1847 | } | |
1848 | ||
a133d952 DC |
1849 | /* |
1850 | * Swap the data forks of the inodes | |
1851 | */ | |
1852 | ifp = &ip->i_df; | |
1853 | tifp = &tip->i_df; | |
1854 | *tempifp = *ifp; /* struct copy */ | |
1855 | *ifp = *tifp; /* struct copy */ | |
1856 | *tifp = *tempifp; /* struct copy */ | |
1857 | ||
1858 | /* | |
1859 | * Fix the on-disk inode values | |
1860 | */ | |
1861 | tmp = (__uint64_t)ip->i_d.di_nblocks; | |
1862 | ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks; | |
1863 | tip->i_d.di_nblocks = tmp + taforkblks - aforkblks; | |
1864 | ||
1865 | tmp = (__uint64_t) ip->i_d.di_nextents; | |
1866 | ip->i_d.di_nextents = tip->i_d.di_nextents; | |
1867 | tip->i_d.di_nextents = tmp; | |
1868 | ||
1869 | tmp = (__uint64_t) ip->i_d.di_format; | |
1870 | ip->i_d.di_format = tip->i_d.di_format; | |
1871 | tip->i_d.di_format = tmp; | |
1872 | ||
1873 | /* | |
1874 | * The extents in the source inode could still contain speculative | |
1875 | * preallocation beyond EOF (e.g. the file is open but not modified | |
1876 | * while defrag is in progress). In that case, we need to copy over the | |
1877 | * number of delalloc blocks the data fork in the source inode is | |
1878 | * tracking beyond EOF so that when the fork is truncated away when the | |
1879 | * temporary inode is unlinked we don't underrun the i_delayed_blks | |
1880 | * counter on that inode. | |
1881 | */ | |
1882 | ASSERT(tip->i_delayed_blks == 0); | |
1883 | tip->i_delayed_blks = ip->i_delayed_blks; | |
1884 | ip->i_delayed_blks = 0; | |
1885 | ||
a133d952 DC |
1886 | switch (ip->i_d.di_format) { |
1887 | case XFS_DINODE_FMT_EXTENTS: | |
1888 | /* If the extents fit in the inode, fix the | |
1889 | * pointer. Otherwise it's already NULL or | |
1890 | * pointing to the extent. | |
1891 | */ | |
1892 | if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) { | |
1893 | ifp->if_u1.if_extents = | |
1894 | ifp->if_u2.if_inline_ext; | |
1895 | } | |
1896 | src_log_flags |= XFS_ILOG_DEXT; | |
1897 | break; | |
1898 | case XFS_DINODE_FMT_BTREE: | |
21b5c978 | 1899 | ASSERT(ip->i_d.di_version < 3 || |
638f4416 | 1900 | (src_log_flags & XFS_ILOG_DOWNER)); |
a133d952 DC |
1901 | src_log_flags |= XFS_ILOG_DBROOT; |
1902 | break; | |
1903 | } | |
1904 | ||
a133d952 DC |
1905 | switch (tip->i_d.di_format) { |
1906 | case XFS_DINODE_FMT_EXTENTS: | |
1907 | /* If the extents fit in the inode, fix the | |
1908 | * pointer. Otherwise it's already NULL or | |
1909 | * pointing to the extent. | |
1910 | */ | |
1911 | if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) { | |
1912 | tifp->if_u1.if_extents = | |
1913 | tifp->if_u2.if_inline_ext; | |
1914 | } | |
1915 | target_log_flags |= XFS_ILOG_DEXT; | |
1916 | break; | |
1917 | case XFS_DINODE_FMT_BTREE: | |
1918 | target_log_flags |= XFS_ILOG_DBROOT; | |
21b5c978 | 1919 | ASSERT(tip->i_d.di_version < 3 || |
638f4416 | 1920 | (target_log_flags & XFS_ILOG_DOWNER)); |
a133d952 DC |
1921 | break; |
1922 | } | |
1923 | ||
a133d952 DC |
1924 | xfs_trans_log_inode(tp, ip, src_log_flags); |
1925 | xfs_trans_log_inode(tp, tip, target_log_flags); | |
1926 | ||
1927 | /* | |
1928 | * If this is a synchronous mount, make sure that the | |
1929 | * transaction goes to disk before returning to the user. | |
1930 | */ | |
1931 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
1932 | xfs_trans_set_sync(tp); | |
1933 | ||
1934 | error = xfs_trans_commit(tp, 0); | |
1935 | ||
1936 | trace_xfs_swap_extent_after(ip, 0); | |
1937 | trace_xfs_swap_extent_after(tip, 1); | |
1938 | out: | |
1939 | kmem_free(tempifp); | |
1940 | return error; | |
1941 | ||
1942 | out_unlock: | |
1943 | xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1944 | xfs_iunlock(tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1945 | goto out; | |
1946 | ||
1947 | out_trans_cancel: | |
1948 | xfs_trans_cancel(tp, 0); | |
1949 | goto out_unlock; | |
1950 | } |