blob: 0f34886cf7269b1cd67eadbc372655372253981b [file] [log] [blame]
Dave Chinner68988112013-08-12 20:49:42 +10001/*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
Dave Chinnerc24b5df2013-08-12 20:49:45 +10003 * Copyright (c) 2012 Red Hat, Inc.
Dave Chinner68988112013-08-12 20:49:42 +10004 * 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"
Dave Chinner70a9883c2013-10-23 10:36:05 +110021#include "xfs_shared.h"
Dave Chinner239880e2013-10-23 10:50:10 +110022#include "xfs_format.h"
23#include "xfs_log_format.h"
24#include "xfs_trans_resv.h"
Dave Chinner68988112013-08-12 20:49:42 +100025#include "xfs_bit.h"
Dave Chinner68988112013-08-12 20:49:42 +100026#include "xfs_mount.h"
Dave Chinner57062782013-10-15 09:17:51 +110027#include "xfs_da_format.h"
Dave Chinner68988112013-08-12 20:49:42 +100028#include "xfs_inode.h"
29#include "xfs_btree.h"
Dave Chinner239880e2013-10-23 10:50:10 +110030#include "xfs_trans.h"
Dave Chinner68988112013-08-12 20:49:42 +100031#include "xfs_extfree_item.h"
32#include "xfs_alloc.h"
33#include "xfs_bmap.h"
34#include "xfs_bmap_util.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110035#include "xfs_bmap_btree.h"
Dave Chinner68988112013-08-12 20:49:42 +100036#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"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100041#include "xfs_icache.h"
Dave Chinner239880e2013-10-23 10:50:10 +110042#include "xfs_log.h"
Dave Chinner68988112013-08-12 20:49:42 +100043
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 */
51xfs_daddr_t
52xfs_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
59/*
60 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
61 * caller. Frees all the extents that need freeing, which must be done
62 * last due to locking considerations. We never free any extents in
63 * the first transaction.
64 *
65 * Return 1 if the given transaction was committed and a new one
66 * started, and 0 otherwise in the committed parameter.
67 */
68int /* error */
69xfs_bmap_finish(
70 xfs_trans_t **tp, /* transaction pointer addr */
71 xfs_bmap_free_t *flist, /* i/o: list extents to free */
72 int *committed) /* xact committed or not */
73{
74 xfs_efd_log_item_t *efd; /* extent free data */
75 xfs_efi_log_item_t *efi; /* extent free intention */
76 int error; /* error return value */
77 xfs_bmap_free_item_t *free; /* free extent item */
Dave Chinner68988112013-08-12 20:49:42 +100078 xfs_mount_t *mp; /* filesystem mount structure */
79 xfs_bmap_free_item_t *next; /* next item on free list */
Dave Chinner68988112013-08-12 20:49:42 +100080
81 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
82 if (flist->xbf_count == 0) {
83 *committed = 0;
84 return 0;
85 }
Christoph Hellwig2e6db6c2015-06-04 13:47:29 +100086 efi = xfs_trans_get_efi(*tp, flist->xbf_count);
Dave Chinner68988112013-08-12 20:49:42 +100087 for (free = flist->xbf_first; free; free = free->xbfi_next)
Christoph Hellwig2e6db6c2015-06-04 13:47:29 +100088 xfs_trans_log_efi_extent(*tp, efi, free->xbfi_startblock,
Dave Chinner68988112013-08-12 20:49:42 +100089 free->xbfi_blockcount);
Jie Liu3d3c8b52013-08-12 20:49:59 +100090
Christoph Hellwig2e6db6c2015-06-04 13:47:29 +100091 error = xfs_trans_roll(tp, NULL);
Dave Chinner68988112013-08-12 20:49:42 +100092 *committed = 1;
93 /*
94 * We have a new transaction, so we should return committed=1,
95 * even though we're returning an error.
96 */
97 if (error)
98 return error;
99
Christoph Hellwig2e6db6c2015-06-04 13:47:29 +1000100 efd = xfs_trans_get_efd(*tp, efi, flist->xbf_count);
Dave Chinner68988112013-08-12 20:49:42 +1000101 for (free = flist->xbf_first; free != NULL; free = next) {
102 next = free->xbfi_next;
Christoph Hellwig2e6db6c2015-06-04 13:47:29 +1000103 if ((error = xfs_free_extent(*tp, free->xbfi_startblock,
Dave Chinner68988112013-08-12 20:49:42 +1000104 free->xbfi_blockcount))) {
105 /*
106 * The bmap free list will be cleaned up at a
107 * higher level. The EFI will be canceled when
108 * this transaction is aborted.
109 * Need to force shutdown here to make sure it
110 * happens, since this transaction may not be
111 * dirty yet.
112 */
Christoph Hellwig2e6db6c2015-06-04 13:47:29 +1000113 mp = (*tp)->t_mountp;
Dave Chinner68988112013-08-12 20:49:42 +1000114 if (!XFS_FORCED_SHUTDOWN(mp))
115 xfs_force_shutdown(mp,
Dave Chinner24513372014-06-25 14:58:08 +1000116 (error == -EFSCORRUPTED) ?
Dave Chinner68988112013-08-12 20:49:42 +1000117 SHUTDOWN_CORRUPT_INCORE :
118 SHUTDOWN_META_IO_ERROR);
119 return error;
120 }
Christoph Hellwig2e6db6c2015-06-04 13:47:29 +1000121 xfs_trans_log_efd_extent(*tp, efd, free->xbfi_startblock,
Dave Chinner68988112013-08-12 20:49:42 +1000122 free->xbfi_blockcount);
123 xfs_bmap_del_free(flist, NULL, free);
124 }
125 return 0;
126}
127
128int
129xfs_bmap_rtalloc(
130 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
131{
132 xfs_alloctype_t atype = 0; /* type for allocation routines */
133 int error; /* error return value */
134 xfs_mount_t *mp; /* mount point structure */
135 xfs_extlen_t prod = 0; /* product factor for allocators */
136 xfs_extlen_t ralen = 0; /* realtime allocation length */
137 xfs_extlen_t align; /* minimum allocation alignment */
138 xfs_rtblock_t rtb;
139
140 mp = ap->ip->i_mount;
141 align = xfs_get_extsz_hint(ap->ip);
142 prod = align / mp->m_sb.sb_rextsize;
143 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
144 align, 1, ap->eof, 0,
145 ap->conv, &ap->offset, &ap->length);
146 if (error)
147 return error;
148 ASSERT(ap->length);
149 ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
150
151 /*
152 * If the offset & length are not perfectly aligned
153 * then kill prod, it will just get us in trouble.
154 */
155 if (do_mod(ap->offset, align) || ap->length % align)
156 prod = 1;
157 /*
158 * Set ralen to be the actual requested length in rtextents.
159 */
160 ralen = ap->length / mp->m_sb.sb_rextsize;
161 /*
162 * If the old value was close enough to MAXEXTLEN that
163 * we rounded up to it, cut it back so it's valid again.
164 * Note that if it's a really large request (bigger than
165 * MAXEXTLEN), we don't hear about that number, and can't
166 * adjust the starting point to match it.
167 */
168 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
169 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
170
171 /*
172 * Lock out other modifications to the RT bitmap inode.
173 */
174 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
175 xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
176
177 /*
178 * If it's an allocation to an empty file at offset 0,
179 * pick an extent that will space things out in the rt area.
180 */
181 if (ap->eof && ap->offset == 0) {
182 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
183
184 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
185 if (error)
186 return error;
187 ap->blkno = rtx * mp->m_sb.sb_rextsize;
188 } else {
189 ap->blkno = 0;
190 }
191
192 xfs_bmap_adjacent(ap);
193
194 /*
195 * Realtime allocation, done through xfs_rtallocate_extent.
196 */
197 atype = ap->blkno == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
198 do_div(ap->blkno, mp->m_sb.sb_rextsize);
199 rtb = ap->blkno;
200 ap->length = ralen;
201 if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
202 &ralen, atype, ap->wasdel, prod, &rtb)))
203 return error;
204 if (rtb == NULLFSBLOCK && prod > 1 &&
205 (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1,
206 ap->length, &ralen, atype,
207 ap->wasdel, 1, &rtb)))
208 return error;
209 ap->blkno = rtb;
210 if (ap->blkno != NULLFSBLOCK) {
211 ap->blkno *= mp->m_sb.sb_rextsize;
212 ralen *= mp->m_sb.sb_rextsize;
213 ap->length = ralen;
214 ap->ip->i_d.di_nblocks += ralen;
215 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
216 if (ap->wasdel)
217 ap->ip->i_delayed_blks -= ralen;
218 /*
219 * Adjust the disk quota also. This was reserved
220 * earlier.
221 */
222 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
223 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
224 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
225 } else {
226 ap->length = 0;
227 }
228 return 0;
229}
230
231/*
Dave Chinner68988112013-08-12 20:49:42 +1000232 * Check if the endoff is outside the last extent. If so the caller will grow
233 * the allocation to a stripe unit boundary. All offsets are considered outside
234 * the end of file for an empty fork, so 1 is returned in *eof in that case.
235 */
236int
237xfs_bmap_eof(
238 struct xfs_inode *ip,
239 xfs_fileoff_t endoff,
240 int whichfork,
241 int *eof)
242{
243 struct xfs_bmbt_irec rec;
244 int error;
245
246 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
247 if (error || *eof)
248 return error;
249
250 *eof = endoff >= rec.br_startoff + rec.br_blockcount;
251 return 0;
252}
253
254/*
255 * Extent tree block counting routines.
256 */
257
258/*
259 * Count leaf blocks given a range of extent records.
260 */
261STATIC void
262xfs_bmap_count_leaves(
263 xfs_ifork_t *ifp,
264 xfs_extnum_t idx,
265 int numrecs,
266 int *count)
267{
268 int b;
269
270 for (b = 0; b < numrecs; b++) {
271 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
272 *count += xfs_bmbt_get_blockcount(frp);
273 }
274}
275
276/*
277 * Count leaf blocks given a range of extent records originally
278 * in btree format.
279 */
280STATIC void
281xfs_bmap_disk_count_leaves(
282 struct xfs_mount *mp,
283 struct xfs_btree_block *block,
284 int numrecs,
285 int *count)
286{
287 int b;
288 xfs_bmbt_rec_t *frp;
289
290 for (b = 1; b <= numrecs; b++) {
291 frp = XFS_BMBT_REC_ADDR(mp, block, b);
292 *count += xfs_bmbt_disk_get_blockcount(frp);
293 }
294}
295
296/*
297 * Recursively walks each level of a btree
Zhi Yong Wu8be11e92013-08-12 03:14:52 +0000298 * to count total fsblocks in use.
Dave Chinner68988112013-08-12 20:49:42 +1000299 */
300STATIC int /* error */
301xfs_bmap_count_tree(
302 xfs_mount_t *mp, /* file system mount point */
303 xfs_trans_t *tp, /* transaction pointer */
304 xfs_ifork_t *ifp, /* inode fork pointer */
305 xfs_fsblock_t blockno, /* file system block number */
306 int levelin, /* level in btree */
307 int *count) /* Count of blocks */
308{
309 int error;
310 xfs_buf_t *bp, *nbp;
311 int level = levelin;
312 __be64 *pp;
313 xfs_fsblock_t bno = blockno;
314 xfs_fsblock_t nextbno;
315 struct xfs_btree_block *block, *nextblock;
316 int numrecs;
317
318 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF,
319 &xfs_bmbt_buf_ops);
320 if (error)
321 return error;
322 *count += 1;
323 block = XFS_BUF_TO_BLOCK(bp);
324
325 if (--level) {
326 /* Not at node above leaves, count this level of nodes */
327 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
328 while (nextbno != NULLFSBLOCK) {
329 error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp,
330 XFS_BMAP_BTREE_REF,
331 &xfs_bmbt_buf_ops);
332 if (error)
333 return error;
334 *count += 1;
335 nextblock = XFS_BUF_TO_BLOCK(nbp);
336 nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
337 xfs_trans_brelse(tp, nbp);
338 }
339
340 /* Dive to the next level */
341 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
342 bno = be64_to_cpu(*pp);
343 if (unlikely((error =
344 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
345 xfs_trans_brelse(tp, bp);
346 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
347 XFS_ERRLEVEL_LOW, mp);
Dave Chinner24513372014-06-25 14:58:08 +1000348 return -EFSCORRUPTED;
Dave Chinner68988112013-08-12 20:49:42 +1000349 }
350 xfs_trans_brelse(tp, bp);
351 } else {
352 /* count all level 1 nodes and their leaves */
353 for (;;) {
354 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
355 numrecs = be16_to_cpu(block->bb_numrecs);
356 xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
357 xfs_trans_brelse(tp, bp);
358 if (nextbno == NULLFSBLOCK)
359 break;
360 bno = nextbno;
361 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
362 XFS_BMAP_BTREE_REF,
363 &xfs_bmbt_buf_ops);
364 if (error)
365 return error;
366 *count += 1;
367 block = XFS_BUF_TO_BLOCK(bp);
368 }
369 }
370 return 0;
371}
372
373/*
374 * Count fsblocks of the given fork.
375 */
376int /* error */
377xfs_bmap_count_blocks(
378 xfs_trans_t *tp, /* transaction pointer */
379 xfs_inode_t *ip, /* incore inode */
380 int whichfork, /* data or attr fork */
381 int *count) /* out: count of blocks */
382{
383 struct xfs_btree_block *block; /* current btree block */
384 xfs_fsblock_t bno; /* block # of "block" */
385 xfs_ifork_t *ifp; /* fork structure */
386 int level; /* btree level, for checking */
387 xfs_mount_t *mp; /* file system mount structure */
388 __be64 *pp; /* pointer to block address */
389
390 bno = NULLFSBLOCK;
391 mp = ip->i_mount;
392 ifp = XFS_IFORK_PTR(ip, whichfork);
393 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
394 xfs_bmap_count_leaves(ifp, 0,
395 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
396 count);
397 return 0;
398 }
399
400 /*
401 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
402 */
403 block = ifp->if_broot;
404 level = be16_to_cpu(block->bb_level);
405 ASSERT(level > 0);
406 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
407 bno = be64_to_cpu(*pp);
Christoph Hellwigd5cf09b2014-07-30 09:12:05 +1000408 ASSERT(bno != NULLFSBLOCK);
Dave Chinner68988112013-08-12 20:49:42 +1000409 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
410 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
411
412 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
413 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
414 mp);
Dave Chinner24513372014-06-25 14:58:08 +1000415 return -EFSCORRUPTED;
Dave Chinner68988112013-08-12 20:49:42 +1000416 }
417
418 return 0;
419}
420
421/*
422 * returns 1 for success, 0 if we failed to map the extent.
423 */
424STATIC int
425xfs_getbmapx_fix_eof_hole(
426 xfs_inode_t *ip, /* xfs incore inode pointer */
427 struct getbmapx *out, /* output structure */
428 int prealloced, /* this is a file with
429 * preallocated data space */
430 __int64_t end, /* last block requested */
431 xfs_fsblock_t startblock)
432{
433 __int64_t fixlen;
434 xfs_mount_t *mp; /* file system mount point */
435 xfs_ifork_t *ifp; /* inode fork pointer */
436 xfs_extnum_t lastx; /* last extent pointer */
437 xfs_fileoff_t fileblock;
438
439 if (startblock == HOLESTARTBLOCK) {
440 mp = ip->i_mount;
441 out->bmv_block = -1;
442 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
443 fixlen -= out->bmv_offset;
444 if (prealloced && out->bmv_offset + out->bmv_length == end) {
445 /* Came to hole at EOF. Trim it. */
446 if (fixlen <= 0)
447 return 0;
448 out->bmv_length = fixlen;
449 }
450 } else {
451 if (startblock == DELAYSTARTBLOCK)
452 out->bmv_block = -2;
453 else
454 out->bmv_block = xfs_fsb_to_db(ip, startblock);
455 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
456 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
457 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
458 (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
459 out->bmv_oflags |= BMV_OF_LAST;
460 }
461
462 return 1;
463}
464
465/*
466 * Get inode's extents as described in bmv, and format for output.
467 * Calls formatter to fill the user's buffer until all extents
468 * are mapped, until the passed-in bmv->bmv_count slots have
469 * been filled, or until the formatter short-circuits the loop,
470 * if it is tracking filled-in extents on its own.
471 */
472int /* error code */
473xfs_getbmap(
474 xfs_inode_t *ip,
475 struct getbmapx *bmv, /* user bmap structure */
476 xfs_bmap_format_t formatter, /* format to user */
477 void *arg) /* formatter arg */
478{
479 __int64_t bmvend; /* last block requested */
480 int error = 0; /* return value */
481 __int64_t fixlen; /* length for -1 case */
482 int i; /* extent number */
483 int lock; /* lock state */
484 xfs_bmbt_irec_t *map; /* buffer for user's data */
485 xfs_mount_t *mp; /* file system mount point */
486 int nex; /* # of user extents can do */
487 int nexleft; /* # of user extents left */
488 int subnex; /* # of bmapi's can do */
489 int nmap; /* number of map entries */
490 struct getbmapx *out; /* output structure */
491 int whichfork; /* data or attr fork */
492 int prealloced; /* this is a file with
493 * preallocated data space */
494 int iflags; /* interface flags */
495 int bmapi_flags; /* flags for xfs_bmapi */
496 int cur_ext = 0;
497
498 mp = ip->i_mount;
499 iflags = bmv->bmv_iflags;
500 whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
501
502 if (whichfork == XFS_ATTR_FORK) {
503 if (XFS_IFORK_Q(ip)) {
504 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
505 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
506 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
Dave Chinner24513372014-06-25 14:58:08 +1000507 return -EINVAL;
Dave Chinner68988112013-08-12 20:49:42 +1000508 } else if (unlikely(
509 ip->i_d.di_aformat != 0 &&
510 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
511 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
512 ip->i_mount);
Dave Chinner24513372014-06-25 14:58:08 +1000513 return -EFSCORRUPTED;
Dave Chinner68988112013-08-12 20:49:42 +1000514 }
515
516 prealloced = 0;
517 fixlen = 1LL << 32;
518 } else {
519 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
520 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
521 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
Dave Chinner24513372014-06-25 14:58:08 +1000522 return -EINVAL;
Dave Chinner68988112013-08-12 20:49:42 +1000523
524 if (xfs_get_extsz_hint(ip) ||
525 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
526 prealloced = 1;
527 fixlen = mp->m_super->s_maxbytes;
528 } else {
529 prealloced = 0;
530 fixlen = XFS_ISIZE(ip);
531 }
532 }
533
534 if (bmv->bmv_length == -1) {
535 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
536 bmv->bmv_length =
537 max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
538 } else if (bmv->bmv_length == 0) {
539 bmv->bmv_entries = 0;
540 return 0;
541 } else if (bmv->bmv_length < 0) {
Dave Chinner24513372014-06-25 14:58:08 +1000542 return -EINVAL;
Dave Chinner68988112013-08-12 20:49:42 +1000543 }
544
545 nex = bmv->bmv_count - 1;
546 if (nex <= 0)
Dave Chinner24513372014-06-25 14:58:08 +1000547 return -EINVAL;
Dave Chinner68988112013-08-12 20:49:42 +1000548 bmvend = bmv->bmv_offset + bmv->bmv_length;
549
550
551 if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
Dave Chinner24513372014-06-25 14:58:08 +1000552 return -ENOMEM;
Dave Chinnerfdd3cce2013-09-02 20:53:00 +1000553 out = kmem_zalloc_large(bmv->bmv_count * sizeof(struct getbmapx), 0);
554 if (!out)
Dave Chinner24513372014-06-25 14:58:08 +1000555 return -ENOMEM;
Dave Chinner68988112013-08-12 20:49:42 +1000556
557 xfs_ilock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigefa70be2013-12-18 02:14:39 -0800558 if (whichfork == XFS_DATA_FORK) {
559 if (!(iflags & BMV_IF_DELALLOC) &&
560 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) {
Dave Chinner24513372014-06-25 14:58:08 +1000561 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
Dave Chinner68988112013-08-12 20:49:42 +1000562 if (error)
563 goto out_unlock_iolock;
Dave Chinner68988112013-08-12 20:49:42 +1000564
Christoph Hellwigefa70be2013-12-18 02:14:39 -0800565 /*
566 * Even after flushing the inode, there can still be
567 * delalloc blocks on the inode beyond EOF due to
568 * speculative preallocation. These are not removed
569 * until the release function is called or the inode
570 * is inactivated. Hence we cannot assert here that
571 * ip->i_delayed_blks == 0.
572 */
573 }
574
575 lock = xfs_ilock_data_map_shared(ip);
576 } else {
577 lock = xfs_ilock_attr_map_shared(ip);
578 }
Dave Chinner68988112013-08-12 20:49:42 +1000579
580 /*
581 * Don't let nex be bigger than the number of extents
582 * we can have assuming alternating holes and real extents.
583 */
584 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
585 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
586
587 bmapi_flags = xfs_bmapi_aflag(whichfork);
588 if (!(iflags & BMV_IF_PREALLOC))
589 bmapi_flags |= XFS_BMAPI_IGSTATE;
590
591 /*
592 * Allocate enough space to handle "subnex" maps at a time.
593 */
Dave Chinner24513372014-06-25 14:58:08 +1000594 error = -ENOMEM;
Dave Chinner68988112013-08-12 20:49:42 +1000595 subnex = 16;
596 map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
597 if (!map)
598 goto out_unlock_ilock;
599
600 bmv->bmv_entries = 0;
601
602 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
603 (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
604 error = 0;
605 goto out_free_map;
606 }
607
608 nexleft = nex;
609
610 do {
611 nmap = (nexleft > subnex) ? subnex : nexleft;
612 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
613 XFS_BB_TO_FSB(mp, bmv->bmv_length),
614 map, &nmap, bmapi_flags);
615 if (error)
616 goto out_free_map;
617 ASSERT(nmap <= subnex);
618
619 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
620 out[cur_ext].bmv_oflags = 0;
621 if (map[i].br_state == XFS_EXT_UNWRITTEN)
622 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
623 else if (map[i].br_startblock == DELAYSTARTBLOCK)
624 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
625 out[cur_ext].bmv_offset =
626 XFS_FSB_TO_BB(mp, map[i].br_startoff);
627 out[cur_ext].bmv_length =
628 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
629 out[cur_ext].bmv_unused1 = 0;
630 out[cur_ext].bmv_unused2 = 0;
631
632 /*
633 * delayed allocation extents that start beyond EOF can
634 * occur due to speculative EOF allocation when the
635 * delalloc extent is larger than the largest freespace
636 * extent at conversion time. These extents cannot be
637 * converted by data writeback, so can exist here even
638 * if we are not supposed to be finding delalloc
639 * extents.
640 */
641 if (map[i].br_startblock == DELAYSTARTBLOCK &&
642 map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
643 ASSERT((iflags & BMV_IF_DELALLOC) != 0);
644
645 if (map[i].br_startblock == HOLESTARTBLOCK &&
646 whichfork == XFS_ATTR_FORK) {
647 /* came to the end of attribute fork */
648 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
649 goto out_free_map;
650 }
651
652 if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
653 prealloced, bmvend,
654 map[i].br_startblock))
655 goto out_free_map;
656
657 bmv->bmv_offset =
658 out[cur_ext].bmv_offset +
659 out[cur_ext].bmv_length;
660 bmv->bmv_length =
661 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
662
663 /*
664 * In case we don't want to return the hole,
665 * don't increase cur_ext so that we can reuse
666 * it in the next loop.
667 */
668 if ((iflags & BMV_IF_NO_HOLES) &&
669 map[i].br_startblock == HOLESTARTBLOCK) {
670 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
671 continue;
672 }
673
674 nexleft--;
675 bmv->bmv_entries++;
676 cur_ext++;
677 }
678 } while (nmap && nexleft && bmv->bmv_length);
679
680 out_free_map:
681 kmem_free(map);
682 out_unlock_ilock:
Christoph Hellwig01f4f322013-12-06 12:30:08 -0800683 xfs_iunlock(ip, lock);
Dave Chinner68988112013-08-12 20:49:42 +1000684 out_unlock_iolock:
685 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
686
687 for (i = 0; i < cur_ext; i++) {
688 int full = 0; /* user array is full */
689
690 /* format results & advance arg */
691 error = formatter(&arg, &out[i], &full);
692 if (error || full)
693 break;
694 }
695
Dave Chinnerfdd3cce2013-09-02 20:53:00 +1000696 kmem_free(out);
Dave Chinner68988112013-08-12 20:49:42 +1000697 return error;
698}
699
700/*
701 * dead simple method of punching delalyed allocation blocks from a range in
702 * the inode. Walks a block at a time so will be slow, but is only executed in
Zhi Yong Wuad4809b2013-08-12 03:14:55 +0000703 * rare error cases so the overhead is not critical. This will always punch out
Dave Chinner68988112013-08-12 20:49:42 +1000704 * both the start and end blocks, even if the ranges only partially overlap
705 * them, so it is up to the caller to ensure that partial blocks are not
706 * passed in.
707 */
708int
709xfs_bmap_punch_delalloc_range(
710 struct xfs_inode *ip,
711 xfs_fileoff_t start_fsb,
712 xfs_fileoff_t length)
713{
714 xfs_fileoff_t remaining = length;
715 int error = 0;
716
717 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
718
719 do {
720 int done;
721 xfs_bmbt_irec_t imap;
722 int nimaps = 1;
723 xfs_fsblock_t firstblock;
724 xfs_bmap_free_t flist;
725
726 /*
727 * Map the range first and check that it is a delalloc extent
728 * before trying to unmap the range. Otherwise we will be
729 * trying to remove a real extent (which requires a
730 * transaction) or a hole, which is probably a bad idea...
731 */
732 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
733 XFS_BMAPI_ENTIRE);
734
735 if (error) {
736 /* something screwed, just bail */
737 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
738 xfs_alert(ip->i_mount,
739 "Failed delalloc mapping lookup ino %lld fsb %lld.",
740 ip->i_ino, start_fsb);
741 }
742 break;
743 }
744 if (!nimaps) {
745 /* nothing there */
746 goto next_block;
747 }
748 if (imap.br_startblock != DELAYSTARTBLOCK) {
749 /* been converted, ignore */
750 goto next_block;
751 }
752 WARN_ON(imap.br_blockcount == 0);
753
754 /*
755 * Note: while we initialise the firstblock/flist pair, they
756 * should never be used because blocks should never be
757 * allocated or freed for a delalloc extent and hence we need
758 * don't cancel or finish them after the xfs_bunmapi() call.
759 */
760 xfs_bmap_init(&flist, &firstblock);
761 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
762 &flist, &done);
763 if (error)
764 break;
765
766 ASSERT(!flist.xbf_count && !flist.xbf_first);
767next_block:
768 start_fsb++;
769 remaining--;
770 } while(remaining > 0);
771
772 return error;
773}
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000774
775/*
776 * Test whether it is appropriate to check an inode for and free post EOF
777 * blocks. The 'force' parameter determines whether we should also consider
778 * regular files that are marked preallocated or append-only.
779 */
780bool
781xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
782{
783 /* prealloc/delalloc exists only on regular files */
784 if (!S_ISREG(ip->i_d.di_mode))
785 return false;
786
787 /*
788 * Zero sized files with no cached pages and delalloc blocks will not
789 * have speculative prealloc/delalloc blocks to remove.
790 */
791 if (VFS_I(ip)->i_size == 0 &&
Dave Chinner2667c6f2014-08-04 13:23:15 +1000792 VFS_I(ip)->i_mapping->nrpages == 0 &&
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000793 ip->i_delayed_blks == 0)
794 return false;
795
796 /* If we haven't read in the extent list, then don't do it now. */
797 if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
798 return false;
799
800 /*
801 * Do not free real preallocated or append-only files unless the file
802 * has delalloc blocks and we are forced to remove them.
803 */
804 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
805 if (!force || ip->i_delayed_blks == 0)
806 return false;
807
808 return true;
809}
810
811/*
812 * This is called by xfs_inactive to free any blocks beyond eof
813 * when the link count isn't zero and by xfs_dm_punch_hole() when
814 * punching a hole to EOF.
815 */
816int
817xfs_free_eofblocks(
818 xfs_mount_t *mp,
819 xfs_inode_t *ip,
820 bool need_iolock)
821{
822 xfs_trans_t *tp;
823 int error;
824 xfs_fileoff_t end_fsb;
825 xfs_fileoff_t last_fsb;
826 xfs_filblks_t map_len;
827 int nimaps;
828 xfs_bmbt_irec_t imap;
829
830 /*
831 * Figure out if there are any blocks beyond the end
832 * of the file. If not, then there is nothing to do.
833 */
834 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
835 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
836 if (last_fsb <= end_fsb)
837 return 0;
838 map_len = last_fsb - end_fsb;
839
840 nimaps = 1;
841 xfs_ilock(ip, XFS_ILOCK_SHARED);
842 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
843 xfs_iunlock(ip, XFS_ILOCK_SHARED);
844
845 if (!error && (nimaps != 0) &&
846 (imap.br_startblock != HOLESTARTBLOCK ||
847 ip->i_delayed_blks)) {
848 /*
849 * Attach the dquots to the inode up front.
850 */
851 error = xfs_qm_dqattach(ip, 0);
852 if (error)
853 return error;
854
855 /*
856 * There are blocks after the end of file.
857 * Free them up now by truncating the file to
858 * its current size.
859 */
860 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
861
862 if (need_iolock) {
863 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
Christoph Hellwig4906e212015-06-04 13:47:56 +1000864 xfs_trans_cancel(tp);
Dave Chinner24513372014-06-25 14:58:08 +1000865 return -EAGAIN;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000866 }
867 }
868
Jie Liu3d3c8b52013-08-12 20:49:59 +1000869 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000870 if (error) {
871 ASSERT(XFS_FORCED_SHUTDOWN(mp));
Christoph Hellwig4906e212015-06-04 13:47:56 +1000872 xfs_trans_cancel(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000873 if (need_iolock)
874 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
875 return error;
876 }
877
878 xfs_ilock(ip, XFS_ILOCK_EXCL);
879 xfs_trans_ijoin(tp, ip, 0);
880
881 /*
882 * Do not update the on-disk file size. If we update the
883 * on-disk file size and then the system crashes before the
884 * contents of the file are flushed to disk then the files
885 * may be full of holes (ie NULL files bug).
886 */
887 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
888 XFS_ISIZE(ip));
889 if (error) {
890 /*
891 * If we get an error at this point we simply don't
892 * bother truncating the file.
893 */
Christoph Hellwig4906e212015-06-04 13:47:56 +1000894 xfs_trans_cancel(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000895 } else {
Christoph Hellwig70393312015-06-04 13:48:08 +1000896 error = xfs_trans_commit(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000897 if (!error)
898 xfs_inode_clear_eofblocks_tag(ip);
899 }
900
901 xfs_iunlock(ip, XFS_ILOCK_EXCL);
902 if (need_iolock)
903 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
904 }
905 return error;
906}
907
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700908int
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000909xfs_alloc_file_space(
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700910 struct xfs_inode *ip,
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000911 xfs_off_t offset,
912 xfs_off_t len,
Christoph Hellwig5f8aca82013-10-12 00:55:06 -0700913 int alloc_type)
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000914{
915 xfs_mount_t *mp = ip->i_mount;
916 xfs_off_t count;
917 xfs_filblks_t allocated_fsb;
918 xfs_filblks_t allocatesize_fsb;
919 xfs_extlen_t extsz, temp;
920 xfs_fileoff_t startoffset_fsb;
921 xfs_fsblock_t firstfsb;
922 int nimaps;
923 int quota_flag;
924 int rt;
925 xfs_trans_t *tp;
926 xfs_bmbt_irec_t imaps[1], *imapp;
927 xfs_bmap_free_t free_list;
928 uint qblocks, resblks, resrtextents;
929 int committed;
930 int error;
931
932 trace_xfs_alloc_file_space(ip);
933
934 if (XFS_FORCED_SHUTDOWN(mp))
Dave Chinner24513372014-06-25 14:58:08 +1000935 return -EIO;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000936
937 error = xfs_qm_dqattach(ip, 0);
938 if (error)
939 return error;
940
941 if (len <= 0)
Dave Chinner24513372014-06-25 14:58:08 +1000942 return -EINVAL;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000943
944 rt = XFS_IS_REALTIME_INODE(ip);
945 extsz = xfs_get_extsz_hint(ip);
946
947 count = len;
948 imapp = &imaps[0];
949 nimaps = 1;
950 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
951 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
952
953 /*
954 * Allocate file space until done or until there is an error
955 */
956 while (allocatesize_fsb && !error) {
957 xfs_fileoff_t s, e;
958
959 /*
960 * Determine space reservations for data/realtime.
961 */
962 if (unlikely(extsz)) {
963 s = startoffset_fsb;
964 do_div(s, extsz);
965 s *= extsz;
966 e = startoffset_fsb + allocatesize_fsb;
967 if ((temp = do_mod(startoffset_fsb, extsz)))
968 e += temp;
969 if ((temp = do_mod(e, extsz)))
970 e += extsz - temp;
971 } else {
972 s = 0;
973 e = allocatesize_fsb;
974 }
975
976 /*
977 * The transaction reservation is limited to a 32-bit block
978 * count, hence we need to limit the number of blocks we are
979 * trying to reserve to avoid an overflow. We can't allocate
980 * more than @nimaps extents, and an extent is limited on disk
981 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
982 */
983 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
984 if (unlikely(rt)) {
985 resrtextents = qblocks = resblks;
986 resrtextents /= mp->m_sb.sb_rextsize;
987 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
988 quota_flag = XFS_QMOPT_RES_RTBLKS;
989 } else {
990 resrtextents = 0;
991 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
992 quota_flag = XFS_QMOPT_RES_REGBLKS;
993 }
994
995 /*
996 * Allocate and setup the transaction.
997 */
998 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
Jie Liu3d3c8b52013-08-12 20:49:59 +1000999 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
1000 resblks, resrtextents);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001001 /*
1002 * Check for running out of space
1003 */
1004 if (error) {
1005 /*
1006 * Free the transaction structure.
1007 */
Dave Chinner24513372014-06-25 14:58:08 +10001008 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
Christoph Hellwig4906e212015-06-04 13:47:56 +10001009 xfs_trans_cancel(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001010 break;
1011 }
1012 xfs_ilock(ip, XFS_ILOCK_EXCL);
1013 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1014 0, quota_flag);
1015 if (error)
1016 goto error1;
1017
1018 xfs_trans_ijoin(tp, ip, 0);
1019
1020 xfs_bmap_init(&free_list, &firstfsb);
1021 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1022 allocatesize_fsb, alloc_type, &firstfsb,
1023 0, imapp, &nimaps, &free_list);
1024 if (error) {
1025 goto error0;
1026 }
1027
1028 /*
1029 * Complete the transaction
1030 */
1031 error = xfs_bmap_finish(&tp, &free_list, &committed);
1032 if (error) {
1033 goto error0;
1034 }
1035
Christoph Hellwig70393312015-06-04 13:48:08 +10001036 error = xfs_trans_commit(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001037 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1038 if (error) {
1039 break;
1040 }
1041
1042 allocated_fsb = imapp->br_blockcount;
1043
1044 if (nimaps == 0) {
Dave Chinner24513372014-06-25 14:58:08 +10001045 error = -ENOSPC;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001046 break;
1047 }
1048
1049 startoffset_fsb += allocated_fsb;
1050 allocatesize_fsb -= allocated_fsb;
1051 }
1052
1053 return error;
1054
1055error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1056 xfs_bmap_cancel(&free_list);
1057 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
1058
1059error1: /* Just cancel transaction */
Christoph Hellwig4906e212015-06-04 13:47:56 +10001060 xfs_trans_cancel(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001061 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1062 return error;
1063}
1064
1065/*
1066 * Zero file bytes between startoff and endoff inclusive.
1067 * The iolock is held exclusive and no blocks are buffered.
1068 *
1069 * This function is used by xfs_free_file_space() to zero
1070 * partial blocks when the range to free is not block aligned.
1071 * When unreserving space with boundaries that are not block
1072 * aligned we round up the start and round down the end
1073 * boundaries and then use this function to zero the parts of
1074 * the blocks that got dropped during the rounding.
1075 */
1076STATIC int
1077xfs_zero_remaining_bytes(
1078 xfs_inode_t *ip,
1079 xfs_off_t startoff,
1080 xfs_off_t endoff)
1081{
1082 xfs_bmbt_irec_t imap;
1083 xfs_fileoff_t offset_fsb;
1084 xfs_off_t lastoffset;
1085 xfs_off_t offset;
1086 xfs_buf_t *bp;
1087 xfs_mount_t *mp = ip->i_mount;
1088 int nimap;
1089 int error = 0;
1090
1091 /*
1092 * Avoid doing I/O beyond eof - it's not necessary
1093 * since nothing can read beyond eof. The space will
1094 * be zeroed when the file is extended anyway.
1095 */
1096 if (startoff >= XFS_ISIZE(ip))
1097 return 0;
1098
1099 if (endoff > XFS_ISIZE(ip))
1100 endoff = XFS_ISIZE(ip);
1101
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001102 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
Christoph Hellwig4f317362013-12-06 12:30:12 -08001103 uint lock_mode;
1104
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001105 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1106 nimap = 1;
Christoph Hellwig4f317362013-12-06 12:30:12 -08001107
1108 lock_mode = xfs_ilock_data_map_shared(ip);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001109 error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0);
Christoph Hellwig4f317362013-12-06 12:30:12 -08001110 xfs_iunlock(ip, lock_mode);
1111
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001112 if (error || nimap < 1)
1113 break;
1114 ASSERT(imap.br_blockcount >= 1);
1115 ASSERT(imap.br_startoff == offset_fsb);
Dave Chinner4f69f572015-06-04 09:19:08 +10001116 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1117
1118 if (imap.br_startblock == HOLESTARTBLOCK ||
1119 imap.br_state == XFS_EXT_UNWRITTEN) {
1120 /* skip the entire extent */
1121 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff +
1122 imap.br_blockcount) - 1;
1123 continue;
1124 }
1125
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001126 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
1127 if (lastoffset > endoff)
1128 lastoffset = endoff;
Dave Chinner4f69f572015-06-04 09:19:08 +10001129
1130 /* DAX can just zero the backing device directly */
1131 if (IS_DAX(VFS_I(ip))) {
1132 error = dax_zero_page_range(VFS_I(ip), offset,
1133 lastoffset - offset + 1,
1134 xfs_get_blocks_direct);
1135 if (error)
1136 return error;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001137 continue;
Dave Chinner4f69f572015-06-04 09:19:08 +10001138 }
Christoph Hellwig83a0adc2013-12-17 00:03:52 -08001139
Christoph Hellwig8c156122014-10-02 09:05:44 +10001140 error = xfs_buf_read_uncached(XFS_IS_REALTIME_INODE(ip) ?
1141 mp->m_rtdev_targp : mp->m_ddev_targp,
1142 xfs_fsb_to_db(ip, imap.br_startblock),
1143 BTOBB(mp->m_sb.sb_blocksize),
1144 0, &bp, NULL);
1145 if (error)
1146 return error;
1147
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001148 memset(bp->b_addr +
Christoph Hellwig8c156122014-10-02 09:05:44 +10001149 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
1150 0, lastoffset - offset + 1);
Christoph Hellwig83a0adc2013-12-17 00:03:52 -08001151
Christoph Hellwig8c156122014-10-02 09:05:44 +10001152 error = xfs_bwrite(bp);
1153 xfs_buf_relse(bp);
1154 if (error)
1155 return error;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001156 }
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001157 return error;
1158}
1159
Christoph Hellwig83aee9e2013-10-12 00:55:07 -07001160int
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001161xfs_free_file_space(
Christoph Hellwig83aee9e2013-10-12 00:55:07 -07001162 struct xfs_inode *ip,
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001163 xfs_off_t offset,
Christoph Hellwig5f8aca82013-10-12 00:55:06 -07001164 xfs_off_t len)
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001165{
1166 int committed;
1167 int done;
1168 xfs_fileoff_t endoffset_fsb;
1169 int error;
1170 xfs_fsblock_t firstfsb;
1171 xfs_bmap_free_t free_list;
1172 xfs_bmbt_irec_t imap;
1173 xfs_off_t ioffset;
Brian Foster8b5279e2014-09-23 15:39:05 +10001174 xfs_off_t iendoffset;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001175 xfs_extlen_t mod=0;
1176 xfs_mount_t *mp;
1177 int nimap;
1178 uint resblks;
1179 xfs_off_t rounding;
1180 int rt;
1181 xfs_fileoff_t startoffset_fsb;
1182 xfs_trans_t *tp;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001183
1184 mp = ip->i_mount;
1185
1186 trace_xfs_free_file_space(ip);
1187
1188 error = xfs_qm_dqattach(ip, 0);
1189 if (error)
1190 return error;
1191
1192 error = 0;
1193 if (len <= 0) /* if nothing being freed */
1194 return error;
1195 rt = XFS_IS_REALTIME_INODE(ip);
1196 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1197 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1198
Christoph Hellwig5f8aca82013-10-12 00:55:06 -07001199 /* wait for the completion of any pending DIOs */
1200 inode_dio_wait(VFS_I(ip));
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001201
1202 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
Brian Foster8b5279e2014-09-23 15:39:05 +10001203 ioffset = round_down(offset, rounding);
1204 iendoffset = round_up(offset + len, rounding) - 1;
1205 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, ioffset,
1206 iendoffset);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001207 if (error)
Christoph Hellwig5f8aca82013-10-12 00:55:06 -07001208 goto out;
Brian Foster8b5279e2014-09-23 15:39:05 +10001209 truncate_pagecache_range(VFS_I(ip), ioffset, iendoffset);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001210
1211 /*
1212 * Need to zero the stuff we're not freeing, on disk.
1213 * If it's a realtime file & can't use unwritten extents then we
1214 * actually need to zero the extent edges. Otherwise xfs_bunmapi
1215 * will take care of it for us.
1216 */
1217 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
1218 nimap = 1;
1219 error = xfs_bmapi_read(ip, startoffset_fsb, 1,
1220 &imap, &nimap, 0);
1221 if (error)
Christoph Hellwig5f8aca82013-10-12 00:55:06 -07001222 goto out;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001223 ASSERT(nimap == 0 || nimap == 1);
1224 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1225 xfs_daddr_t block;
1226
1227 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1228 block = imap.br_startblock;
1229 mod = do_div(block, mp->m_sb.sb_rextsize);
1230 if (mod)
1231 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
1232 }
1233 nimap = 1;
1234 error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1,
1235 &imap, &nimap, 0);
1236 if (error)
Christoph Hellwig5f8aca82013-10-12 00:55:06 -07001237 goto out;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001238 ASSERT(nimap == 0 || nimap == 1);
1239 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1240 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1241 mod++;
1242 if (mod && (mod != mp->m_sb.sb_rextsize))
1243 endoffset_fsb -= mod;
1244 }
1245 }
1246 if ((done = (endoffset_fsb <= startoffset_fsb)))
1247 /*
1248 * One contiguous piece to clear
1249 */
1250 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
1251 else {
1252 /*
1253 * Some full blocks, possibly two pieces to clear
1254 */
1255 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
1256 error = xfs_zero_remaining_bytes(ip, offset,
1257 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
1258 if (!error &&
1259 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
1260 error = xfs_zero_remaining_bytes(ip,
1261 XFS_FSB_TO_B(mp, endoffset_fsb),
1262 offset + len - 1);
1263 }
1264
1265 /*
1266 * free file space until done or until there is an error
1267 */
1268 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1269 while (!error && !done) {
1270
1271 /*
1272 * allocate and setup the transaction. Allow this
1273 * transaction to dip into the reserve blocks to ensure
1274 * the freeing of the space succeeds at ENOSPC.
1275 */
1276 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
Jie Liu3d3c8b52013-08-12 20:49:59 +10001277 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, resblks, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001278
1279 /*
1280 * check for running out of space
1281 */
1282 if (error) {
1283 /*
1284 * Free the transaction structure.
1285 */
Dave Chinner24513372014-06-25 14:58:08 +10001286 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
Christoph Hellwig4906e212015-06-04 13:47:56 +10001287 xfs_trans_cancel(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001288 break;
1289 }
1290 xfs_ilock(ip, XFS_ILOCK_EXCL);
1291 error = xfs_trans_reserve_quota(tp, mp,
1292 ip->i_udquot, ip->i_gdquot, ip->i_pdquot,
1293 resblks, 0, XFS_QMOPT_RES_REGBLKS);
1294 if (error)
1295 goto error1;
1296
1297 xfs_trans_ijoin(tp, ip, 0);
1298
1299 /*
1300 * issue the bunmapi() call to free the blocks
1301 */
1302 xfs_bmap_init(&free_list, &firstfsb);
1303 error = xfs_bunmapi(tp, ip, startoffset_fsb,
1304 endoffset_fsb - startoffset_fsb,
1305 0, 2, &firstfsb, &free_list, &done);
1306 if (error) {
1307 goto error0;
1308 }
1309
1310 /*
1311 * complete the transaction
1312 */
1313 error = xfs_bmap_finish(&tp, &free_list, &committed);
1314 if (error) {
1315 goto error0;
1316 }
1317
Christoph Hellwig70393312015-06-04 13:48:08 +10001318 error = xfs_trans_commit(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001319 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1320 }
1321
Christoph Hellwig5f8aca82013-10-12 00:55:06 -07001322 out:
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001323 return error;
1324
1325 error0:
1326 xfs_bmap_cancel(&free_list);
1327 error1:
Christoph Hellwig4906e212015-06-04 13:47:56 +10001328 xfs_trans_cancel(tp);
Christoph Hellwig5f8aca82013-10-12 00:55:06 -07001329 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1330 goto out;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001331}
1332
Brian Foster5d11fb42014-10-30 10:35:11 +11001333/*
1334 * Preallocate and zero a range of a file. This mechanism has the allocation
1335 * semantics of fallocate and in addition converts data in the range to zeroes.
1336 */
Christoph Hellwig865e9442013-10-12 00:55:08 -07001337int
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001338xfs_zero_file_space(
1339 struct xfs_inode *ip,
1340 xfs_off_t offset,
Christoph Hellwig5f8aca82013-10-12 00:55:06 -07001341 xfs_off_t len)
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001342{
1343 struct xfs_mount *mp = ip->i_mount;
Brian Foster5d11fb42014-10-30 10:35:11 +11001344 uint blksize;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001345 int error;
1346
Dave Chinner897b73b2014-04-14 18:15:11 +10001347 trace_xfs_zero_file_space(ip);
1348
Brian Foster5d11fb42014-10-30 10:35:11 +11001349 blksize = 1 << mp->m_sb.sb_blocklog;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001350
1351 /*
Brian Foster5d11fb42014-10-30 10:35:11 +11001352 * Punch a hole and prealloc the range. We use hole punch rather than
1353 * unwritten extent conversion for two reasons:
1354 *
1355 * 1.) Hole punch handles partial block zeroing for us.
1356 *
1357 * 2.) If prealloc returns ENOSPC, the file range is still zero-valued
1358 * by virtue of the hole punch.
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001359 */
Brian Foster5d11fb42014-10-30 10:35:11 +11001360 error = xfs_free_file_space(ip, offset, len);
1361 if (error)
1362 goto out;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001363
Brian Foster5d11fb42014-10-30 10:35:11 +11001364 error = xfs_alloc_file_space(ip, round_down(offset, blksize),
1365 round_up(offset + len, blksize) -
1366 round_down(offset, blksize),
1367 XFS_BMAPI_PREALLOC);
Christoph Hellwig5f8aca82013-10-12 00:55:06 -07001368out:
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001369 return error;
1370
1371}
1372
1373/*
Namjae Jeona904b1c2015-03-25 15:08:56 +11001374 * @next_fsb will keep track of the extent currently undergoing shift.
1375 * @stop_fsb will keep track of the extent at which we have to stop.
1376 * If we are shifting left, we will start with block (offset + len) and
1377 * shift each extent till last extent.
1378 * If we are shifting right, we will start with last extent inside file space
1379 * and continue until we reach the block corresponding to offset.
Namjae Jeone1d8fb82014-02-24 10:58:19 +11001380 */
kbuild test robot72c1a732015-04-13 11:25:04 +10001381static int
Namjae Jeona904b1c2015-03-25 15:08:56 +11001382xfs_shift_file_space(
1383 struct xfs_inode *ip,
1384 xfs_off_t offset,
1385 xfs_off_t len,
1386 enum shift_direction direction)
Namjae Jeone1d8fb82014-02-24 10:58:19 +11001387{
1388 int done = 0;
1389 struct xfs_mount *mp = ip->i_mount;
1390 struct xfs_trans *tp;
1391 int error;
Namjae Jeone1d8fb82014-02-24 10:58:19 +11001392 struct xfs_bmap_free free_list;
1393 xfs_fsblock_t first_block;
1394 int committed;
Namjae Jeona904b1c2015-03-25 15:08:56 +11001395 xfs_fileoff_t stop_fsb;
Brian Foster2c845f52014-09-23 15:37:09 +10001396 xfs_fileoff_t next_fsb;
Namjae Jeone1d8fb82014-02-24 10:58:19 +11001397 xfs_fileoff_t shift_fsb;
1398
Namjae Jeona904b1c2015-03-25 15:08:56 +11001399 ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
Namjae Jeone1d8fb82014-02-24 10:58:19 +11001400
Namjae Jeona904b1c2015-03-25 15:08:56 +11001401 if (direction == SHIFT_LEFT) {
1402 next_fsb = XFS_B_TO_FSB(mp, offset + len);
1403 stop_fsb = XFS_B_TO_FSB(mp, VFS_I(ip)->i_size);
1404 } else {
1405 /*
1406 * If right shift, delegate the work of initialization of
1407 * next_fsb to xfs_bmap_shift_extent as it has ilock held.
1408 */
1409 next_fsb = NULLFSBLOCK;
1410 stop_fsb = XFS_B_TO_FSB(mp, offset);
1411 }
Namjae Jeone1d8fb82014-02-24 10:58:19 +11001412
Namjae Jeone1d8fb82014-02-24 10:58:19 +11001413 shift_fsb = XFS_B_TO_FSB(mp, len);
1414
Brian Fosterf71721d2014-09-23 15:39:05 +10001415 /*
1416 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1417 * into the accessible region of the file.
1418 */
Brian Foster41b9d722014-09-02 12:12:53 +10001419 if (xfs_can_free_eofblocks(ip, true)) {
1420 error = xfs_free_eofblocks(mp, ip, false);
1421 if (error)
1422 return error;
1423 }
Dave Chinner1669a8c2014-09-02 12:12:53 +10001424
Brian Fosterf71721d2014-09-23 15:39:05 +10001425 /*
1426 * Writeback and invalidate cache for the remainder of the file as we're
Namjae Jeona904b1c2015-03-25 15:08:56 +11001427 * about to shift down every extent from offset to EOF.
Brian Fosterf71721d2014-09-23 15:39:05 +10001428 */
1429 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
Namjae Jeona904b1c2015-03-25 15:08:56 +11001430 offset, -1);
Brian Fosterf71721d2014-09-23 15:39:05 +10001431 if (error)
1432 return error;
1433 error = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
Namjae Jeona904b1c2015-03-25 15:08:56 +11001434 offset >> PAGE_CACHE_SHIFT, -1);
Namjae Jeone1d8fb82014-02-24 10:58:19 +11001435 if (error)
1436 return error;
1437
Namjae Jeona904b1c2015-03-25 15:08:56 +11001438 /*
1439 * The extent shiting code works on extent granularity. So, if
1440 * stop_fsb is not the starting block of extent, we need to split
1441 * the extent at stop_fsb.
1442 */
1443 if (direction == SHIFT_RIGHT) {
1444 error = xfs_bmap_split_extent(ip, stop_fsb);
1445 if (error)
1446 return error;
1447 }
1448
Namjae Jeone1d8fb82014-02-24 10:58:19 +11001449 while (!error && !done) {
1450 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
Namjae Jeone1d8fb82014-02-24 10:58:19 +11001451 /*
1452 * We would need to reserve permanent block for transaction.
1453 * This will come into picture when after shifting extent into
1454 * hole we found that adjacent extents can be merged which
1455 * may lead to freeing of a block during record update.
1456 */
1457 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
1458 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0);
1459 if (error) {
Christoph Hellwig4906e212015-06-04 13:47:56 +10001460 xfs_trans_cancel(tp);
Namjae Jeone1d8fb82014-02-24 10:58:19 +11001461 break;
1462 }
1463
1464 xfs_ilock(ip, XFS_ILOCK_EXCL);
1465 error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot,
1466 ip->i_gdquot, ip->i_pdquot,
1467 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0,
1468 XFS_QMOPT_RES_REGBLKS);
1469 if (error)
1470 goto out;
1471
Namjae Jeona904b1c2015-03-25 15:08:56 +11001472 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
Namjae Jeone1d8fb82014-02-24 10:58:19 +11001473
1474 xfs_bmap_init(&free_list, &first_block);
1475
1476 /*
1477 * We are using the write transaction in which max 2 bmbt
1478 * updates are allowed
1479 */
Namjae Jeona904b1c2015-03-25 15:08:56 +11001480 error = xfs_bmap_shift_extents(tp, ip, &next_fsb, shift_fsb,
1481 &done, stop_fsb, &first_block, &free_list,
1482 direction, XFS_BMAP_MAX_SHIFT_EXTENTS);
Namjae Jeone1d8fb82014-02-24 10:58:19 +11001483 if (error)
1484 goto out;
1485
1486 error = xfs_bmap_finish(&tp, &free_list, &committed);
1487 if (error)
1488 goto out;
1489
Christoph Hellwig70393312015-06-04 13:48:08 +10001490 error = xfs_trans_commit(tp);
Namjae Jeone1d8fb82014-02-24 10:58:19 +11001491 }
1492
1493 return error;
1494
1495out:
Christoph Hellwig4906e212015-06-04 13:47:56 +10001496 xfs_trans_cancel(tp);
Namjae Jeone1d8fb82014-02-24 10:58:19 +11001497 return error;
1498}
1499
1500/*
Namjae Jeona904b1c2015-03-25 15:08:56 +11001501 * xfs_collapse_file_space()
1502 * This routine frees disk space and shift extent for the given file.
1503 * The first thing we do is to free data blocks in the specified range
1504 * by calling xfs_free_file_space(). It would also sync dirty data
1505 * and invalidate page cache over the region on which collapse range
1506 * is working. And Shift extent records to the left to cover a hole.
1507 * RETURNS:
1508 * 0 on success
1509 * errno on error
1510 *
1511 */
1512int
1513xfs_collapse_file_space(
1514 struct xfs_inode *ip,
1515 xfs_off_t offset,
1516 xfs_off_t len)
1517{
1518 int error;
1519
1520 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1521 trace_xfs_collapse_file_space(ip);
1522
1523 error = xfs_free_file_space(ip, offset, len);
1524 if (error)
1525 return error;
1526
1527 return xfs_shift_file_space(ip, offset, len, SHIFT_LEFT);
1528}
1529
1530/*
1531 * xfs_insert_file_space()
1532 * This routine create hole space by shifting extents for the given file.
1533 * The first thing we do is to sync dirty data and invalidate page cache
1534 * over the region on which insert range is working. And split an extent
1535 * to two extents at given offset by calling xfs_bmap_split_extent.
1536 * And shift all extent records which are laying between [offset,
1537 * last allocated extent] to the right to reserve hole range.
1538 * RETURNS:
1539 * 0 on success
1540 * errno on error
1541 */
1542int
1543xfs_insert_file_space(
1544 struct xfs_inode *ip,
1545 loff_t offset,
1546 loff_t len)
1547{
1548 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1549 trace_xfs_insert_file_space(ip);
1550
1551 return xfs_shift_file_space(ip, offset, len, SHIFT_RIGHT);
1552}
1553
1554/*
Dave Chinnera133d952013-08-12 20:49:48 +10001555 * We need to check that the format of the data fork in the temporary inode is
1556 * valid for the target inode before doing the swap. This is not a problem with
1557 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1558 * data fork depending on the space the attribute fork is taking so we can get
1559 * invalid formats on the target inode.
1560 *
1561 * E.g. target has space for 7 extents in extent format, temp inode only has
1562 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1563 * btree, but when swapped it needs to be in extent format. Hence we can't just
1564 * blindly swap data forks on attr2 filesystems.
1565 *
1566 * Note that we check the swap in both directions so that we don't end up with
1567 * a corrupt temporary inode, either.
1568 *
1569 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1570 * inode will prevent this situation from occurring, so all we do here is
1571 * reject and log the attempt. basically we are putting the responsibility on
1572 * userspace to get this right.
1573 */
1574static int
1575xfs_swap_extents_check_format(
1576 xfs_inode_t *ip, /* target inode */
1577 xfs_inode_t *tip) /* tmp inode */
1578{
1579
1580 /* Should never get a local format */
1581 if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
1582 tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
Dave Chinner24513372014-06-25 14:58:08 +10001583 return -EINVAL;
Dave Chinnera133d952013-08-12 20:49:48 +10001584
1585 /*
1586 * if the target inode has less extents that then temporary inode then
1587 * why did userspace call us?
1588 */
1589 if (ip->i_d.di_nextents < tip->i_d.di_nextents)
Dave Chinner24513372014-06-25 14:58:08 +10001590 return -EINVAL;
Dave Chinnera133d952013-08-12 20:49:48 +10001591
1592 /*
1593 * if the target inode is in extent form and the temp inode is in btree
1594 * form then we will end up with the target inode in the wrong format
1595 * as we already know there are less extents in the temp inode.
1596 */
1597 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1598 tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
Dave Chinner24513372014-06-25 14:58:08 +10001599 return -EINVAL;
Dave Chinnera133d952013-08-12 20:49:48 +10001600
1601 /* Check temp in extent form to max in target */
1602 if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1603 XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
1604 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
Dave Chinner24513372014-06-25 14:58:08 +10001605 return -EINVAL;
Dave Chinnera133d952013-08-12 20:49:48 +10001606
1607 /* Check target in extent form to max in temp */
1608 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1609 XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
1610 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
Dave Chinner24513372014-06-25 14:58:08 +10001611 return -EINVAL;
Dave Chinnera133d952013-08-12 20:49:48 +10001612
1613 /*
1614 * If we are in a btree format, check that the temp root block will fit
1615 * in the target and that it has enough extents to be in btree format
1616 * in the target.
1617 *
1618 * Note that we have to be careful to allow btree->extent conversions
1619 * (a common defrag case) which will occur when the temp inode is in
1620 * extent format...
1621 */
1622 if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1623 if (XFS_IFORK_BOFF(ip) &&
1624 XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip))
Dave Chinner24513372014-06-25 14:58:08 +10001625 return -EINVAL;
Dave Chinnera133d952013-08-12 20:49:48 +10001626 if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
1627 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
Dave Chinner24513372014-06-25 14:58:08 +10001628 return -EINVAL;
Dave Chinnera133d952013-08-12 20:49:48 +10001629 }
1630
1631 /* Reciprocal target->temp btree format checks */
1632 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1633 if (XFS_IFORK_BOFF(tip) &&
1634 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
Dave Chinner24513372014-06-25 14:58:08 +10001635 return -EINVAL;
Dave Chinnera133d952013-08-12 20:49:48 +10001636 if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
1637 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
Dave Chinner24513372014-06-25 14:58:08 +10001638 return -EINVAL;
Dave Chinnera133d952013-08-12 20:49:48 +10001639 }
1640
1641 return 0;
1642}
1643
Dave Chinner7abbb8f2014-09-23 16:20:11 +10001644static int
Dave Chinner4ef897a2014-08-04 13:44:08 +10001645xfs_swap_extent_flush(
1646 struct xfs_inode *ip)
1647{
1648 int error;
1649
1650 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1651 if (error)
1652 return error;
1653 truncate_pagecache_range(VFS_I(ip), 0, -1);
1654
1655 /* Verify O_DIRECT for ftmp */
1656 if (VFS_I(ip)->i_mapping->nrpages)
1657 return -EINVAL;
Dave Chinner4ef897a2014-08-04 13:44:08 +10001658 return 0;
1659}
1660
1661int
Dave Chinnera133d952013-08-12 20:49:48 +10001662xfs_swap_extents(
1663 xfs_inode_t *ip, /* target inode */
1664 xfs_inode_t *tip, /* tmp inode */
1665 xfs_swapext_t *sxp)
1666{
1667 xfs_mount_t *mp = ip->i_mount;
1668 xfs_trans_t *tp;
1669 xfs_bstat_t *sbp = &sxp->sx_stat;
1670 xfs_ifork_t *tempifp, *ifp, *tifp;
1671 int src_log_flags, target_log_flags;
1672 int error = 0;
1673 int aforkblks = 0;
1674 int taforkblks = 0;
1675 __uint64_t tmp;
Dave Chinner81217682014-08-04 13:29:32 +10001676 int lock_flags;
Dave Chinnera133d952013-08-12 20:49:48 +10001677
Dave Chinnera133d952013-08-12 20:49:48 +10001678 tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
1679 if (!tempifp) {
Dave Chinner24513372014-06-25 14:58:08 +10001680 error = -ENOMEM;
Dave Chinnera133d952013-08-12 20:49:48 +10001681 goto out;
1682 }
1683
1684 /*
Dave Chinner723cac42015-02-23 21:47:29 +11001685 * Lock the inodes against other IO, page faults and truncate to
1686 * begin with. Then we can ensure the inodes are flushed and have no
1687 * page cache safely. Once we have done this we can take the ilocks and
1688 * do the rest of the checks.
Dave Chinnera133d952013-08-12 20:49:48 +10001689 */
Dave Chinner723cac42015-02-23 21:47:29 +11001690 lock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
Dave Chinnera133d952013-08-12 20:49:48 +10001691 xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL);
Dave Chinner723cac42015-02-23 21:47:29 +11001692 xfs_lock_two_inodes(ip, tip, XFS_MMAPLOCK_EXCL);
Dave Chinnera133d952013-08-12 20:49:48 +10001693
1694 /* Verify that both files have the same format */
1695 if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
Dave Chinner24513372014-06-25 14:58:08 +10001696 error = -EINVAL;
Dave Chinnera133d952013-08-12 20:49:48 +10001697 goto out_unlock;
1698 }
1699
1700 /* Verify both files are either real-time or non-realtime */
1701 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
Dave Chinner24513372014-06-25 14:58:08 +10001702 error = -EINVAL;
Dave Chinnera133d952013-08-12 20:49:48 +10001703 goto out_unlock;
1704 }
1705
Dave Chinner4ef897a2014-08-04 13:44:08 +10001706 error = xfs_swap_extent_flush(ip);
Dave Chinnera133d952013-08-12 20:49:48 +10001707 if (error)
1708 goto out_unlock;
Dave Chinner4ef897a2014-08-04 13:44:08 +10001709 error = xfs_swap_extent_flush(tip);
1710 if (error)
1711 goto out_unlock;
Dave Chinnera133d952013-08-12 20:49:48 +10001712
Dave Chinner4ef897a2014-08-04 13:44:08 +10001713 tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
1714 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
1715 if (error) {
Christoph Hellwig4906e212015-06-04 13:47:56 +10001716 xfs_trans_cancel(tp);
Dave Chinnera133d952013-08-12 20:49:48 +10001717 goto out_unlock;
1718 }
Dave Chinner723cac42015-02-23 21:47:29 +11001719
1720 /*
1721 * Lock and join the inodes to the tansaction so that transaction commit
1722 * or cancel will unlock the inodes from this point onwards.
1723 */
Dave Chinner4ef897a2014-08-04 13:44:08 +10001724 xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
1725 lock_flags |= XFS_ILOCK_EXCL;
Dave Chinner723cac42015-02-23 21:47:29 +11001726 xfs_trans_ijoin(tp, ip, lock_flags);
1727 xfs_trans_ijoin(tp, tip, lock_flags);
1728
Dave Chinnera133d952013-08-12 20:49:48 +10001729
1730 /* Verify all data are being swapped */
1731 if (sxp->sx_offset != 0 ||
1732 sxp->sx_length != ip->i_d.di_size ||
1733 sxp->sx_length != tip->i_d.di_size) {
Dave Chinner24513372014-06-25 14:58:08 +10001734 error = -EFAULT;
Dave Chinner4ef897a2014-08-04 13:44:08 +10001735 goto out_trans_cancel;
Dave Chinnera133d952013-08-12 20:49:48 +10001736 }
1737
1738 trace_xfs_swap_extent_before(ip, 0);
1739 trace_xfs_swap_extent_before(tip, 1);
1740
1741 /* check inode formats now that data is flushed */
1742 error = xfs_swap_extents_check_format(ip, tip);
1743 if (error) {
1744 xfs_notice(mp,
1745 "%s: inode 0x%llx format is incompatible for exchanging.",
1746 __func__, ip->i_ino);
Dave Chinner4ef897a2014-08-04 13:44:08 +10001747 goto out_trans_cancel;
Dave Chinnera133d952013-08-12 20:49:48 +10001748 }
1749
1750 /*
1751 * Compare the current change & modify times with that
1752 * passed in. If they differ, we abort this swap.
1753 * This is the mechanism used to ensure the calling
1754 * process that the file was not changed out from
1755 * under it.
1756 */
1757 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
1758 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
1759 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
1760 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
Dave Chinner24513372014-06-25 14:58:08 +10001761 error = -EBUSY;
Dave Chinner81217682014-08-04 13:29:32 +10001762 goto out_trans_cancel;
Dave Chinnera133d952013-08-12 20:49:48 +10001763 }
Dave Chinnera133d952013-08-12 20:49:48 +10001764 /*
1765 * Count the number of extended attribute blocks
1766 */
1767 if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
1768 (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1769 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
1770 if (error)
1771 goto out_trans_cancel;
1772 }
1773 if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
1774 (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1775 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
1776 &taforkblks);
1777 if (error)
1778 goto out_trans_cancel;
1779 }
1780
Dave Chinner21b5c972013-08-30 10:23:44 +10001781 /*
1782 * Before we've swapped the forks, lets set the owners of the forks
1783 * appropriately. We have to do this as we are demand paging the btree
1784 * buffers, and so the validation done on read will expect the owner
1785 * field to be correctly set. Once we change the owners, we can swap the
1786 * inode forks.
1787 *
1788 * Note the trickiness in setting the log flags - we set the owner log
1789 * flag on the opposite inode (i.e. the inode we are setting the new
1790 * owner to be) because once we swap the forks and log that, log
1791 * recovery is going to see the fork as owned by the swapped inode,
1792 * not the pre-swapped inodes.
1793 */
1794 src_log_flags = XFS_ILOG_CORE;
1795 target_log_flags = XFS_ILOG_CORE;
1796 if (ip->i_d.di_version == 3 &&
1797 ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
Dave Chinner638f44162013-08-30 10:23:45 +10001798 target_log_flags |= XFS_ILOG_DOWNER;
1799 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK,
1800 tip->i_ino, NULL);
Dave Chinner21b5c972013-08-30 10:23:44 +10001801 if (error)
1802 goto out_trans_cancel;
1803 }
1804
1805 if (tip->i_d.di_version == 3 &&
1806 tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
Dave Chinner638f44162013-08-30 10:23:45 +10001807 src_log_flags |= XFS_ILOG_DOWNER;
1808 error = xfs_bmbt_change_owner(tp, tip, XFS_DATA_FORK,
1809 ip->i_ino, NULL);
Dave Chinner21b5c972013-08-30 10:23:44 +10001810 if (error)
1811 goto out_trans_cancel;
1812 }
1813
Dave Chinnera133d952013-08-12 20:49:48 +10001814 /*
1815 * Swap the data forks of the inodes
1816 */
1817 ifp = &ip->i_df;
1818 tifp = &tip->i_df;
1819 *tempifp = *ifp; /* struct copy */
1820 *ifp = *tifp; /* struct copy */
1821 *tifp = *tempifp; /* struct copy */
1822
1823 /*
1824 * Fix the on-disk inode values
1825 */
1826 tmp = (__uint64_t)ip->i_d.di_nblocks;
1827 ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1828 tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1829
1830 tmp = (__uint64_t) ip->i_d.di_nextents;
1831 ip->i_d.di_nextents = tip->i_d.di_nextents;
1832 tip->i_d.di_nextents = tmp;
1833
1834 tmp = (__uint64_t) ip->i_d.di_format;
1835 ip->i_d.di_format = tip->i_d.di_format;
1836 tip->i_d.di_format = tmp;
1837
1838 /*
1839 * The extents in the source inode could still contain speculative
1840 * preallocation beyond EOF (e.g. the file is open but not modified
1841 * while defrag is in progress). In that case, we need to copy over the
1842 * number of delalloc blocks the data fork in the source inode is
1843 * tracking beyond EOF so that when the fork is truncated away when the
1844 * temporary inode is unlinked we don't underrun the i_delayed_blks
1845 * counter on that inode.
1846 */
1847 ASSERT(tip->i_delayed_blks == 0);
1848 tip->i_delayed_blks = ip->i_delayed_blks;
1849 ip->i_delayed_blks = 0;
1850
Dave Chinnera133d952013-08-12 20:49:48 +10001851 switch (ip->i_d.di_format) {
1852 case XFS_DINODE_FMT_EXTENTS:
1853 /* If the extents fit in the inode, fix the
1854 * pointer. Otherwise it's already NULL or
1855 * pointing to the extent.
1856 */
1857 if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
1858 ifp->if_u1.if_extents =
1859 ifp->if_u2.if_inline_ext;
1860 }
1861 src_log_flags |= XFS_ILOG_DEXT;
1862 break;
1863 case XFS_DINODE_FMT_BTREE:
Dave Chinner21b5c972013-08-30 10:23:44 +10001864 ASSERT(ip->i_d.di_version < 3 ||
Dave Chinner638f44162013-08-30 10:23:45 +10001865 (src_log_flags & XFS_ILOG_DOWNER));
Dave Chinnera133d952013-08-12 20:49:48 +10001866 src_log_flags |= XFS_ILOG_DBROOT;
1867 break;
1868 }
1869
Dave Chinnera133d952013-08-12 20:49:48 +10001870 switch (tip->i_d.di_format) {
1871 case XFS_DINODE_FMT_EXTENTS:
1872 /* If the extents fit in the inode, fix the
1873 * pointer. Otherwise it's already NULL or
1874 * pointing to the extent.
1875 */
1876 if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
1877 tifp->if_u1.if_extents =
1878 tifp->if_u2.if_inline_ext;
1879 }
1880 target_log_flags |= XFS_ILOG_DEXT;
1881 break;
1882 case XFS_DINODE_FMT_BTREE:
1883 target_log_flags |= XFS_ILOG_DBROOT;
Dave Chinner21b5c972013-08-30 10:23:44 +10001884 ASSERT(tip->i_d.di_version < 3 ||
Dave Chinner638f44162013-08-30 10:23:45 +10001885 (target_log_flags & XFS_ILOG_DOWNER));
Dave Chinnera133d952013-08-12 20:49:48 +10001886 break;
1887 }
1888
Dave Chinnera133d952013-08-12 20:49:48 +10001889 xfs_trans_log_inode(tp, ip, src_log_flags);
1890 xfs_trans_log_inode(tp, tip, target_log_flags);
1891
1892 /*
1893 * If this is a synchronous mount, make sure that the
1894 * transaction goes to disk before returning to the user.
1895 */
1896 if (mp->m_flags & XFS_MOUNT_WSYNC)
1897 xfs_trans_set_sync(tp);
1898
Christoph Hellwig70393312015-06-04 13:48:08 +10001899 error = xfs_trans_commit(tp);
Dave Chinnera133d952013-08-12 20:49:48 +10001900
1901 trace_xfs_swap_extent_after(ip, 0);
1902 trace_xfs_swap_extent_after(tip, 1);
1903out:
1904 kmem_free(tempifp);
1905 return error;
1906
1907out_unlock:
Dave Chinner81217682014-08-04 13:29:32 +10001908 xfs_iunlock(ip, lock_flags);
1909 xfs_iunlock(tip, lock_flags);
Dave Chinnera133d952013-08-12 20:49:48 +10001910 goto out;
1911
1912out_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10001913 xfs_trans_cancel(tp);
Dave Chinner723cac42015-02-23 21:47:29 +11001914 goto out;
Dave Chinnera133d952013-08-12 20:49:48 +10001915}