blob: a63f61c256bdc54f0a8ee6088702dbd5e91ee9aa [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Olaf Weber3e57ecf2006-06-09 14:48:12 +10002 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
Christoph Hellwig51446f52016-09-19 11:10:21 +10003 * Copyright (c) 2016 Christoph Hellwig.
Nathan Scott7b718762005-11-02 14:58:39 +11004 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 *
Nathan Scott7b718762005-11-02 14:58:39 +11006 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * published by the Free Software Foundation.
9 *
Nathan Scott7b718762005-11-02 14:58:39 +110010 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070014 *
Nathan Scott7b718762005-11-02 14:58:39 +110015 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070018 */
Christoph Hellwig3b3dce02016-06-21 09:52:47 +100019#include <linux/iomap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include "xfs_fs.h"
Dave Chinner70a98832013-10-23 10:36:05 +110022#include "xfs_shared.h"
Dave Chinner239880e2013-10-23 10:50:10 +110023#include "xfs_format.h"
24#include "xfs_log_format.h"
25#include "xfs_trans_resv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_mount.h"
Darrick J. Wong3ab78df2016-08-03 11:15:38 +100027#include "xfs_defer.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110029#include "xfs_btree.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110030#include "xfs_bmap_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_bmap.h"
Dave Chinner68988112013-08-12 20:49:42 +100032#include "xfs_bmap_util.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_error.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110034#include "xfs_trans.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_trans_space.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_iomap.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000037#include "xfs_trace.h"
Brian Foster27b52862012-11-06 09:50:38 -050038#include "xfs_icache.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110039#include "xfs_quota.h"
Brian Foster76a42022013-03-18 10:51:47 -040040#include "xfs_dquot_item.h"
41#include "xfs_dquot.h"
Darrick J. Wong2a067052016-10-03 09:11:33 -070042#include "xfs_reflink.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
45#define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \
46 << mp->m_writeio_log)
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
Christoph Hellwige9c49732016-09-19 11:09:12 +100048void
49xfs_bmbt_to_iomap(
50 struct xfs_inode *ip,
51 struct iomap *iomap,
52 struct xfs_bmbt_irec *imap)
53{
54 struct xfs_mount *mp = ip->i_mount;
55
56 if (imap->br_startblock == HOLESTARTBLOCK) {
57 iomap->blkno = IOMAP_NULL_BLOCK;
58 iomap->type = IOMAP_HOLE;
59 } else if (imap->br_startblock == DELAYSTARTBLOCK) {
60 iomap->blkno = IOMAP_NULL_BLOCK;
61 iomap->type = IOMAP_DELALLOC;
62 } else {
63 iomap->blkno = xfs_fsb_to_db(ip, imap->br_startblock);
64 if (imap->br_state == XFS_EXT_UNWRITTEN)
65 iomap->type = IOMAP_UNWRITTEN;
66 else
67 iomap->type = IOMAP_MAPPED;
68 }
69 iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
70 iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
71 iomap->bdev = xfs_find_bdev_for_inode(VFS_I(ip));
72}
73
Darrick J. Wongf7ca3522016-10-03 09:11:43 -070074xfs_extlen_t
Christoph Hellwigf8e3a822016-09-19 11:09:28 +100075xfs_eof_alignment(
76 struct xfs_inode *ip,
77 xfs_extlen_t extsize)
Nathan Scottdd9f4382006-01-11 15:28:28 +110078{
Christoph Hellwigf8e3a822016-09-19 11:09:28 +100079 struct xfs_mount *mp = ip->i_mount;
80 xfs_extlen_t align = 0;
Nathan Scottdd9f4382006-01-11 15:28:28 +110081
Christoph Hellwigbf322d92011-12-18 20:00:05 +000082 if (!XFS_IS_REALTIME_INODE(ip)) {
83 /*
84 * Round up the allocation request to a stripe unit
85 * (m_dalign) boundary if the file size is >= stripe unit
86 * size, and we are allocating past the allocation eof.
87 *
88 * If mounted with the "-o swalloc" option the alignment is
89 * increased from the strip unit size to the stripe width.
90 */
91 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
92 align = mp->m_swidth;
93 else if (mp->m_dalign)
94 align = mp->m_dalign;
95
Peter Watkins76b57302014-12-04 09:30:51 +110096 if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
97 align = 0;
Christoph Hellwigbf322d92011-12-18 20:00:05 +000098 }
Nathan Scottdd9f4382006-01-11 15:28:28 +110099
100 /*
101 * Always round up the allocation request to an extent boundary
102 * (when file on a real-time subvolume or has di_extsize hint).
103 */
104 if (extsize) {
Peter Watkins76b57302014-12-04 09:30:51 +1100105 if (align)
106 align = roundup_64(align, extsize);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100107 else
108 align = extsize;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100109 }
110
Christoph Hellwigf8e3a822016-09-19 11:09:28 +1000111 return align;
112}
113
114STATIC int
115xfs_iomap_eof_align_last_fsb(
116 struct xfs_inode *ip,
117 xfs_extlen_t extsize,
118 xfs_fileoff_t *last_fsb)
119{
120 xfs_extlen_t align = xfs_eof_alignment(ip, extsize);
121
Peter Watkins76b57302014-12-04 09:30:51 +1100122 if (align) {
123 xfs_fileoff_t new_last_fsb = roundup_64(*last_fsb, align);
Christoph Hellwigf8e3a822016-09-19 11:09:28 +1000124 int eof, error;
125
Lachlan McIlroy541d7d32007-10-11 17:34:33 +1000126 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100127 if (error)
128 return error;
129 if (eof)
130 *last_fsb = new_last_fsb;
131 }
132 return 0;
133}
134
135STATIC int
Dave Chinner6d4a8ec2011-03-07 10:06:35 +1100136xfs_alert_fsblock_zero(
Nathan Scott572d95f2006-09-28 11:03:20 +1000137 xfs_inode_t *ip,
138 xfs_bmbt_irec_t *imap)
139{
Dave Chinner6a19d932011-03-07 10:02:35 +1100140 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
Nathan Scott572d95f2006-09-28 11:03:20 +1000141 "Access to block zero in inode %llu "
142 "start_block: %llx start_off: %llx "
Eric Sandeen08e96e12013-10-11 20:59:05 -0500143 "blkcnt: %llx extent-state: %x",
Nathan Scott572d95f2006-09-28 11:03:20 +1000144 (unsigned long long)ip->i_ino,
145 (unsigned long long)imap->br_startblock,
146 (unsigned long long)imap->br_startoff,
147 (unsigned long long)imap->br_blockcount,
148 imap->br_state);
Dave Chinner24513372014-06-25 14:58:08 +1000149 return -EFSCORRUPTED;
Nathan Scott572d95f2006-09-28 11:03:20 +1000150}
151
Christoph Hellwiga206c812010-12-10 08:42:20 +0000152int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153xfs_iomap_write_direct(
154 xfs_inode_t *ip,
Nathan Scottf403b7f2005-05-05 13:33:40 -0700155 xfs_off_t offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 size_t count,
Christoph Hellwig30704512010-06-24 11:42:19 +1000157 xfs_bmbt_irec_t *imap,
Christoph Hellwig405f8042010-12-10 08:42:19 +0000158 int nmaps)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159{
160 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 xfs_fileoff_t offset_fsb;
162 xfs_fileoff_t last_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100163 xfs_filblks_t count_fsb, resaligned;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 xfs_fsblock_t firstfsb;
Christoph Hellwigf13eb202017-02-06 10:42:26 -0800165 xfs_extlen_t extsz;
Eric Sandeen0116d932005-11-02 15:00:01 +1100166 int nimaps;
Nathan Scott06d10dd2005-06-21 15:48:47 +1000167 int quota_flag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 int rt;
169 xfs_trans_t *tp;
Darrick J. Wong2c3234d2016-08-03 11:19:29 +1000170 struct xfs_defer_ops dfops;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100171 uint qblocks, resblks, resrtextents;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100172 int error;
Brian Foster009c6e82015-10-12 15:34:20 +1100173 int lockmode;
Dave Chinner1ca19152015-11-03 12:37:00 +1100174 int bmapi_flags = XFS_BMAPI_PREALLOC;
Christoph Hellwig253f4912016-04-06 09:19:55 +1000175 uint tflags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176
Nathan Scottdd9f4382006-01-11 15:28:28 +1100177 rt = XFS_IS_REALTIME_INODE(ip);
David Chinner957d0eb2007-06-18 16:50:37 +1000178 extsz = xfs_get_extsz_hint(ip);
Brian Foster009c6e82015-10-12 15:34:20 +1100179 lockmode = XFS_ILOCK_SHARED; /* locked by caller */
180
181 ASSERT(xfs_isilocked(ip, lockmode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
David Chinner957d0eb2007-06-18 16:50:37 +1000183 offset_fsb = XFS_B_TO_FSBT(mp, offset);
184 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000185 if ((offset + count) > XFS_ISIZE(ip)) {
Brian Foster009c6e82015-10-12 15:34:20 +1100186 /*
187 * Assert that the in-core extent list is present since this can
188 * call xfs_iread_extents() and we only have the ilock shared.
189 * This should be safe because the lock was held around a bmapi
190 * call in the caller and we only need it to access the in-core
191 * list.
192 */
193 ASSERT(XFS_IFORK_PTR(ip, XFS_DATA_FORK)->if_flags &
194 XFS_IFEXTENTS);
Christoph Hellwigf8e3a822016-09-19 11:09:28 +1000195 error = xfs_iomap_eof_align_last_fsb(ip, extsz, &last_fsb);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100196 if (error)
Brian Foster009c6e82015-10-12 15:34:20 +1100197 goto out_unlock;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100198 } else {
Christoph Hellwig405f8042010-12-10 08:42:19 +0000199 if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
Nathan Scottdd9f4382006-01-11 15:28:28 +1100200 last_fsb = MIN(last_fsb, (xfs_fileoff_t)
Christoph Hellwig30704512010-06-24 11:42:19 +1000201 imap->br_blockcount +
202 imap->br_startoff);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100203 }
204 count_fsb = last_fsb - offset_fsb;
205 ASSERT(count_fsb > 0);
Christoph Hellwigf13eb202017-02-06 10:42:26 -0800206 resaligned = xfs_aligned_fsb_count(offset_fsb, count_fsb, extsz);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100207
208 if (unlikely(rt)) {
209 resrtextents = qblocks = resaligned;
210 resrtextents /= mp->m_sb.sb_rextsize;
David Chinner84e1e992007-06-18 16:50:27 +1000211 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
212 quota_flag = XFS_QMOPT_RES_RTBLKS;
213 } else {
214 resrtextents = 0;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100215 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
David Chinner84e1e992007-06-18 16:50:27 +1000216 quota_flag = XFS_QMOPT_RES_REGBLKS;
217 }
Nathan Scottdd9f4382006-01-11 15:28:28 +1100218
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 /*
Brian Foster009c6e82015-10-12 15:34:20 +1100220 * Drop the shared lock acquired by the caller, attach the dquot if
221 * necessary and move on to transaction setup.
222 */
223 xfs_iunlock(ip, lockmode);
224 error = xfs_qm_dqattach(ip, 0);
225 if (error)
226 return error;
227
228 /*
Dave Chinner1ca19152015-11-03 12:37:00 +1100229 * For DAX, we do not allocate unwritten extents, but instead we zero
230 * the block before we commit the transaction. Ideally we'd like to do
231 * this outside the transaction context, but if we commit and then crash
232 * we may not have zeroed the blocks and this will be exposed on
233 * recovery of the allocation. Hence we must zero before commit.
Dave Chinner3b0fe472016-01-04 16:22:45 +1100234 *
Dave Chinner1ca19152015-11-03 12:37:00 +1100235 * Further, if we are mapping unwritten extents here, we need to zero
236 * and convert them to written so that we don't need an unwritten extent
237 * callback for DAX. This also means that we need to be able to dip into
Dave Chinner3b0fe472016-01-04 16:22:45 +1100238 * the reserve block pool for bmbt block allocation if there is no space
239 * left but we need to do unwritten extent conversion.
Dave Chinner1ca19152015-11-03 12:37:00 +1100240 */
241 if (IS_DAX(VFS_I(ip))) {
242 bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
Christoph Hellwig63fbb4c2017-03-28 14:53:36 -0700243 if (imap->br_state == XFS_EXT_UNWRITTEN) {
Christoph Hellwig253f4912016-04-06 09:19:55 +1000244 tflags |= XFS_TRANS_RESERVE;
Dave Chinner3b0fe472016-01-04 16:22:45 +1100245 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
246 }
Dave Chinner1ca19152015-11-03 12:37:00 +1100247 }
Christoph Hellwig253f4912016-04-06 09:19:55 +1000248 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, resrtextents,
249 tflags, &tp);
250 if (error)
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000251 return error;
Dave Chinner507630b2012-03-27 10:34:50 -0400252
Brian Foster009c6e82015-10-12 15:34:20 +1100253 lockmode = XFS_ILOCK_EXCL;
254 xfs_ilock(ip, lockmode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
Christoph Hellwig7d095252009-06-08 15:33:32 +0200256 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100257 if (error)
Dave Chinner507630b2012-03-27 10:34:50 -0400258 goto out_trans_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
Christoph Hellwigddc34152011-09-19 15:00:54 +0000260 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 /*
Christoph Hellwig30704512010-06-24 11:42:19 +1000263 * From this point onwards we overwrite the imap pointer that the
264 * caller gave to us.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 */
Darrick J. Wong2c3234d2016-08-03 11:19:29 +1000266 xfs_defer_init(&dfops, &firstfsb);
Nathan Scott06d10dd2005-06-21 15:48:47 +1000267 nimaps = 1;
Christoph Hellwigd531d912014-02-10 10:27:43 +1100268 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
Dave Chinner264e89a2015-11-03 13:28:41 +1100269 bmapi_flags, &firstfsb, resblks, imap,
Darrick J. Wong2c3234d2016-08-03 11:19:29 +1000270 &nimaps, &dfops);
Nathan Scott06d10dd2005-06-21 15:48:47 +1000271 if (error)
Dave Chinner507630b2012-03-27 10:34:50 -0400272 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
274 /*
Nathan Scott06d10dd2005-06-21 15:48:47 +1000275 * Complete the transaction
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 */
Darrick J. Wong2c3234d2016-08-03 11:19:29 +1000277 error = xfs_defer_finish(&tp, &dfops, NULL);
Nathan Scott06d10dd2005-06-21 15:48:47 +1000278 if (error)
Dave Chinner507630b2012-03-27 10:34:50 -0400279 goto out_bmap_cancel;
Dave Chinner1ca19152015-11-03 12:37:00 +1100280
Christoph Hellwig70393312015-06-04 13:48:08 +1000281 error = xfs_trans_commit(tp);
Nathan Scott06d10dd2005-06-21 15:48:47 +1000282 if (error)
Dave Chinner507630b2012-03-27 10:34:50 -0400283 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284
Nathan Scott06d10dd2005-06-21 15:48:47 +1000285 /*
286 * Copy any maps to caller's array and return any error.
287 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 if (nimaps == 0) {
Dave Chinner24513372014-06-25 14:58:08 +1000289 error = -ENOSPC;
Dave Chinner507630b2012-03-27 10:34:50 -0400290 goto out_unlock;
Nathan Scott572d95f2006-09-28 11:03:20 +1000291 }
292
Dave Chinner507630b2012-03-27 10:34:50 -0400293 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
Dave Chinner6d4a8ec2011-03-07 10:06:35 +1100294 error = xfs_alert_fsblock_zero(ip, imap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295
Dave Chinner507630b2012-03-27 10:34:50 -0400296out_unlock:
Brian Foster009c6e82015-10-12 15:34:20 +1100297 xfs_iunlock(ip, lockmode);
Dave Chinner507630b2012-03-27 10:34:50 -0400298 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
Dave Chinner507630b2012-03-27 10:34:50 -0400300out_bmap_cancel:
Darrick J. Wong2c3234d2016-08-03 11:19:29 +1000301 xfs_defer_cancel(&dfops);
Dave Chinnerea562ed2012-05-08 20:48:53 +1000302 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
Dave Chinner507630b2012-03-27 10:34:50 -0400303out_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +1000304 xfs_trans_cancel(tp);
Dave Chinner507630b2012-03-27 10:34:50 -0400305 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306}
307
Brian Foster76a42022013-03-18 10:51:47 -0400308STATIC bool
309xfs_quota_need_throttle(
310 struct xfs_inode *ip,
311 int type,
312 xfs_fsblock_t alloc_blocks)
313{
314 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
315
316 if (!dq || !xfs_this_quota_on(ip->i_mount, type))
317 return false;
318
319 /* no hi watermark, no throttle */
320 if (!dq->q_prealloc_hi_wmark)
321 return false;
322
323 /* under the lo watermark, no throttle */
324 if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark)
325 return false;
326
327 return true;
328}
329
330STATIC void
331xfs_quota_calc_throttle(
332 struct xfs_inode *ip,
333 int type,
334 xfs_fsblock_t *qblocks,
Brian Fosterf0740512014-07-24 19:56:08 +1000335 int *qshift,
336 int64_t *qfreesp)
Brian Foster76a42022013-03-18 10:51:47 -0400337{
338 int64_t freesp;
339 int shift = 0;
340 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
341
Eric Sandeen5cca3f62014-10-02 09:27:09 +1000342 /* no dq, or over hi wmark, squash the prealloc completely */
343 if (!dq || dq->q_res_bcount >= dq->q_prealloc_hi_wmark) {
Brian Foster76a42022013-03-18 10:51:47 -0400344 *qblocks = 0;
Brian Fosterf0740512014-07-24 19:56:08 +1000345 *qfreesp = 0;
Brian Foster76a42022013-03-18 10:51:47 -0400346 return;
347 }
348
349 freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount;
350 if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
351 shift = 2;
352 if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
353 shift += 2;
354 if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
355 shift += 2;
356 }
357
Brian Fosterf0740512014-07-24 19:56:08 +1000358 if (freesp < *qfreesp)
359 *qfreesp = freesp;
360
Brian Foster76a42022013-03-18 10:51:47 -0400361 /* only overwrite the throttle values if we are more aggressive */
362 if ((freesp >> shift) < (*qblocks >> *qshift)) {
363 *qblocks = freesp;
364 *qshift = shift;
365 }
366}
367
Dave Chinnera1e16c22013-02-11 16:05:01 +1100368/*
Christoph Hellwig51446f52016-09-19 11:10:21 +1000369 * If we are doing a write at the end of the file and there are no allocations
370 * past this one, then extend the allocation out to the file system's write
371 * iosize.
372 *
Dave Chinner055388a2011-01-04 11:35:03 +1100373 * If we don't have a user specified preallocation size, dynamically increase
Christoph Hellwig51446f52016-09-19 11:10:21 +1000374 * the preallocation size as the size of the file grows. Cap the maximum size
Dave Chinner055388a2011-01-04 11:35:03 +1100375 * at a single extent or less if the filesystem is near full. The closer the
376 * filesystem is to full, the smaller the maximum prealocation.
Christoph Hellwig51446f52016-09-19 11:10:21 +1000377 *
378 * As an exception we don't do any preallocation at all if the file is smaller
379 * than the minimum preallocation and we are using the default dynamic
380 * preallocation scheme, as it is likely this is the only write to the file that
381 * is going to be done.
382 *
383 * We clean up any extra space left over when the file is closed in
384 * xfs_inactive().
Dave Chinner055388a2011-01-04 11:35:03 +1100385 */
386STATIC xfs_fsblock_t
387xfs_iomap_prealloc_size(
Dave Chinnera1e16c22013-02-11 16:05:01 +1100388 struct xfs_inode *ip,
Christoph Hellwig51446f52016-09-19 11:10:21 +1000389 loff_t offset,
390 loff_t count,
Christoph Hellwig656152e2016-11-24 11:39:44 +1100391 xfs_extnum_t idx)
Dave Chinner055388a2011-01-04 11:35:03 +1100392{
Christoph Hellwig51446f52016-09-19 11:10:21 +1000393 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwig656152e2016-11-24 11:39:44 +1100394 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
Christoph Hellwig51446f52016-09-19 11:10:21 +1000395 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
Christoph Hellwig656152e2016-11-24 11:39:44 +1100396 struct xfs_bmbt_irec prev;
Brian Foster3c58b5f2013-03-18 10:51:43 -0400397 int shift = 0;
398 int64_t freesp;
Brian Foster76a42022013-03-18 10:51:47 -0400399 xfs_fsblock_t qblocks;
400 int qshift = 0;
Christoph Hellwig51446f52016-09-19 11:10:21 +1000401 xfs_fsblock_t alloc_blocks = 0;
Dave Chinner055388a2011-01-04 11:35:03 +1100402
Christoph Hellwig51446f52016-09-19 11:10:21 +1000403 if (offset + count <= XFS_ISIZE(ip))
404 return 0;
405
406 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) &&
407 (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks)))
408 return 0;
409
410 /*
411 * If an explicit allocsize is set, the file is small, or we
412 * are writing behind a hole, then use the minimum prealloc:
413 */
414 if ((mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) ||
415 XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
Christoph Hellwig656152e2016-11-24 11:39:44 +1100416 !xfs_iext_get_extent(ifp, idx - 1, &prev) ||
417 prev.br_startoff + prev.br_blockcount < offset_fsb)
Christoph Hellwig51446f52016-09-19 11:10:21 +1000418 return mp->m_writeio_blocks;
419
420 /*
421 * Determine the initial size of the preallocation. We are beyond the
422 * current EOF here, but we need to take into account whether this is
423 * a sparse write or an extending write when determining the
424 * preallocation size. Hence we need to look up the extent that ends
425 * at the current write offset and use the result to determine the
426 * preallocation size.
427 *
428 * If the extent is a hole, then preallocation is essentially disabled.
429 * Otherwise we take the size of the preceding data extent as the basis
430 * for the preallocation size. If the size of the extent is greater than
431 * half the maximum extent length, then use the current offset as the
432 * basis. This ensures that for large files the preallocation size
433 * always extends to MAXEXTLEN rather than falling short due to things
434 * like stripe unit/width alignment of real extents.
435 */
Christoph Hellwig656152e2016-11-24 11:39:44 +1100436 if (prev.br_blockcount <= (MAXEXTLEN >> 1))
437 alloc_blocks = prev.br_blockcount << 1;
Christoph Hellwig51446f52016-09-19 11:10:21 +1000438 else
439 alloc_blocks = XFS_B_TO_FSB(mp, offset);
Brian Foster3c58b5f2013-03-18 10:51:43 -0400440 if (!alloc_blocks)
441 goto check_writeio;
Brian Foster76a42022013-03-18 10:51:47 -0400442 qblocks = alloc_blocks;
Dave Chinner055388a2011-01-04 11:35:03 +1100443
Brian Fosterc9bdbdc2013-03-18 10:51:44 -0400444 /*
445 * MAXEXTLEN is not a power of two value but we round the prealloc down
446 * to the nearest power of two value after throttling. To prevent the
447 * round down from unconditionally reducing the maximum supported prealloc
448 * size, we round up first, apply appropriate throttling, round down and
449 * cap the value to MAXEXTLEN.
450 */
451 alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
452 alloc_blocks);
Dave Chinner055388a2011-01-04 11:35:03 +1100453
Dave Chinner0d485ad2015-02-23 21:22:03 +1100454 freesp = percpu_counter_read_positive(&mp->m_fdblocks);
Brian Foster3c58b5f2013-03-18 10:51:43 -0400455 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
456 shift = 2;
457 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
458 shift++;
459 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
460 shift++;
461 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
462 shift++;
463 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
464 shift++;
Dave Chinner055388a2011-01-04 11:35:03 +1100465 }
Brian Foster76a42022013-03-18 10:51:47 -0400466
467 /*
Brian Fosterf0740512014-07-24 19:56:08 +1000468 * Check each quota to cap the prealloc size, provide a shift value to
469 * throttle with and adjust amount of available space.
Brian Foster76a42022013-03-18 10:51:47 -0400470 */
471 if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks))
Brian Fosterf0740512014-07-24 19:56:08 +1000472 xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift,
473 &freesp);
Brian Foster76a42022013-03-18 10:51:47 -0400474 if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks))
Brian Fosterf0740512014-07-24 19:56:08 +1000475 xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift,
476 &freesp);
Brian Foster76a42022013-03-18 10:51:47 -0400477 if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks))
Brian Fosterf0740512014-07-24 19:56:08 +1000478 xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift,
479 &freesp);
Brian Foster76a42022013-03-18 10:51:47 -0400480
481 /*
482 * The final prealloc size is set to the minimum of free space available
483 * in each of the quotas and the overall filesystem.
484 *
485 * The shift throttle value is set to the maximum value as determined by
486 * the global low free space values and per-quota low free space values.
487 */
488 alloc_blocks = MIN(alloc_blocks, qblocks);
489 shift = MAX(shift, qshift);
490
Brian Foster3c58b5f2013-03-18 10:51:43 -0400491 if (shift)
492 alloc_blocks >>= shift;
Brian Fosterc9bdbdc2013-03-18 10:51:44 -0400493 /*
494 * rounddown_pow_of_two() returns an undefined result if we pass in
495 * alloc_blocks = 0.
496 */
497 if (alloc_blocks)
498 alloc_blocks = rounddown_pow_of_two(alloc_blocks);
499 if (alloc_blocks > MAXEXTLEN)
500 alloc_blocks = MAXEXTLEN;
Dave Chinner055388a2011-01-04 11:35:03 +1100501
Brian Foster3c58b5f2013-03-18 10:51:43 -0400502 /*
503 * If we are still trying to allocate more space than is
504 * available, squash the prealloc hard. This can happen if we
505 * have a large file on a small filesystem and the above
506 * lowspace thresholds are smaller than MAXEXTLEN.
507 */
508 while (alloc_blocks && alloc_blocks >= freesp)
509 alloc_blocks >>= 4;
Brian Foster3c58b5f2013-03-18 10:51:43 -0400510check_writeio:
Dave Chinner055388a2011-01-04 11:35:03 +1100511 if (alloc_blocks < mp->m_writeio_blocks)
512 alloc_blocks = mp->m_writeio_blocks;
Brian Foster19cb7e32013-03-18 10:51:48 -0400513 trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
514 mp->m_writeio_blocks);
Dave Chinner055388a2011-01-04 11:35:03 +1100515 return alloc_blocks;
516}
517
Christoph Hellwig51446f52016-09-19 11:10:21 +1000518static int
519xfs_file_iomap_begin_delay(
520 struct inode *inode,
521 loff_t offset,
522 loff_t count,
523 unsigned flags,
524 struct iomap *iomap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525{
Christoph Hellwig51446f52016-09-19 11:10:21 +1000526 struct xfs_inode *ip = XFS_I(inode);
527 struct xfs_mount *mp = ip->i_mount;
528 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
529 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
530 xfs_fileoff_t maxbytes_fsb =
531 XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
Brian Fosterf7820882016-11-28 14:57:42 +1100532 xfs_fileoff_t end_fsb;
Christoph Hellwig51446f52016-09-19 11:10:21 +1000533 int error = 0, eof = 0;
534 struct xfs_bmbt_irec got;
Christoph Hellwig51446f52016-09-19 11:10:21 +1000535 xfs_extnum_t idx;
Brian Fosterf7820882016-11-28 14:57:42 +1100536 xfs_fsblock_t prealloc_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537
Christoph Hellwig51446f52016-09-19 11:10:21 +1000538 ASSERT(!XFS_IS_REALTIME_INODE(ip));
539 ASSERT(!xfs_get_extsz_hint(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
Christoph Hellwig51446f52016-09-19 11:10:21 +1000541 xfs_ilock(ip, XFS_ILOCK_EXCL);
542
543 if (unlikely(XFS_TEST_ERROR(
544 (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
545 XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
546 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
547 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
548 error = -EFSCORRUPTED;
549 goto out_unlock;
550 }
551
552 XFS_STATS_INC(mp, xs_blk_mapw);
553
554 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
555 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
556 if (error)
557 goto out_unlock;
558 }
559
Christoph Hellwig656152e2016-11-24 11:39:44 +1100560 eof = !xfs_iext_lookup_extent(ip, ifp, offset_fsb, &idx, &got);
Christoph Hellwig51446f52016-09-19 11:10:21 +1000561 if (!eof && got.br_startoff <= offset_fsb) {
Christoph Hellwig3ba020b2016-10-20 15:53:50 +1100562 if (xfs_is_reflink_inode(ip)) {
563 bool shared;
564
565 end_fsb = min(XFS_B_TO_FSB(mp, offset + count),
566 maxbytes_fsb);
567 xfs_trim_extent(&got, offset_fsb, end_fsb - offset_fsb);
568 error = xfs_reflink_reserve_cow(ip, &got, &shared);
569 if (error)
570 goto out_unlock;
571 }
572
Christoph Hellwig51446f52016-09-19 11:10:21 +1000573 trace_xfs_iomap_found(ip, offset, count, 0, &got);
574 goto done;
575 }
576
Christoph Hellwig7d095252009-06-08 15:33:32 +0200577 error = xfs_qm_dqattach_locked(ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 if (error)
Christoph Hellwig51446f52016-09-19 11:10:21 +1000579 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
Christoph Hellwig51446f52016-09-19 11:10:21 +1000581 /*
582 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES pages
583 * to keep the chunks of work done where somewhat symmetric with the
584 * work writeback does. This is a completely arbitrary number pulled
585 * out of thin air as a best guess for initial testing.
586 *
587 * Note that the values needs to be less than 32-bits wide until
588 * the lower level functions are updated.
589 */
590 count = min_t(loff_t, count, 1024 * PAGE_SIZE);
Brian Fosterf7820882016-11-28 14:57:42 +1100591 end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100592
Christoph Hellwig51446f52016-09-19 11:10:21 +1000593 if (eof) {
Christoph Hellwig656152e2016-11-24 11:39:44 +1100594 prealloc_blocks = xfs_iomap_prealloc_size(ip, offset, count, idx);
Christoph Hellwig51446f52016-09-19 11:10:21 +1000595 if (prealloc_blocks) {
596 xfs_extlen_t align;
597 xfs_off_t end_offset;
Brian Fosterf7820882016-11-28 14:57:42 +1100598 xfs_fileoff_t p_end_fsb;
Christoph Hellwig51446f52016-09-19 11:10:21 +1000599
600 end_offset = XFS_WRITEIO_ALIGN(mp, offset + count - 1);
Brian Fosterf7820882016-11-28 14:57:42 +1100601 p_end_fsb = XFS_B_TO_FSBT(mp, end_offset) +
602 prealloc_blocks;
Christoph Hellwig51446f52016-09-19 11:10:21 +1000603
604 align = xfs_eof_alignment(ip, 0);
605 if (align)
Brian Fosterf7820882016-11-28 14:57:42 +1100606 p_end_fsb = roundup_64(p_end_fsb, align);
Christoph Hellwig51446f52016-09-19 11:10:21 +1000607
Brian Fosterf7820882016-11-28 14:57:42 +1100608 p_end_fsb = min(p_end_fsb, maxbytes_fsb);
609 ASSERT(p_end_fsb > offset_fsb);
610 prealloc_blocks = p_end_fsb - end_fsb;
Christoph Hellwig51446f52016-09-19 11:10:21 +1000611 }
612 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
Dave Chinner8de2bf92009-04-06 18:49:12 +0200614retry:
Darrick J. Wongbe51f812016-10-03 09:11:32 -0700615 error = xfs_bmapi_reserve_delalloc(ip, XFS_DATA_FORK, offset_fsb,
Brian Fosterf7820882016-11-28 14:57:42 +1100616 end_fsb - offset_fsb, prealloc_blocks, &got, &idx, eof);
Dave Chinner055388a2011-01-04 11:35:03 +1100617 switch (error) {
618 case 0:
Christoph Hellwig51446f52016-09-19 11:10:21 +1000619 break;
Dave Chinner24513372014-06-25 14:58:08 +1000620 case -ENOSPC:
621 case -EDQUOT:
Christoph Hellwig51446f52016-09-19 11:10:21 +1000622 /* retry without any preallocation */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000623 trace_xfs_delalloc_enospc(ip, offset, count);
Brian Fosterf7820882016-11-28 14:57:42 +1100624 if (prealloc_blocks) {
625 prealloc_blocks = 0;
Dave Chinner9aa05002012-10-08 21:56:04 +1100626 goto retry;
Dave Chinner055388a2011-01-04 11:35:03 +1100627 }
Christoph Hellwig51446f52016-09-19 11:10:21 +1000628 /*FALLTHRU*/
629 default:
630 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 }
632
Brian Fosterf65e6fa2017-03-08 09:58:08 -0800633 /*
634 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch
635 * them out if the write happens to fail.
636 */
637 iomap->flags = IOMAP_F_NEW;
Christoph Hellwig51446f52016-09-19 11:10:21 +1000638 trace_xfs_iomap_alloc(ip, offset, count, 0, &got);
639done:
640 if (isnullstartblock(got.br_startblock))
641 got.br_startblock = DELAYSTARTBLOCK;
642
643 if (!got.br_startblock) {
644 error = xfs_alert_fsblock_zero(ip, &got);
645 if (error)
646 goto out_unlock;
647 }
648
649 xfs_bmbt_to_iomap(ip, iomap, &got);
650
651out_unlock:
652 xfs_iunlock(ip, XFS_ILOCK_EXCL);
653 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654}
655
656/*
657 * Pass in a delayed allocate extent, convert it to real extents;
658 * return to the caller the extent we create which maps on top of
659 * the originating callers request.
660 *
661 * Called without a lock on the inode.
David Chinnere4143a12007-11-23 16:29:11 +1100662 *
663 * We no longer bother to look at the incoming map - all we have to
664 * guarantee is that whatever we allocate fills the required range.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 */
Christoph Hellwiga206c812010-12-10 08:42:20 +0000666int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667xfs_iomap_write_allocate(
668 xfs_inode_t *ip,
Darrick J. Wong60b49842016-10-03 09:11:34 -0700669 int whichfork,
Nathan Scottf403b7f2005-05-05 13:33:40 -0700670 xfs_off_t offset,
Christoph Hellwig405f8042010-12-10 08:42:19 +0000671 xfs_bmbt_irec_t *imap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672{
673 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 xfs_fileoff_t offset_fsb, last_block;
675 xfs_fileoff_t end_fsb, map_start_fsb;
676 xfs_fsblock_t first_block;
Darrick J. Wong2c3234d2016-08-03 11:19:29 +1000677 struct xfs_defer_ops dfops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 xfs_filblks_t count_fsb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 xfs_trans_t *tp;
Eric Sandeenf6106ef2016-01-11 11:34:01 +1100680 int nimaps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 int error = 0;
Christoph Hellwigd2b39642017-01-20 09:31:54 -0800682 int flags = XFS_BMAPI_DELALLOC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 int nres;
684
Darrick J. Wong60b49842016-10-03 09:11:34 -0700685 if (whichfork == XFS_COW_FORK)
Darrick J. Wong5eda4302017-02-02 15:14:02 -0800686 flags |= XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC;
Darrick J. Wong60b49842016-10-03 09:11:34 -0700687
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 /*
689 * Make sure that the dquots are there.
690 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200691 error = xfs_qm_dqattach(ip, 0);
692 if (error)
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000693 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
Nathan Scott24e17b52005-05-05 13:33:20 -0700695 offset_fsb = XFS_B_TO_FSBT(mp, offset);
Christoph Hellwig30704512010-06-24 11:42:19 +1000696 count_fsb = imap->br_blockcount;
697 map_start_fsb = imap->br_startoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100699 XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
701 while (count_fsb != 0) {
702 /*
703 * Set up a transaction with which to allocate the
704 * backing store for the file. Do allocations in a
705 * loop until we get some space in the range we are
706 * interested in. The other space that might be allocated
707 * is in the delayed allocation extent on which we sit
708 * but before our buffer starts.
709 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 nimaps = 0;
711 while (nimaps == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
Christoph Hellwig0af32fb2016-08-17 08:30:28 +1000713 /*
714 * We have already reserved space for the extent and any
715 * indirect blocks when creating the delalloc extent,
716 * there is no need to reserve space in this transaction
717 * again.
718 */
719 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0,
Christoph Hellwig253f4912016-04-06 09:19:55 +1000720 0, XFS_TRANS_RESERVE, &tp);
721 if (error)
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000722 return error;
Christoph Hellwig253f4912016-04-06 09:19:55 +1000723
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigddc34152011-09-19 15:00:54 +0000725 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726
Darrick J. Wong2c3234d2016-08-03 11:19:29 +1000727 xfs_defer_init(&dfops, &first_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 /*
David Chinnere4143a12007-11-23 16:29:11 +1100730 * it is possible that the extents have changed since
731 * we did the read call as we dropped the ilock for a
732 * while. We have to be careful about truncates or hole
733 * punchs here - we are not allowed to allocate
734 * non-delalloc blocks here.
735 *
736 * The only protection against truncation is the pages
737 * for the range we are being asked to convert are
738 * locked and hence a truncate will block on them
739 * first.
740 *
741 * As a result, if we go beyond the range we really
742 * need and hit an delalloc extent boundary followed by
743 * a hole while we have excess blocks in the map, we
744 * will fill the hole incorrectly and overrun the
745 * transaction reservation.
746 *
747 * Using a single map prevents this as we are forced to
748 * check each map we look for overlap with the desired
749 * range and abort as soon as we find it. Also, given
750 * that we only return a single map, having one beyond
751 * what we can return is probably a bit silly.
752 *
753 * We also need to check that we don't go beyond EOF;
754 * this is a truncate optimisation as a truncate sets
755 * the new file size before block on the pages we
756 * currently have locked under writeback. Because they
757 * are about to be tossed, we don't need to write them
758 * back....
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 */
David Chinnere4143a12007-11-23 16:29:11 +1100760 nimaps = 1;
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000761 end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
Eric Sandeen7fb2cd42014-04-14 18:58:05 +1000762 error = xfs_bmap_last_offset(ip, &last_block,
David Chinner7c9ef852008-04-10 12:21:59 +1000763 XFS_DATA_FORK);
764 if (error)
765 goto trans_cancel;
766
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
768 if ((map_start_fsb + count_fsb) > last_block) {
769 count_fsb = last_block - map_start_fsb;
770 if (count_fsb == 0) {
Dave Chinner24513372014-06-25 14:58:08 +1000771 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 goto trans_cancel;
773 }
774 }
775
Christoph Hellwig30704512010-06-24 11:42:19 +1000776 /*
Christoph Hellwig30704512010-06-24 11:42:19 +1000777 * From this point onwards we overwrite the imap
778 * pointer that the caller gave to us.
779 */
Dave Chinnerc0dc7822011-09-18 20:40:52 +0000780 error = xfs_bmapi_write(tp, ip, map_start_fsb,
Darrick J. Wong60b49842016-10-03 09:11:34 -0700781 count_fsb, flags, &first_block,
Brian Fosterdbd5c8c2015-10-12 16:04:13 +1100782 nres, imap, &nimaps,
Darrick J. Wong2c3234d2016-08-03 11:19:29 +1000783 &dfops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 if (error)
785 goto trans_cancel;
786
Darrick J. Wong2c3234d2016-08-03 11:19:29 +1000787 error = xfs_defer_finish(&tp, &dfops, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 if (error)
789 goto trans_cancel;
790
Christoph Hellwig70393312015-06-04 13:48:08 +1000791 error = xfs_trans_commit(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 if (error)
793 goto error0;
794
795 xfs_iunlock(ip, XFS_ILOCK_EXCL);
796 }
797
798 /*
799 * See if we were able to allocate an extent that
800 * covers at least part of the callers request
801 */
Christoph Hellwig30704512010-06-24 11:42:19 +1000802 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
Dave Chinner6d4a8ec2011-03-07 10:06:35 +1100803 return xfs_alert_fsblock_zero(ip, imap);
David Chinner86c4d622008-04-29 12:53:21 +1000804
Christoph Hellwig30704512010-06-24 11:42:19 +1000805 if ((offset_fsb >= imap->br_startoff) &&
806 (offset_fsb < (imap->br_startoff +
807 imap->br_blockcount))) {
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100808 XFS_STATS_INC(mp, xs_xstrat_quick);
David Chinnere4143a12007-11-23 16:29:11 +1100809 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 }
811
David Chinnere4143a12007-11-23 16:29:11 +1100812 /*
813 * So far we have not mapped the requested part of the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 * file, just surrounding data, try again.
815 */
Christoph Hellwig30704512010-06-24 11:42:19 +1000816 count_fsb -= imap->br_blockcount;
817 map_start_fsb = imap->br_startoff + imap->br_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 }
819
820trans_cancel:
Darrick J. Wong2c3234d2016-08-03 11:19:29 +1000821 xfs_defer_cancel(&dfops);
Christoph Hellwig4906e212015-06-04 13:47:56 +1000822 xfs_trans_cancel(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823error0:
824 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000825 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826}
827
828int
829xfs_iomap_write_unwritten(
830 xfs_inode_t *ip,
Nathan Scottf403b7f2005-05-05 13:33:40 -0700831 xfs_off_t offset,
Christoph Hellwigd32057f2015-01-09 10:48:12 +1100832 xfs_off_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833{
834 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 xfs_fileoff_t offset_fsb;
836 xfs_filblks_t count_fsb;
837 xfs_filblks_t numblks_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100838 xfs_fsblock_t firstfsb;
839 int nimaps;
840 xfs_trans_t *tp;
841 xfs_bmbt_irec_t imap;
Darrick J. Wong2c3234d2016-08-03 11:19:29 +1000842 struct xfs_defer_ops dfops;
Christoph Hellwig84803fb2012-02-29 09:53:50 +0000843 xfs_fsize_t i_size;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100844 uint resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000847 trace_xfs_unwritten_convert(ip, offset, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
849 offset_fsb = XFS_B_TO_FSBT(mp, offset);
850 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
851 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
852
Lachlan McIlroy4ddd8bb2008-06-27 13:32:53 +1000853 /*
854 * Reserve enough blocks in this transaction for two complete extent
855 * btree splits. We may be converting the middle part of an unwritten
856 * extent and in this case we will insert two new extents in the btree
857 * each of which could cause a full split.
858 *
859 * This reservation amount will be used in the first call to
860 * xfs_bmbt_split() to select an AG with enough space to satisfy the
861 * rest of the operation.
862 */
Nathan Scottdd9f4382006-01-11 15:28:28 +1100863 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
Nathan Scottdd9f4382006-01-11 15:28:28 +1100865 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 /*
Christoph Hellwig253f4912016-04-06 09:19:55 +1000867 * Set up a transaction to convert the range of extents
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 * from unwritten to real. Do allocations in a loop until
869 * we have covered the range passed in.
Christoph Hellwig80641dc2009-10-19 04:00:03 +0000870 *
Christoph Hellwig253f4912016-04-06 09:19:55 +1000871 * Note that we can't risk to recursing back into the filesystem
872 * here as we might be asked to write out the same inode that we
873 * complete here and might deadlock on the iolock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 */
Christoph Hellwig253f4912016-04-06 09:19:55 +1000875 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0,
876 XFS_TRANS_RESERVE | XFS_TRANS_NOFS, &tp);
877 if (error)
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000878 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
880 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigddc34152011-09-19 15:00:54 +0000881 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882
883 /*
884 * Modify the unwritten extent state of the buffer.
885 */
Darrick J. Wong2c3234d2016-08-03 11:19:29 +1000886 xfs_defer_init(&dfops, &firstfsb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 nimaps = 1;
Dave Chinnerc0dc7822011-09-18 20:40:52 +0000888 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
Brian Fosterdbd5c8c2015-10-12 16:04:13 +1100889 XFS_BMAPI_CONVERT, &firstfsb, resblks,
Darrick J. Wong2c3234d2016-08-03 11:19:29 +1000890 &imap, &nimaps, &dfops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 if (error)
892 goto error_on_bmapi_transaction;
893
Christoph Hellwig84803fb2012-02-29 09:53:50 +0000894 /*
895 * Log the updated inode size as we go. We have to be careful
896 * to only log it up to the actual write offset if it is
897 * halfway into a block.
898 */
899 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
900 if (i_size > offset + count)
901 i_size = offset + count;
902
903 i_size = xfs_new_eof(ip, i_size);
904 if (i_size) {
905 ip->i_d.di_size = i_size;
906 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
907 }
908
Darrick J. Wong2c3234d2016-08-03 11:19:29 +1000909 error = xfs_defer_finish(&tp, &dfops, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 if (error)
911 goto error_on_bmapi_transaction;
912
Christoph Hellwig70393312015-06-04 13:48:08 +1000913 error = xfs_trans_commit(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 xfs_iunlock(ip, XFS_ILOCK_EXCL);
915 if (error)
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000916 return error;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100917
David Chinner86c4d622008-04-29 12:53:21 +1000918 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
Dave Chinner6d4a8ec2011-03-07 10:06:35 +1100919 return xfs_alert_fsblock_zero(ip, &imap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
921 if ((numblks_fsb = imap.br_blockcount) == 0) {
922 /*
923 * The numblks_fsb value should always get
924 * smaller, otherwise the loop is stuck.
925 */
926 ASSERT(imap.br_blockcount);
927 break;
928 }
929 offset_fsb += numblks_fsb;
930 count_fsb -= numblks_fsb;
931 } while (count_fsb > 0);
932
933 return 0;
934
935error_on_bmapi_transaction:
Darrick J. Wong2c3234d2016-08-03 11:19:29 +1000936 xfs_defer_cancel(&dfops);
Christoph Hellwig4906e212015-06-04 13:47:56 +1000937 xfs_trans_cancel(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000939 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940}
Christoph Hellwig3b3dce02016-06-21 09:52:47 +1000941
Christoph Hellwig6c31f492016-09-19 11:28:38 +1000942static inline bool imap_needs_alloc(struct inode *inode,
943 struct xfs_bmbt_irec *imap, int nimaps)
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +1000944{
945 return !nimaps ||
946 imap->br_startblock == HOLESTARTBLOCK ||
Christoph Hellwig6c31f492016-09-19 11:28:38 +1000947 imap->br_startblock == DELAYSTARTBLOCK ||
Christoph Hellwig63fbb4c2017-03-28 14:53:36 -0700948 (IS_DAX(inode) && imap->br_state == XFS_EXT_UNWRITTEN);
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +1000949}
950
Christoph Hellwigacdda3a2016-11-30 14:37:15 +1100951static inline bool need_excl_ilock(struct xfs_inode *ip, unsigned flags)
952{
953 /*
954 * COW writes will allocate delalloc space, so we need to make sure
955 * to take the lock exclusively here.
956 */
957 if (xfs_is_reflink_inode(ip) && (flags & (IOMAP_WRITE | IOMAP_ZERO)))
958 return true;
959 if ((flags & IOMAP_DIRECT) && (flags & IOMAP_WRITE))
960 return true;
961 return false;
962}
963
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +1000964static int
965xfs_file_iomap_begin(
966 struct inode *inode,
967 loff_t offset,
968 loff_t length,
969 unsigned flags,
970 struct iomap *iomap)
971{
972 struct xfs_inode *ip = XFS_I(inode);
973 struct xfs_mount *mp = ip->i_mount;
974 struct xfs_bmbt_irec imap;
975 xfs_fileoff_t offset_fsb, end_fsb;
976 int nimaps = 1, error = 0;
Christoph Hellwig3ba020b2016-10-20 15:53:50 +1100977 bool shared = false, trimmed = false;
Christoph Hellwig66642c52016-09-19 11:26:39 +1000978 unsigned lockmode;
Dan Williamsfa5d9322017-01-27 12:04:59 -0800979 struct block_device *bdev;
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +1000980
981 if (XFS_FORCED_SHUTDOWN(mp))
982 return -EIO;
983
Christoph Hellwigacdda3a2016-11-30 14:37:15 +1100984 if (((flags & (IOMAP_WRITE | IOMAP_DIRECT)) == IOMAP_WRITE) &&
985 !IS_DAX(inode) && !xfs_get_extsz_hint(ip)) {
Darrick J. Wong2a067052016-10-03 09:11:33 -0700986 /* Reserve delalloc blocks for regular writeback. */
Christoph Hellwig51446f52016-09-19 11:10:21 +1000987 return xfs_file_iomap_begin_delay(inode, offset, length, flags,
988 iomap);
989 }
990
Christoph Hellwigacdda3a2016-11-30 14:37:15 +1100991 if (need_excl_ilock(ip, flags)) {
Christoph Hellwig3ba020b2016-10-20 15:53:50 +1100992 lockmode = XFS_ILOCK_EXCL;
993 xfs_ilock(ip, XFS_ILOCK_EXCL);
994 } else {
995 lockmode = xfs_ilock_data_map_shared(ip);
996 }
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +1000997
998 ASSERT(offset <= mp->m_super->s_maxbytes);
999 if ((xfs_fsize_t)offset + length > mp->m_super->s_maxbytes)
1000 length = mp->m_super->s_maxbytes - offset;
1001 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1002 end_fsb = XFS_B_TO_FSB(mp, offset + length);
1003
1004 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
Darrick J. Wongdb1327b2016-10-03 09:11:36 -07001005 &nimaps, 0);
Christoph Hellwig3ba020b2016-10-20 15:53:50 +11001006 if (error)
1007 goto out_unlock;
Darrick J. Wongdb1327b2016-10-03 09:11:36 -07001008
Christoph Hellwig3c68d442017-02-06 10:51:03 -08001009 if (flags & IOMAP_REPORT) {
Christoph Hellwig5f9268c2016-10-20 15:53:32 +11001010 /* Trim the mapping to the nearest shared extent boundary. */
1011 error = xfs_reflink_trim_around_shared(ip, &imap, &shared,
1012 &trimmed);
Christoph Hellwig3ba020b2016-10-20 15:53:50 +11001013 if (error)
1014 goto out_unlock;
1015 }
1016
1017 if ((flags & (IOMAP_WRITE | IOMAP_ZERO)) && xfs_is_reflink_inode(ip)) {
Christoph Hellwig3c68d442017-02-06 10:51:03 -08001018 if (flags & IOMAP_DIRECT) {
1019 /* may drop and re-acquire the ilock */
1020 error = xfs_reflink_allocate_cow(ip, &imap, &shared,
1021 &lockmode);
1022 if (error)
1023 goto out_unlock;
1024 } else {
1025 error = xfs_reflink_reserve_cow(ip, &imap, &shared);
1026 if (error)
1027 goto out_unlock;
1028 }
Christoph Hellwig3ba020b2016-10-20 15:53:50 +11001029
1030 end_fsb = imap.br_startoff + imap.br_blockcount;
1031 length = XFS_FSB_TO_B(mp, end_fsb) - offset;
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +10001032 }
1033
Christoph Hellwig6c31f492016-09-19 11:28:38 +10001034 if ((flags & IOMAP_WRITE) && imap_needs_alloc(inode, &imap, nimaps)) {
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +10001035 /*
1036 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
1037 * pages to keep the chunks of work done where somewhat symmetric
1038 * with the work writeback does. This is a completely arbitrary
1039 * number pulled out of thin air as a best guess for initial
1040 * testing.
1041 *
1042 * Note that the values needs to be less than 32-bits wide until
1043 * the lower level functions are updated.
1044 */
1045 length = min_t(loff_t, length, 1024 * PAGE_SIZE);
Christoph Hellwig51446f52016-09-19 11:10:21 +10001046 /*
1047 * xfs_iomap_write_direct() expects the shared lock. It
1048 * is unlocked on return.
1049 */
Christoph Hellwig66642c52016-09-19 11:26:39 +10001050 if (lockmode == XFS_ILOCK_EXCL)
1051 xfs_ilock_demote(ip, lockmode);
Christoph Hellwig51446f52016-09-19 11:10:21 +10001052 error = xfs_iomap_write_direct(ip, offset, length, &imap,
1053 nimaps);
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +10001054 if (error)
1055 return error;
1056
Christoph Hellwigecd50722016-09-19 11:24:37 +10001057 iomap->flags = IOMAP_F_NEW;
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +10001058 trace_xfs_iomap_alloc(ip, offset, length, 0, &imap);
Christoph Hellwigb95a2122016-08-17 08:44:52 +10001059 } else {
1060 ASSERT(nimaps);
1061
Christoph Hellwig66642c52016-09-19 11:26:39 +10001062 xfs_iunlock(ip, lockmode);
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +10001063 trace_xfs_iomap_found(ip, offset, length, 0, &imap);
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +10001064 }
1065
Christoph Hellwigb95a2122016-08-17 08:44:52 +10001066 xfs_bmbt_to_iomap(ip, iomap, &imap);
Dan Williamsfa5d9322017-01-27 12:04:59 -08001067
1068 /* optionally associate a dax device with the iomap bdev */
1069 bdev = iomap->bdev;
1070 if (blk_queue_dax(bdev->bd_queue))
1071 iomap->dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
1072 else
1073 iomap->dax_dev = NULL;
1074
Darrick J. Wongdb1327b2016-10-03 09:11:36 -07001075 if (shared)
1076 iomap->flags |= IOMAP_F_SHARED;
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +10001077 return 0;
Christoph Hellwig3ba020b2016-10-20 15:53:50 +11001078out_unlock:
1079 xfs_iunlock(ip, lockmode);
1080 return error;
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +10001081}
1082
1083static int
1084xfs_file_iomap_end_delalloc(
1085 struct xfs_inode *ip,
1086 loff_t offset,
1087 loff_t length,
Brian Fosterf65e6fa2017-03-08 09:58:08 -08001088 ssize_t written,
1089 struct iomap *iomap)
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +10001090{
1091 struct xfs_mount *mp = ip->i_mount;
1092 xfs_fileoff_t start_fsb;
1093 xfs_fileoff_t end_fsb;
1094 int error = 0;
1095
Brian Fosterf65e6fa2017-03-08 09:58:08 -08001096 /*
1097 * Behave as if the write failed if drop writes is enabled. Set the NEW
1098 * flag to force delalloc cleanup.
1099 */
1100 if (xfs_mp_drop_writes(mp)) {
1101 iomap->flags |= IOMAP_F_NEW;
Brian Foster9dbddd72017-02-13 22:48:17 -08001102 written = 0;
Brian Fosterf65e6fa2017-03-08 09:58:08 -08001103 }
Brian Foster9dbddd72017-02-13 22:48:17 -08001104
Brian Fosterfa7f1382017-02-16 17:19:12 -08001105 /*
1106 * start_fsb refers to the first unused block after a short write. If
1107 * nothing was written, round offset down to point at the first block in
1108 * the range.
1109 */
1110 if (unlikely(!written))
1111 start_fsb = XFS_B_TO_FSBT(mp, offset);
1112 else
1113 start_fsb = XFS_B_TO_FSB(mp, offset + written);
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +10001114 end_fsb = XFS_B_TO_FSB(mp, offset + length);
1115
1116 /*
Brian Fosterf65e6fa2017-03-08 09:58:08 -08001117 * Trim delalloc blocks if they were allocated by this write and we
1118 * didn't manage to write the whole range.
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +10001119 *
1120 * We don't need to care about racing delalloc as we hold i_mutex
1121 * across the reserve/allocate/unreserve calls. If there are delalloc
1122 * blocks in the range, they are ours.
1123 */
Brian Fosterf65e6fa2017-03-08 09:58:08 -08001124 if ((iomap->flags & IOMAP_F_NEW) && start_fsb < end_fsb) {
Brian Fosterfa7f1382017-02-16 17:19:12 -08001125 truncate_pagecache_range(VFS_I(ip), XFS_FSB_TO_B(mp, start_fsb),
1126 XFS_FSB_TO_B(mp, end_fsb) - 1);
1127
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +10001128 xfs_ilock(ip, XFS_ILOCK_EXCL);
1129 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1130 end_fsb - start_fsb);
1131 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1132
1133 if (error && !XFS_FORCED_SHUTDOWN(mp)) {
1134 xfs_alert(mp, "%s: unable to clean up ino %lld",
1135 __func__, ip->i_ino);
1136 return error;
1137 }
1138 }
1139
1140 return 0;
1141}
1142
1143static int
1144xfs_file_iomap_end(
1145 struct inode *inode,
1146 loff_t offset,
1147 loff_t length,
1148 ssize_t written,
1149 unsigned flags,
1150 struct iomap *iomap)
1151{
Dan Williamsfa5d9322017-01-27 12:04:59 -08001152 put_dax(iomap->dax_dev);
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +10001153 if ((flags & IOMAP_WRITE) && iomap->type == IOMAP_DELALLOC)
1154 return xfs_file_iomap_end_delalloc(XFS_I(inode), offset,
Brian Fosterf65e6fa2017-03-08 09:58:08 -08001155 length, written, iomap);
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +10001156 return 0;
1157}
1158
Christoph Hellwig8ff6daa2017-01-27 23:20:26 -08001159const struct iomap_ops xfs_iomap_ops = {
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +10001160 .iomap_begin = xfs_file_iomap_begin,
1161 .iomap_end = xfs_file_iomap_end,
1162};
Christoph Hellwig1d4795e2016-08-17 08:45:30 +10001163
1164static int
1165xfs_xattr_iomap_begin(
1166 struct inode *inode,
1167 loff_t offset,
1168 loff_t length,
1169 unsigned flags,
1170 struct iomap *iomap)
1171{
1172 struct xfs_inode *ip = XFS_I(inode);
1173 struct xfs_mount *mp = ip->i_mount;
1174 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1175 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length);
1176 struct xfs_bmbt_irec imap;
1177 int nimaps = 1, error = 0;
1178 unsigned lockmode;
1179
1180 if (XFS_FORCED_SHUTDOWN(mp))
1181 return -EIO;
1182
Darrick J. Wong84358536d2017-04-06 16:00:39 -07001183 lockmode = xfs_ilock_attr_map_shared(ip);
Christoph Hellwig1d4795e2016-08-17 08:45:30 +10001184
1185 /* if there are no attribute fork or extents, return ENOENT */
Darrick J. Wong84358536d2017-04-06 16:00:39 -07001186 if (!XFS_IFORK_Q(ip) || !ip->i_d.di_anextents) {
Christoph Hellwig1d4795e2016-08-17 08:45:30 +10001187 error = -ENOENT;
1188 goto out_unlock;
1189 }
1190
1191 ASSERT(ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL);
1192 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1193 &nimaps, XFS_BMAPI_ENTIRE | XFS_BMAPI_ATTRFORK);
1194out_unlock:
1195 xfs_iunlock(ip, lockmode);
1196
1197 if (!error) {
1198 ASSERT(nimaps);
1199 xfs_bmbt_to_iomap(ip, iomap, &imap);
1200 }
1201
1202 return error;
1203}
1204
Christoph Hellwig8ff6daa2017-01-27 23:20:26 -08001205const struct iomap_ops xfs_xattr_iomap_ops = {
Christoph Hellwig1d4795e2016-08-17 08:45:30 +10001206 .iomap_begin = xfs_xattr_iomap_begin,
1207};