blob: 8f8aaee7f3791457c9ec43ea46927bc05f624f04 [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.
Nathan Scott7b718762005-11-02 14:58:39 +11003 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_log.h"
21#include "xfs_trans.h"
22#include "xfs_sb.h"
23#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_quota.h"
26#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110028#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_dinode.h"
31#include "xfs_inode.h"
Christoph Hellwig84803fb2012-02-29 09:53:50 +000032#include "xfs_inode_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110033#include "xfs_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_bmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_rtalloc.h"
36#include "xfs_error.h"
37#include "xfs_itable.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include "xfs_attr.h"
39#include "xfs_buf_item.h"
40#include "xfs_trans_space.h"
41#include "xfs_utils.h"
42#include "xfs_iomap.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000043#include "xfs_trace.h"
Brian Foster27b52862012-11-06 09:50:38 -050044#include "xfs_icache.h"
Brian Foster76a42022013-03-18 10:51:47 -040045#include "xfs_dquot_item.h"
46#include "xfs_dquot.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
49#define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \
50 << mp->m_writeio_log)
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP
52
Linus Torvalds1da177e2005-04-16 15:20:36 -070053STATIC int
Nathan Scottdd9f4382006-01-11 15:28:28 +110054xfs_iomap_eof_align_last_fsb(
55 xfs_mount_t *mp,
Lachlan McIlroy541d7d32007-10-11 17:34:33 +100056 xfs_inode_t *ip,
Nathan Scottdd9f4382006-01-11 15:28:28 +110057 xfs_extlen_t extsize,
58 xfs_fileoff_t *last_fsb)
59{
60 xfs_fileoff_t new_last_fsb = 0;
Christoph Hellwigbf322d92011-12-18 20:00:05 +000061 xfs_extlen_t align = 0;
Nathan Scottdd9f4382006-01-11 15:28:28 +110062 int eof, error;
63
Christoph Hellwigbf322d92011-12-18 20:00:05 +000064 if (!XFS_IS_REALTIME_INODE(ip)) {
65 /*
66 * Round up the allocation request to a stripe unit
67 * (m_dalign) boundary if the file size is >= stripe unit
68 * size, and we are allocating past the allocation eof.
69 *
70 * If mounted with the "-o swalloc" option the alignment is
71 * increased from the strip unit size to the stripe width.
72 */
73 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
74 align = mp->m_swidth;
75 else if (mp->m_dalign)
76 align = mp->m_dalign;
77
Christoph Hellwigce7ae1512011-12-18 20:00:11 +000078 if (align && XFS_ISIZE(ip) >= XFS_FSB_TO_B(mp, align))
Christoph Hellwigbf322d92011-12-18 20:00:05 +000079 new_last_fsb = roundup_64(*last_fsb, align);
80 }
Nathan Scottdd9f4382006-01-11 15:28:28 +110081
82 /*
83 * Always round up the allocation request to an extent boundary
84 * (when file on a real-time subvolume or has di_extsize hint).
85 */
86 if (extsize) {
87 if (new_last_fsb)
88 align = roundup_64(new_last_fsb, extsize);
89 else
90 align = extsize;
91 new_last_fsb = roundup_64(*last_fsb, align);
92 }
93
94 if (new_last_fsb) {
Lachlan McIlroy541d7d32007-10-11 17:34:33 +100095 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
Nathan Scottdd9f4382006-01-11 15:28:28 +110096 if (error)
97 return error;
98 if (eof)
99 *last_fsb = new_last_fsb;
100 }
101 return 0;
102}
103
104STATIC int
Dave Chinner6d4a8ec2011-03-07 10:06:35 +1100105xfs_alert_fsblock_zero(
Nathan Scott572d95f2006-09-28 11:03:20 +1000106 xfs_inode_t *ip,
107 xfs_bmbt_irec_t *imap)
108{
Dave Chinner6a19d932011-03-07 10:02:35 +1100109 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
Nathan Scott572d95f2006-09-28 11:03:20 +1000110 "Access to block zero in inode %llu "
111 "start_block: %llx start_off: %llx "
112 "blkcnt: %llx extent-state: %x\n",
113 (unsigned long long)ip->i_ino,
114 (unsigned long long)imap->br_startblock,
115 (unsigned long long)imap->br_startoff,
116 (unsigned long long)imap->br_blockcount,
117 imap->br_state);
118 return EFSCORRUPTED;
119}
120
Christoph Hellwiga206c812010-12-10 08:42:20 +0000121int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122xfs_iomap_write_direct(
123 xfs_inode_t *ip,
Nathan Scottf403b7f2005-05-05 13:33:40 -0700124 xfs_off_t offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 size_t count,
Christoph Hellwig30704512010-06-24 11:42:19 +1000126 xfs_bmbt_irec_t *imap,
Christoph Hellwig405f8042010-12-10 08:42:19 +0000127 int nmaps)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128{
129 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 xfs_fileoff_t offset_fsb;
131 xfs_fileoff_t last_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100132 xfs_filblks_t count_fsb, resaligned;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 xfs_fsblock_t firstfsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100134 xfs_extlen_t extsz, temp;
Eric Sandeen0116d932005-11-02 15:00:01 +1100135 int nimaps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 int bmapi_flag;
Nathan Scott06d10dd2005-06-21 15:48:47 +1000137 int quota_flag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 int rt;
139 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 xfs_bmap_free_t free_list;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100141 uint qblocks, resblks, resrtextents;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 int committed;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100143 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
Dave Chinner507630b2012-03-27 10:34:50 -0400145 error = xfs_qm_dqattach(ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 if (error)
147 return XFS_ERROR(error);
148
Nathan Scottdd9f4382006-01-11 15:28:28 +1100149 rt = XFS_IS_REALTIME_INODE(ip);
David Chinner957d0eb2007-06-18 16:50:37 +1000150 extsz = xfs_get_extsz_hint(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
David Chinner957d0eb2007-06-18 16:50:37 +1000152 offset_fsb = XFS_B_TO_FSBT(mp, offset);
153 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000154 if ((offset + count) > XFS_ISIZE(ip)) {
Lachlan McIlroy9f6c92b2008-12-22 17:56:49 +1100155 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100156 if (error)
Dave Chinner507630b2012-03-27 10:34:50 -0400157 return XFS_ERROR(error);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100158 } else {
Christoph Hellwig405f8042010-12-10 08:42:19 +0000159 if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
Nathan Scottdd9f4382006-01-11 15:28:28 +1100160 last_fsb = MIN(last_fsb, (xfs_fileoff_t)
Christoph Hellwig30704512010-06-24 11:42:19 +1000161 imap->br_blockcount +
162 imap->br_startoff);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100163 }
164 count_fsb = last_fsb - offset_fsb;
165 ASSERT(count_fsb > 0);
166
167 resaligned = count_fsb;
168 if (unlikely(extsz)) {
169 if ((temp = do_mod(offset_fsb, extsz)))
170 resaligned += temp;
171 if ((temp = do_mod(resaligned, extsz)))
172 resaligned += extsz - temp;
173 }
174
175 if (unlikely(rt)) {
176 resrtextents = qblocks = resaligned;
177 resrtextents /= mp->m_sb.sb_rextsize;
David Chinner84e1e992007-06-18 16:50:27 +1000178 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
179 quota_flag = XFS_QMOPT_RES_RTBLKS;
180 } else {
181 resrtextents = 0;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100182 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
David Chinner84e1e992007-06-18 16:50:27 +1000183 quota_flag = XFS_QMOPT_RES_REGBLKS;
184 }
Nathan Scottdd9f4382006-01-11 15:28:28 +1100185
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 /*
Nathan Scott06d10dd2005-06-21 15:48:47 +1000187 * Allocate and setup the transaction
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 error = xfs_trans_reserve(tp, resblks,
Nathan Scottd52b44d2005-09-02 16:41:32 +1000191 XFS_WRITE_LOG_RES(mp), resrtextents,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 XFS_TRANS_PERM_LOG_RES,
193 XFS_WRITE_LOG_COUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 /*
Nathan Scott06d10dd2005-06-21 15:48:47 +1000195 * Check for running out of space, note: need lock to return
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 */
Dave Chinner507630b2012-03-27 10:34:50 -0400197 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 xfs_trans_cancel(tp, 0);
Dave Chinner507630b2012-03-27 10:34:50 -0400199 return XFS_ERROR(error);
200 }
201
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 xfs_ilock(ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
Christoph Hellwig7d095252009-06-08 15:33:32 +0200204 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100205 if (error)
Dave Chinner507630b2012-03-27 10:34:50 -0400206 goto out_trans_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207
Christoph Hellwigddc34152011-09-19 15:00:54 +0000208 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
Dave Chinnerc0dc7822011-09-18 20:40:52 +0000210 bmapi_flag = 0;
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000211 if (offset < XFS_ISIZE(ip) || extsz)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 bmapi_flag |= XFS_BMAPI_PREALLOC;
213
214 /*
Christoph Hellwig30704512010-06-24 11:42:19 +1000215 * From this point onwards we overwrite the imap pointer that the
216 * caller gave to us.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 */
Eric Sandeen9d87c312009-01-14 23:22:07 -0600218 xfs_bmap_init(&free_list, &firstfsb);
Nathan Scott06d10dd2005-06-21 15:48:47 +1000219 nimaps = 1;
Dave Chinnerc0dc7822011-09-18 20:40:52 +0000220 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flag,
221 &firstfsb, 0, imap, &nimaps, &free_list);
Nathan Scott06d10dd2005-06-21 15:48:47 +1000222 if (error)
Dave Chinner507630b2012-03-27 10:34:50 -0400223 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224
225 /*
Nathan Scott06d10dd2005-06-21 15:48:47 +1000226 * Complete the transaction
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +1100228 error = xfs_bmap_finish(&tp, &free_list, &committed);
Nathan Scott06d10dd2005-06-21 15:48:47 +1000229 if (error)
Dave Chinner507630b2012-03-27 10:34:50 -0400230 goto out_bmap_cancel;
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000231 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Nathan Scott06d10dd2005-06-21 15:48:47 +1000232 if (error)
Dave Chinner507630b2012-03-27 10:34:50 -0400233 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
Nathan Scott06d10dd2005-06-21 15:48:47 +1000235 /*
236 * Copy any maps to caller's array and return any error.
237 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 if (nimaps == 0) {
Dave Chinner507630b2012-03-27 10:34:50 -0400239 error = XFS_ERROR(ENOSPC);
240 goto out_unlock;
Nathan Scott572d95f2006-09-28 11:03:20 +1000241 }
242
Dave Chinner507630b2012-03-27 10:34:50 -0400243 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
Dave Chinner6d4a8ec2011-03-07 10:06:35 +1100244 error = xfs_alert_fsblock_zero(ip, imap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
Dave Chinner507630b2012-03-27 10:34:50 -0400246out_unlock:
247 xfs_iunlock(ip, XFS_ILOCK_EXCL);
248 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
Dave Chinner507630b2012-03-27 10:34:50 -0400250out_bmap_cancel:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 xfs_bmap_cancel(&free_list);
Dave Chinnerea562ed2012-05-08 20:48:53 +1000252 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
Dave Chinner507630b2012-03-27 10:34:50 -0400253out_trans_cancel:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
Dave Chinner507630b2012-03-27 10:34:50 -0400255 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256}
257
Nathan Scottdd9f4382006-01-11 15:28:28 +1100258/*
Dave Chinner8de2bf92009-04-06 18:49:12 +0200259 * If the caller is doing a write at the end of the file, then extend the
260 * allocation out to the file system's write iosize. We clean up any extra
261 * space left over when the file is closed in xfs_inactive().
Dave Chinner055388a2011-01-04 11:35:03 +1100262 *
263 * If we find we already have delalloc preallocation beyond EOF, don't do more
264 * preallocation as it it not needed.
Nathan Scottdd9f4382006-01-11 15:28:28 +1100265 */
266STATIC int
267xfs_iomap_eof_want_preallocate(
268 xfs_mount_t *mp,
Lachlan McIlroy541d7d32007-10-11 17:34:33 +1000269 xfs_inode_t *ip,
Nathan Scottdd9f4382006-01-11 15:28:28 +1100270 xfs_off_t offset,
271 size_t count,
Nathan Scottdd9f4382006-01-11 15:28:28 +1100272 xfs_bmbt_irec_t *imap,
273 int nimaps,
274 int *prealloc)
275{
276 xfs_fileoff_t start_fsb;
277 xfs_filblks_t count_fsb;
278 xfs_fsblock_t firstblock;
279 int n, error, imaps;
Dave Chinner055388a2011-01-04 11:35:03 +1100280 int found_delalloc = 0;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100281
282 *prealloc = 0;
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000283 if (offset + count <= XFS_ISIZE(ip))
Nathan Scottdd9f4382006-01-11 15:28:28 +1100284 return 0;
285
286 /*
287 * If there are any real blocks past eof, then don't
288 * do any speculative allocation.
289 */
290 start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
Dave Chinner32972382012-06-08 15:44:54 +1000291 count_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100292 while (count_fsb > 0) {
293 imaps = nimaps;
Nathan Scott3ddb8fa2006-01-11 15:33:02 +1100294 firstblock = NULLFSBLOCK;
Dave Chinner5c8ed202011-09-18 20:40:45 +0000295 error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps,
296 0);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100297 if (error)
298 return error;
299 for (n = 0; n < imaps; n++) {
300 if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
301 (imap[n].br_startblock != DELAYSTARTBLOCK))
302 return 0;
303 start_fsb += imap[n].br_blockcount;
304 count_fsb -= imap[n].br_blockcount;
Dave Chinner055388a2011-01-04 11:35:03 +1100305
306 if (imap[n].br_startblock == DELAYSTARTBLOCK)
307 found_delalloc = 1;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100308 }
309 }
Dave Chinner055388a2011-01-04 11:35:03 +1100310 if (!found_delalloc)
311 *prealloc = 1;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100312 return 0;
313}
314
Dave Chinner055388a2011-01-04 11:35:03 +1100315/*
Dave Chinnera1e16c22013-02-11 16:05:01 +1100316 * Determine the initial size of the preallocation. We are beyond the current
317 * EOF here, but we need to take into account whether this is a sparse write or
318 * an extending write when determining the preallocation size. Hence we need to
319 * look up the extent that ends at the current write offset and use the result
320 * to determine the preallocation size.
321 *
322 * If the extent is a hole, then preallocation is essentially disabled.
323 * Otherwise we take the size of the preceeding data extent as the basis for the
324 * preallocation size. If the size of the extent is greater than half the
325 * maximum extent length, then use the current offset as the basis. This ensures
326 * that for large files the preallocation size always extends to MAXEXTLEN
327 * rather than falling short due to things like stripe unit/width alignment of
328 * real extents.
329 */
Mark Tinguelye8108ce2013-02-24 13:04:37 -0600330STATIC xfs_fsblock_t
Dave Chinnera1e16c22013-02-11 16:05:01 +1100331xfs_iomap_eof_prealloc_initial_size(
332 struct xfs_mount *mp,
333 struct xfs_inode *ip,
334 xfs_off_t offset,
335 xfs_bmbt_irec_t *imap,
336 int nimaps)
337{
338 xfs_fileoff_t start_fsb;
339 int imaps = 1;
340 int error;
341
342 ASSERT(nimaps >= imaps);
343
344 /* if we are using a specific prealloc size, return now */
345 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
346 return 0;
347
348 /*
349 * As we write multiple pages, the offset will always align to the
350 * start of a page and hence point to a hole at EOF. i.e. if the size is
351 * 4096 bytes, we only have one block at FSB 0, but XFS_B_TO_FSB(4096)
352 * will return FSB 1. Hence if there are blocks in the file, we want to
353 * point to the block prior to the EOF block and not the hole that maps
354 * directly at @offset.
355 */
356 start_fsb = XFS_B_TO_FSB(mp, offset);
357 if (start_fsb)
358 start_fsb--;
359 error = xfs_bmapi_read(ip, start_fsb, 1, imap, &imaps, XFS_BMAPI_ENTIRE);
360 if (error)
361 return 0;
362
363 ASSERT(imaps == 1);
364 if (imap[0].br_startblock == HOLESTARTBLOCK)
365 return 0;
366 if (imap[0].br_blockcount <= (MAXEXTLEN >> 1))
Brian Fostere114b5f2013-02-19 10:24:41 -0500367 return imap[0].br_blockcount << 1;
Dave Chinnera1e16c22013-02-11 16:05:01 +1100368 return XFS_B_TO_FSB(mp, offset);
369}
370
Brian Foster76a42022013-03-18 10:51:47 -0400371STATIC bool
372xfs_quota_need_throttle(
373 struct xfs_inode *ip,
374 int type,
375 xfs_fsblock_t alloc_blocks)
376{
377 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
378
379 if (!dq || !xfs_this_quota_on(ip->i_mount, type))
380 return false;
381
382 /* no hi watermark, no throttle */
383 if (!dq->q_prealloc_hi_wmark)
384 return false;
385
386 /* under the lo watermark, no throttle */
387 if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark)
388 return false;
389
390 return true;
391}
392
393STATIC void
394xfs_quota_calc_throttle(
395 struct xfs_inode *ip,
396 int type,
397 xfs_fsblock_t *qblocks,
398 int *qshift)
399{
400 int64_t freesp;
401 int shift = 0;
402 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
403
404 /* over hi wmark, squash the prealloc completely */
405 if (dq->q_res_bcount >= dq->q_prealloc_hi_wmark) {
406 *qblocks = 0;
407 return;
408 }
409
410 freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount;
411 if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
412 shift = 2;
413 if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
414 shift += 2;
415 if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
416 shift += 2;
417 }
418
419 /* only overwrite the throttle values if we are more aggressive */
420 if ((freesp >> shift) < (*qblocks >> *qshift)) {
421 *qblocks = freesp;
422 *qshift = shift;
423 }
424}
425
Dave Chinnera1e16c22013-02-11 16:05:01 +1100426/*
Dave Chinner055388a2011-01-04 11:35:03 +1100427 * If we don't have a user specified preallocation size, dynamically increase
428 * the preallocation size as the size of the file grows. Cap the maximum size
429 * at a single extent or less if the filesystem is near full. The closer the
430 * filesystem is to full, the smaller the maximum prealocation.
431 */
432STATIC xfs_fsblock_t
433xfs_iomap_prealloc_size(
434 struct xfs_mount *mp,
Dave Chinnera1e16c22013-02-11 16:05:01 +1100435 struct xfs_inode *ip,
436 xfs_off_t offset,
437 struct xfs_bmbt_irec *imap,
438 int nimaps)
Dave Chinner055388a2011-01-04 11:35:03 +1100439{
440 xfs_fsblock_t alloc_blocks = 0;
Brian Foster3c58b5f2013-03-18 10:51:43 -0400441 int shift = 0;
442 int64_t freesp;
Brian Foster76a42022013-03-18 10:51:47 -0400443 xfs_fsblock_t qblocks;
444 int qshift = 0;
Dave Chinner055388a2011-01-04 11:35:03 +1100445
Dave Chinnera1e16c22013-02-11 16:05:01 +1100446 alloc_blocks = xfs_iomap_eof_prealloc_initial_size(mp, ip, offset,
447 imap, nimaps);
Brian Foster3c58b5f2013-03-18 10:51:43 -0400448 if (!alloc_blocks)
449 goto check_writeio;
Brian Foster76a42022013-03-18 10:51:47 -0400450 qblocks = alloc_blocks;
Dave Chinner055388a2011-01-04 11:35:03 +1100451
Brian Fosterc9bdbdc2013-03-18 10:51:44 -0400452 /*
453 * MAXEXTLEN is not a power of two value but we round the prealloc down
454 * to the nearest power of two value after throttling. To prevent the
455 * round down from unconditionally reducing the maximum supported prealloc
456 * size, we round up first, apply appropriate throttling, round down and
457 * cap the value to MAXEXTLEN.
458 */
459 alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
460 alloc_blocks);
Dave Chinner055388a2011-01-04 11:35:03 +1100461
Brian Foster3c58b5f2013-03-18 10:51:43 -0400462 xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
463 freesp = mp->m_sb.sb_fdblocks;
464 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
465 shift = 2;
466 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
467 shift++;
468 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
469 shift++;
470 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
471 shift++;
472 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
473 shift++;
Dave Chinner055388a2011-01-04 11:35:03 +1100474 }
Brian Foster76a42022013-03-18 10:51:47 -0400475
476 /*
477 * Check each quota to cap the prealloc size and provide a shift
478 * value to throttle with.
479 */
480 if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks))
481 xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift);
482 if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks))
483 xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift);
484 if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks))
485 xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift);
486
487 /*
488 * The final prealloc size is set to the minimum of free space available
489 * in each of the quotas and the overall filesystem.
490 *
491 * The shift throttle value is set to the maximum value as determined by
492 * the global low free space values and per-quota low free space values.
493 */
494 alloc_blocks = MIN(alloc_blocks, qblocks);
495 shift = MAX(shift, qshift);
496
Brian Foster3c58b5f2013-03-18 10:51:43 -0400497 if (shift)
498 alloc_blocks >>= shift;
Brian Fosterc9bdbdc2013-03-18 10:51:44 -0400499 /*
500 * rounddown_pow_of_two() returns an undefined result if we pass in
501 * alloc_blocks = 0.
502 */
503 if (alloc_blocks)
504 alloc_blocks = rounddown_pow_of_two(alloc_blocks);
505 if (alloc_blocks > MAXEXTLEN)
506 alloc_blocks = MAXEXTLEN;
Dave Chinner055388a2011-01-04 11:35:03 +1100507
Brian Foster3c58b5f2013-03-18 10:51:43 -0400508 /*
509 * If we are still trying to allocate more space than is
510 * available, squash the prealloc hard. This can happen if we
511 * have a large file on a small filesystem and the above
512 * lowspace thresholds are smaller than MAXEXTLEN.
513 */
514 while (alloc_blocks && alloc_blocks >= freesp)
515 alloc_blocks >>= 4;
516
517check_writeio:
Dave Chinner055388a2011-01-04 11:35:03 +1100518 if (alloc_blocks < mp->m_writeio_blocks)
519 alloc_blocks = mp->m_writeio_blocks;
520
Brian Foster19cb7e32013-03-18 10:51:48 -0400521 trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
522 mp->m_writeio_blocks);
523
Dave Chinner055388a2011-01-04 11:35:03 +1100524 return alloc_blocks;
525}
526
Christoph Hellwiga206c812010-12-10 08:42:20 +0000527int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528xfs_iomap_write_delay(
529 xfs_inode_t *ip,
Nathan Scottf403b7f2005-05-05 13:33:40 -0700530 xfs_off_t offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 size_t count,
Christoph Hellwig405f8042010-12-10 08:42:19 +0000532 xfs_bmbt_irec_t *ret_imap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533{
534 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 xfs_fileoff_t offset_fsb;
536 xfs_fileoff_t last_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100537 xfs_off_t aligned_offset;
538 xfs_fileoff_t ioalign;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100539 xfs_extlen_t extsz;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 int nimaps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
Dave Chinner9aa05002012-10-08 21:56:04 +1100542 int prealloc;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100543 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000545 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
547 /*
548 * Make sure that the dquots are there. This doesn't hold
549 * the ilock across a disk read.
550 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200551 error = xfs_qm_dqattach_locked(ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 if (error)
553 return XFS_ERROR(error);
554
David Chinner957d0eb2007-06-18 16:50:37 +1000555 extsz = xfs_get_extsz_hint(ip);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100556 offset_fsb = XFS_B_TO_FSBT(mp, offset);
557
Lachlan McIlroy9f6c92b2008-12-22 17:56:49 +1100558 error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
Christoph Hellwig405f8042010-12-10 08:42:19 +0000559 imap, XFS_WRITE_IMAPS, &prealloc);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100560 if (error)
561 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562
Dave Chinner8de2bf92009-04-06 18:49:12 +0200563retry:
Nathan Scottdd9f4382006-01-11 15:28:28 +1100564 if (prealloc) {
Dave Chinnera1e16c22013-02-11 16:05:01 +1100565 xfs_fsblock_t alloc_blocks;
566
567 alloc_blocks = xfs_iomap_prealloc_size(mp, ip, offset, imap,
568 XFS_WRITE_IMAPS);
Dave Chinner055388a2011-01-04 11:35:03 +1100569
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
571 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
Dave Chinner055388a2011-01-04 11:35:03 +1100572 last_fsb = ioalign + alloc_blocks;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100573 } else {
574 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 }
Nathan Scottdd9f4382006-01-11 15:28:28 +1100576
577 if (prealloc || extsz) {
Lachlan McIlroy9f6c92b2008-12-22 17:56:49 +1100578 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100579 if (error)
580 return error;
581 }
582
Dave Chinner3ed91162012-04-29 22:43:19 +1000583 /*
584 * Make sure preallocation does not create extents beyond the range we
585 * actually support in this filesystem.
586 */
Dave Chinnerd2c28192012-06-08 15:44:53 +1000587 if (last_fsb > XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes))
588 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
Dave Chinner3ed91162012-04-29 22:43:19 +1000589
590 ASSERT(last_fsb > offset_fsb);
591
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 nimaps = XFS_WRITE_IMAPS;
Christoph Hellwig44032802011-09-18 20:40:48 +0000593 error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb,
594 imap, &nimaps, XFS_BMAPI_ENTIRE);
Dave Chinner055388a2011-01-04 11:35:03 +1100595 switch (error) {
596 case 0:
597 case ENOSPC:
598 case EDQUOT:
599 break;
600 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 return XFS_ERROR(error);
Dave Chinner055388a2011-01-04 11:35:03 +1100602 }
Nathan Scottdd9f4382006-01-11 15:28:28 +1100603
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 /*
Dave Chinner9aa05002012-10-08 21:56:04 +1100605 * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. Retry
Dave Chinner055388a2011-01-04 11:35:03 +1100606 * without EOF preallocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 */
608 if (nimaps == 0) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000609 trace_xfs_delalloc_enospc(ip, offset, count);
Dave Chinner9aa05002012-10-08 21:56:04 +1100610 if (prealloc) {
611 prealloc = 0;
612 error = 0;
613 goto retry;
Dave Chinner055388a2011-01-04 11:35:03 +1100614 }
Dave Chinner9aa05002012-10-08 21:56:04 +1100615 return XFS_ERROR(error ? error : ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 }
617
David Chinner86c4d622008-04-29 12:53:21 +1000618 if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
Dave Chinner6d4a8ec2011-03-07 10:06:35 +1100619 return xfs_alert_fsblock_zero(ip, &imap[0]);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100620
Brian Foster27b52862012-11-06 09:50:38 -0500621 /*
622 * Tag the inode as speculatively preallocated so we can reclaim this
623 * space on demand, if necessary.
624 */
625 if (prealloc)
626 xfs_inode_set_eofblocks_tag(ip);
627
Nathan Scottdd9f4382006-01-11 15:28:28 +1100628 *ret_imap = imap[0];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 return 0;
630}
631
632/*
633 * Pass in a delayed allocate extent, convert it to real extents;
634 * return to the caller the extent we create which maps on top of
635 * the originating callers request.
636 *
637 * Called without a lock on the inode.
David Chinnere4143a12007-11-23 16:29:11 +1100638 *
639 * We no longer bother to look at the incoming map - all we have to
640 * guarantee is that whatever we allocate fills the required range.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 */
Christoph Hellwiga206c812010-12-10 08:42:20 +0000642int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643xfs_iomap_write_allocate(
644 xfs_inode_t *ip,
Nathan Scottf403b7f2005-05-05 13:33:40 -0700645 xfs_off_t offset,
Nathan Scott24e17b52005-05-05 13:33:20 -0700646 size_t count,
Christoph Hellwig405f8042010-12-10 08:42:19 +0000647 xfs_bmbt_irec_t *imap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648{
649 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 xfs_fileoff_t offset_fsb, last_block;
651 xfs_fileoff_t end_fsb, map_start_fsb;
652 xfs_fsblock_t first_block;
653 xfs_bmap_free_t free_list;
654 xfs_filblks_t count_fsb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 xfs_trans_t *tp;
David Chinnere4143a12007-11-23 16:29:11 +1100656 int nimaps, committed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 int error = 0;
658 int nres;
659
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 /*
661 * Make sure that the dquots are there.
662 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200663 error = xfs_qm_dqattach(ip, 0);
664 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 return XFS_ERROR(error);
666
Nathan Scott24e17b52005-05-05 13:33:20 -0700667 offset_fsb = XFS_B_TO_FSBT(mp, offset);
Christoph Hellwig30704512010-06-24 11:42:19 +1000668 count_fsb = imap->br_blockcount;
669 map_start_fsb = imap->br_startoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670
671 XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
672
673 while (count_fsb != 0) {
674 /*
675 * Set up a transaction with which to allocate the
676 * backing store for the file. Do allocations in a
677 * loop until we get some space in the range we are
678 * interested in. The other space that might be allocated
679 * is in the delayed allocation extent on which we sit
680 * but before our buffer starts.
681 */
682
683 nimaps = 0;
684 while (nimaps == 0) {
685 tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
David Chinner84e1e992007-06-18 16:50:27 +1000686 tp->t_flags |= XFS_TRANS_RESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
688 error = xfs_trans_reserve(tp, nres,
689 XFS_WRITE_LOG_RES(mp),
690 0, XFS_TRANS_PERM_LOG_RES,
691 XFS_WRITE_LOG_COUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 if (error) {
693 xfs_trans_cancel(tp, 0);
694 return XFS_ERROR(error);
695 }
696 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigddc34152011-09-19 15:00:54 +0000697 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
Eric Sandeen9d87c312009-01-14 23:22:07 -0600699 xfs_bmap_init(&free_list, &first_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 /*
David Chinnere4143a12007-11-23 16:29:11 +1100702 * it is possible that the extents have changed since
703 * we did the read call as we dropped the ilock for a
704 * while. We have to be careful about truncates or hole
705 * punchs here - we are not allowed to allocate
706 * non-delalloc blocks here.
707 *
708 * The only protection against truncation is the pages
709 * for the range we are being asked to convert are
710 * locked and hence a truncate will block on them
711 * first.
712 *
713 * As a result, if we go beyond the range we really
714 * need and hit an delalloc extent boundary followed by
715 * a hole while we have excess blocks in the map, we
716 * will fill the hole incorrectly and overrun the
717 * transaction reservation.
718 *
719 * Using a single map prevents this as we are forced to
720 * check each map we look for overlap with the desired
721 * range and abort as soon as we find it. Also, given
722 * that we only return a single map, having one beyond
723 * what we can return is probably a bit silly.
724 *
725 * We also need to check that we don't go beyond EOF;
726 * this is a truncate optimisation as a truncate sets
727 * the new file size before block on the pages we
728 * currently have locked under writeback. Because they
729 * are about to be tossed, we don't need to write them
730 * back....
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 */
David Chinnere4143a12007-11-23 16:29:11 +1100732 nimaps = 1;
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000733 end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
David Chinner7c9ef852008-04-10 12:21:59 +1000734 error = xfs_bmap_last_offset(NULL, ip, &last_block,
735 XFS_DATA_FORK);
736 if (error)
737 goto trans_cancel;
738
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
740 if ((map_start_fsb + count_fsb) > last_block) {
741 count_fsb = last_block - map_start_fsb;
742 if (count_fsb == 0) {
743 error = EAGAIN;
744 goto trans_cancel;
745 }
746 }
747
Christoph Hellwig30704512010-06-24 11:42:19 +1000748 /*
Christoph Hellwig30704512010-06-24 11:42:19 +1000749 * From this point onwards we overwrite the imap
750 * pointer that the caller gave to us.
751 */
Dave Chinnerc0dc7822011-09-18 20:40:52 +0000752 error = xfs_bmapi_write(tp, ip, map_start_fsb,
Dave Chinner24558812012-10-05 11:06:58 +1000753 count_fsb,
754 XFS_BMAPI_STACK_SWITCH,
755 &first_block, 1,
Dave Chinnerc0dc7822011-09-18 20:40:52 +0000756 imap, &nimaps, &free_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 if (error)
758 goto trans_cancel;
759
Eric Sandeenf7c99b62007-02-10 18:37:16 +1100760 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 if (error)
762 goto trans_cancel;
763
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000764 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 if (error)
766 goto error0;
767
768 xfs_iunlock(ip, XFS_ILOCK_EXCL);
769 }
770
771 /*
772 * See if we were able to allocate an extent that
773 * covers at least part of the callers request
774 */
Christoph Hellwig30704512010-06-24 11:42:19 +1000775 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
Dave Chinner6d4a8ec2011-03-07 10:06:35 +1100776 return xfs_alert_fsblock_zero(ip, imap);
David Chinner86c4d622008-04-29 12:53:21 +1000777
Christoph Hellwig30704512010-06-24 11:42:19 +1000778 if ((offset_fsb >= imap->br_startoff) &&
779 (offset_fsb < (imap->br_startoff +
780 imap->br_blockcount))) {
David Chinnere4143a12007-11-23 16:29:11 +1100781 XFS_STATS_INC(xs_xstrat_quick);
782 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 }
784
David Chinnere4143a12007-11-23 16:29:11 +1100785 /*
786 * So far we have not mapped the requested part of the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 * file, just surrounding data, try again.
788 */
Christoph Hellwig30704512010-06-24 11:42:19 +1000789 count_fsb -= imap->br_blockcount;
790 map_start_fsb = imap->br_startoff + imap->br_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 }
792
793trans_cancel:
794 xfs_bmap_cancel(&free_list);
795 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
796error0:
797 xfs_iunlock(ip, XFS_ILOCK_EXCL);
798 return XFS_ERROR(error);
799}
800
801int
802xfs_iomap_write_unwritten(
803 xfs_inode_t *ip,
Nathan Scottf403b7f2005-05-05 13:33:40 -0700804 xfs_off_t offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 size_t count)
806{
807 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 xfs_fileoff_t offset_fsb;
809 xfs_filblks_t count_fsb;
810 xfs_filblks_t numblks_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100811 xfs_fsblock_t firstfsb;
812 int nimaps;
813 xfs_trans_t *tp;
814 xfs_bmbt_irec_t imap;
815 xfs_bmap_free_t free_list;
Christoph Hellwig84803fb2012-02-29 09:53:50 +0000816 xfs_fsize_t i_size;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100817 uint resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 int committed;
819 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000821 trace_xfs_unwritten_convert(ip, offset, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822
823 offset_fsb = XFS_B_TO_FSBT(mp, offset);
824 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
825 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
826
Lachlan McIlroy4ddd8bb2008-06-27 13:32:53 +1000827 /*
828 * Reserve enough blocks in this transaction for two complete extent
829 * btree splits. We may be converting the middle part of an unwritten
830 * extent and in this case we will insert two new extents in the btree
831 * each of which could cause a full split.
832 *
833 * This reservation amount will be used in the first call to
834 * xfs_bmbt_split() to select an AG with enough space to satisfy the
835 * rest of the operation.
836 */
Nathan Scottdd9f4382006-01-11 15:28:28 +1100837 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
Nathan Scottdd9f4382006-01-11 15:28:28 +1100839 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 /*
841 * set up a transaction to convert the range of extents
842 * from unwritten to real. Do allocations in a loop until
843 * we have covered the range passed in.
Christoph Hellwig80641dc2009-10-19 04:00:03 +0000844 *
845 * Note that we open code the transaction allocation here
846 * to pass KM_NOFS--we can't risk to recursing back into
847 * the filesystem here as we might be asked to write out
848 * the same inode that we complete here and might deadlock
849 * on the iolock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 */
Jan Karad9457dc2012-06-12 16:20:39 +0200851 sb_start_intwrite(mp->m_super);
Christoph Hellwig80641dc2009-10-19 04:00:03 +0000852 tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS);
Jan Karad9457dc2012-06-12 16:20:39 +0200853 tp->t_flags |= XFS_TRANS_RESERVE | XFS_TRANS_FREEZE_PROT;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100854 error = xfs_trans_reserve(tp, resblks,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 XFS_WRITE_LOG_RES(mp), 0,
856 XFS_TRANS_PERM_LOG_RES,
857 XFS_WRITE_LOG_COUNT);
858 if (error) {
859 xfs_trans_cancel(tp, 0);
Nathan Scott572d95f2006-09-28 11:03:20 +1000860 return XFS_ERROR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 }
862
863 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigddc34152011-09-19 15:00:54 +0000864 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865
866 /*
867 * Modify the unwritten extent state of the buffer.
868 */
Eric Sandeen9d87c312009-01-14 23:22:07 -0600869 xfs_bmap_init(&free_list, &firstfsb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 nimaps = 1;
Dave Chinnerc0dc7822011-09-18 20:40:52 +0000871 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
872 XFS_BMAPI_CONVERT, &firstfsb,
Christoph Hellwigb4e91812010-06-23 18:11:15 +1000873 1, &imap, &nimaps, &free_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 if (error)
875 goto error_on_bmapi_transaction;
876
Christoph Hellwig84803fb2012-02-29 09:53:50 +0000877 /*
878 * Log the updated inode size as we go. We have to be careful
879 * to only log it up to the actual write offset if it is
880 * halfway into a block.
881 */
882 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
883 if (i_size > offset + count)
884 i_size = offset + count;
885
886 i_size = xfs_new_eof(ip, i_size);
887 if (i_size) {
888 ip->i_d.di_size = i_size;
889 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
890 }
891
892 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 if (error)
894 goto error_on_bmapi_transaction;
895
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000896 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 xfs_iunlock(ip, XFS_ILOCK_EXCL);
898 if (error)
Nathan Scott572d95f2006-09-28 11:03:20 +1000899 return XFS_ERROR(error);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100900
David Chinner86c4d622008-04-29 12:53:21 +1000901 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
Dave Chinner6d4a8ec2011-03-07 10:06:35 +1100902 return xfs_alert_fsblock_zero(ip, &imap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
904 if ((numblks_fsb = imap.br_blockcount) == 0) {
905 /*
906 * The numblks_fsb value should always get
907 * smaller, otherwise the loop is stuck.
908 */
909 ASSERT(imap.br_blockcount);
910 break;
911 }
912 offset_fsb += numblks_fsb;
913 count_fsb -= numblks_fsb;
914 } while (count_fsb > 0);
915
916 return 0;
917
918error_on_bmapi_transaction:
919 xfs_bmap_cancel(&free_list);
920 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
921 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 return XFS_ERROR(error);
923}