blob: 1f86033171c84ef4b12a201e112c9c37c7c69bdf [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"
Dave Chinner70a98832013-10-23 10:36:05 +110020#include "xfs_shared.h"
Dave Chinner239880e2013-10-23 10:50:10 +110021#include "xfs_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110026#include "xfs_btree.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110027#include "xfs_bmap_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_bmap.h"
Dave Chinner68988112013-08-12 20:49:42 +100029#include "xfs_bmap_util.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_error.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110031#include "xfs_trans.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_trans_space.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_iomap.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000034#include "xfs_trace.h"
Brian Foster27b52862012-11-06 09:50:38 -050035#include "xfs_icache.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110036#include "xfs_quota.h"
Brian Foster76a42022013-03-18 10:51:47 -040037#include "xfs_dquot_item.h"
38#include "xfs_dquot.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41#define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \
42 << mp->m_writeio_log)
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP
44
Linus Torvalds1da177e2005-04-16 15:20:36 -070045STATIC int
Nathan Scottdd9f4382006-01-11 15:28:28 +110046xfs_iomap_eof_align_last_fsb(
47 xfs_mount_t *mp,
Lachlan McIlroy541d7d32007-10-11 17:34:33 +100048 xfs_inode_t *ip,
Nathan Scottdd9f4382006-01-11 15:28:28 +110049 xfs_extlen_t extsize,
50 xfs_fileoff_t *last_fsb)
51{
Christoph Hellwigbf322d92011-12-18 20:00:05 +000052 xfs_extlen_t align = 0;
Nathan Scottdd9f4382006-01-11 15:28:28 +110053 int eof, error;
54
Christoph Hellwigbf322d92011-12-18 20:00:05 +000055 if (!XFS_IS_REALTIME_INODE(ip)) {
56 /*
57 * Round up the allocation request to a stripe unit
58 * (m_dalign) boundary if the file size is >= stripe unit
59 * size, and we are allocating past the allocation eof.
60 *
61 * If mounted with the "-o swalloc" option the alignment is
62 * increased from the strip unit size to the stripe width.
63 */
64 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
65 align = mp->m_swidth;
66 else if (mp->m_dalign)
67 align = mp->m_dalign;
68
Peter Watkins76b57302014-12-04 09:30:51 +110069 if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
70 align = 0;
Christoph Hellwigbf322d92011-12-18 20:00:05 +000071 }
Nathan Scottdd9f4382006-01-11 15:28:28 +110072
73 /*
74 * Always round up the allocation request to an extent boundary
75 * (when file on a real-time subvolume or has di_extsize hint).
76 */
77 if (extsize) {
Peter Watkins76b57302014-12-04 09:30:51 +110078 if (align)
79 align = roundup_64(align, extsize);
Nathan Scottdd9f4382006-01-11 15:28:28 +110080 else
81 align = extsize;
Nathan Scottdd9f4382006-01-11 15:28:28 +110082 }
83
Peter Watkins76b57302014-12-04 09:30:51 +110084 if (align) {
85 xfs_fileoff_t new_last_fsb = roundup_64(*last_fsb, align);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +100086 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
Nathan Scottdd9f4382006-01-11 15:28:28 +110087 if (error)
88 return error;
89 if (eof)
90 *last_fsb = new_last_fsb;
91 }
92 return 0;
93}
94
95STATIC int
Dave Chinner6d4a8ec2011-03-07 10:06:35 +110096xfs_alert_fsblock_zero(
Nathan Scott572d95f2006-09-28 11:03:20 +100097 xfs_inode_t *ip,
98 xfs_bmbt_irec_t *imap)
99{
Dave Chinner6a19d932011-03-07 10:02:35 +1100100 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
Nathan Scott572d95f2006-09-28 11:03:20 +1000101 "Access to block zero in inode %llu "
102 "start_block: %llx start_off: %llx "
Eric Sandeen08e96e12013-10-11 20:59:05 -0500103 "blkcnt: %llx extent-state: %x",
Nathan Scott572d95f2006-09-28 11:03:20 +1000104 (unsigned long long)ip->i_ino,
105 (unsigned long long)imap->br_startblock,
106 (unsigned long long)imap->br_startoff,
107 (unsigned long long)imap->br_blockcount,
108 imap->br_state);
Dave Chinner24513372014-06-25 14:58:08 +1000109 return -EFSCORRUPTED;
Nathan Scott572d95f2006-09-28 11:03:20 +1000110}
111
Christoph Hellwiga206c812010-12-10 08:42:20 +0000112int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113xfs_iomap_write_direct(
114 xfs_inode_t *ip,
Nathan Scottf403b7f2005-05-05 13:33:40 -0700115 xfs_off_t offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 size_t count,
Christoph Hellwig30704512010-06-24 11:42:19 +1000117 xfs_bmbt_irec_t *imap,
Christoph Hellwig405f8042010-12-10 08:42:19 +0000118 int nmaps)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119{
120 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 xfs_fileoff_t offset_fsb;
122 xfs_fileoff_t last_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100123 xfs_filblks_t count_fsb, resaligned;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 xfs_fsblock_t firstfsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100125 xfs_extlen_t extsz, temp;
Eric Sandeen0116d932005-11-02 15:00:01 +1100126 int nimaps;
Nathan Scott06d10dd2005-06-21 15:48:47 +1000127 int quota_flag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 int rt;
129 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 xfs_bmap_free_t free_list;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100131 uint qblocks, resblks, resrtextents;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 int committed;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100133 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134
Dave Chinner507630b2012-03-27 10:34:50 -0400135 error = xfs_qm_dqattach(ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 if (error)
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000137 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
Nathan Scottdd9f4382006-01-11 15:28:28 +1100139 rt = XFS_IS_REALTIME_INODE(ip);
David Chinner957d0eb2007-06-18 16:50:37 +1000140 extsz = xfs_get_extsz_hint(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
David Chinner957d0eb2007-06-18 16:50:37 +1000142 offset_fsb = XFS_B_TO_FSBT(mp, offset);
143 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000144 if ((offset + count) > XFS_ISIZE(ip)) {
Lachlan McIlroy9f6c92b2008-12-22 17:56:49 +1100145 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100146 if (error)
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000147 return error;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100148 } else {
Christoph Hellwig405f8042010-12-10 08:42:19 +0000149 if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
Nathan Scottdd9f4382006-01-11 15:28:28 +1100150 last_fsb = MIN(last_fsb, (xfs_fileoff_t)
Christoph Hellwig30704512010-06-24 11:42:19 +1000151 imap->br_blockcount +
152 imap->br_startoff);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100153 }
154 count_fsb = last_fsb - offset_fsb;
155 ASSERT(count_fsb > 0);
156
157 resaligned = count_fsb;
158 if (unlikely(extsz)) {
159 if ((temp = do_mod(offset_fsb, extsz)))
160 resaligned += temp;
161 if ((temp = do_mod(resaligned, extsz)))
162 resaligned += extsz - temp;
163 }
164
165 if (unlikely(rt)) {
166 resrtextents = qblocks = resaligned;
167 resrtextents /= mp->m_sb.sb_rextsize;
David Chinner84e1e992007-06-18 16:50:27 +1000168 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
169 quota_flag = XFS_QMOPT_RES_RTBLKS;
170 } else {
171 resrtextents = 0;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100172 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
David Chinner84e1e992007-06-18 16:50:27 +1000173 quota_flag = XFS_QMOPT_RES_REGBLKS;
174 }
Nathan Scottdd9f4382006-01-11 15:28:28 +1100175
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 /*
Nathan Scott06d10dd2005-06-21 15:48:47 +1000177 * Allocate and setup the transaction
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
Jie Liu3d3c8b52013-08-12 20:49:59 +1000180 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
181 resblks, resrtextents);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 /*
Nathan Scott06d10dd2005-06-21 15:48:47 +1000183 * Check for running out of space, note: need lock to return
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 */
Dave Chinner507630b2012-03-27 10:34:50 -0400185 if (error) {
Christoph Hellwig4906e212015-06-04 13:47:56 +1000186 xfs_trans_cancel(tp);
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000187 return error;
Dave Chinner507630b2012-03-27 10:34:50 -0400188 }
189
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 xfs_ilock(ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191
Christoph Hellwig7d095252009-06-08 15:33:32 +0200192 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100193 if (error)
Dave Chinner507630b2012-03-27 10:34:50 -0400194 goto out_trans_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
Christoph Hellwigddc34152011-09-19 15:00:54 +0000196 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 /*
Christoph Hellwig30704512010-06-24 11:42:19 +1000199 * From this point onwards we overwrite the imap pointer that the
200 * caller gave to us.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 */
Eric Sandeen9d87c312009-01-14 23:22:07 -0600202 xfs_bmap_init(&free_list, &firstfsb);
Nathan Scott06d10dd2005-06-21 15:48:47 +1000203 nimaps = 1;
Christoph Hellwigd531d912014-02-10 10:27:43 +1100204 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
205 XFS_BMAPI_PREALLOC, &firstfsb, 0,
206 imap, &nimaps, &free_list);
Nathan Scott06d10dd2005-06-21 15:48:47 +1000207 if (error)
Dave Chinner507630b2012-03-27 10:34:50 -0400208 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
210 /*
Nathan Scott06d10dd2005-06-21 15:48:47 +1000211 * Complete the transaction
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +1100213 error = xfs_bmap_finish(&tp, &free_list, &committed);
Nathan Scott06d10dd2005-06-21 15:48:47 +1000214 if (error)
Dave Chinner507630b2012-03-27 10:34:50 -0400215 goto out_bmap_cancel;
Christoph Hellwig70393312015-06-04 13:48:08 +1000216 error = xfs_trans_commit(tp);
Nathan Scott06d10dd2005-06-21 15:48:47 +1000217 if (error)
Dave Chinner507630b2012-03-27 10:34:50 -0400218 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
Nathan Scott06d10dd2005-06-21 15:48:47 +1000220 /*
221 * Copy any maps to caller's array and return any error.
222 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 if (nimaps == 0) {
Dave Chinner24513372014-06-25 14:58:08 +1000224 error = -ENOSPC;
Dave Chinner507630b2012-03-27 10:34:50 -0400225 goto out_unlock;
Nathan Scott572d95f2006-09-28 11:03:20 +1000226 }
227
Dave Chinner507630b2012-03-27 10:34:50 -0400228 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
Dave Chinner6d4a8ec2011-03-07 10:06:35 +1100229 error = xfs_alert_fsblock_zero(ip, imap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
Dave Chinner507630b2012-03-27 10:34:50 -0400231out_unlock:
232 xfs_iunlock(ip, XFS_ILOCK_EXCL);
233 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
Dave Chinner507630b2012-03-27 10:34:50 -0400235out_bmap_cancel:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 xfs_bmap_cancel(&free_list);
Dave Chinnerea562ed2012-05-08 20:48:53 +1000237 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
Dave Chinner507630b2012-03-27 10:34:50 -0400238out_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +1000239 xfs_trans_cancel(tp);
Dave Chinner507630b2012-03-27 10:34:50 -0400240 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241}
242
Nathan Scottdd9f4382006-01-11 15:28:28 +1100243/*
Dave Chinner8de2bf92009-04-06 18:49:12 +0200244 * If the caller is doing a write at the end of the file, then extend the
245 * allocation out to the file system's write iosize. We clean up any extra
246 * space left over when the file is closed in xfs_inactive().
Dave Chinner055388a2011-01-04 11:35:03 +1100247 *
248 * If we find we already have delalloc preallocation beyond EOF, don't do more
249 * preallocation as it it not needed.
Nathan Scottdd9f4382006-01-11 15:28:28 +1100250 */
251STATIC int
252xfs_iomap_eof_want_preallocate(
253 xfs_mount_t *mp,
Lachlan McIlroy541d7d32007-10-11 17:34:33 +1000254 xfs_inode_t *ip,
Nathan Scottdd9f4382006-01-11 15:28:28 +1100255 xfs_off_t offset,
256 size_t count,
Nathan Scottdd9f4382006-01-11 15:28:28 +1100257 xfs_bmbt_irec_t *imap,
258 int nimaps,
259 int *prealloc)
260{
261 xfs_fileoff_t start_fsb;
262 xfs_filblks_t count_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100263 int n, error, imaps;
Dave Chinner055388a2011-01-04 11:35:03 +1100264 int found_delalloc = 0;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100265
266 *prealloc = 0;
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000267 if (offset + count <= XFS_ISIZE(ip))
Nathan Scottdd9f4382006-01-11 15:28:28 +1100268 return 0;
269
270 /*
Dave Chinner133eeb12013-06-27 16:04:48 +1000271 * If the file is smaller than the minimum prealloc and we are using
272 * dynamic preallocation, don't do any preallocation at all as it is
273 * likely this is the only write to the file that is going to be done.
274 */
275 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) &&
276 XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks))
277 return 0;
278
279 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +1100280 * If there are any real blocks past eof, then don't
281 * do any speculative allocation.
282 */
283 start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
Dave Chinner32972382012-06-08 15:44:54 +1000284 count_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100285 while (count_fsb > 0) {
286 imaps = nimaps;
Dave Chinner5c8ed202011-09-18 20:40:45 +0000287 error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps,
288 0);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100289 if (error)
290 return error;
291 for (n = 0; n < imaps; n++) {
292 if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
293 (imap[n].br_startblock != DELAYSTARTBLOCK))
294 return 0;
295 start_fsb += imap[n].br_blockcount;
296 count_fsb -= imap[n].br_blockcount;
Dave Chinner055388a2011-01-04 11:35:03 +1100297
298 if (imap[n].br_startblock == DELAYSTARTBLOCK)
299 found_delalloc = 1;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100300 }
301 }
Dave Chinner055388a2011-01-04 11:35:03 +1100302 if (!found_delalloc)
303 *prealloc = 1;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100304 return 0;
305}
306
Dave Chinner055388a2011-01-04 11:35:03 +1100307/*
Dave Chinnera1e16c22013-02-11 16:05:01 +1100308 * Determine the initial size of the preallocation. We are beyond the current
309 * EOF here, but we need to take into account whether this is a sparse write or
310 * an extending write when determining the preallocation size. Hence we need to
311 * look up the extent that ends at the current write offset and use the result
312 * to determine the preallocation size.
313 *
314 * If the extent is a hole, then preallocation is essentially disabled.
315 * Otherwise we take the size of the preceeding data extent as the basis for the
316 * preallocation size. If the size of the extent is greater than half the
317 * maximum extent length, then use the current offset as the basis. This ensures
318 * that for large files the preallocation size always extends to MAXEXTLEN
319 * rather than falling short due to things like stripe unit/width alignment of
320 * real extents.
321 */
Mark Tinguelye8108ce2013-02-24 13:04:37 -0600322STATIC xfs_fsblock_t
Dave Chinnera1e16c22013-02-11 16:05:01 +1100323xfs_iomap_eof_prealloc_initial_size(
324 struct xfs_mount *mp,
325 struct xfs_inode *ip,
326 xfs_off_t offset,
327 xfs_bmbt_irec_t *imap,
328 int nimaps)
329{
330 xfs_fileoff_t start_fsb;
331 int imaps = 1;
332 int error;
333
334 ASSERT(nimaps >= imaps);
335
336 /* if we are using a specific prealloc size, return now */
337 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
338 return 0;
339
Dave Chinner133eeb12013-06-27 16:04:48 +1000340 /* If the file is small, then use the minimum prealloc */
341 if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign))
342 return 0;
343
Dave Chinnera1e16c22013-02-11 16:05:01 +1100344 /*
345 * As we write multiple pages, the offset will always align to the
346 * start of a page and hence point to a hole at EOF. i.e. if the size is
347 * 4096 bytes, we only have one block at FSB 0, but XFS_B_TO_FSB(4096)
348 * will return FSB 1. Hence if there are blocks in the file, we want to
349 * point to the block prior to the EOF block and not the hole that maps
350 * directly at @offset.
351 */
352 start_fsb = XFS_B_TO_FSB(mp, offset);
353 if (start_fsb)
354 start_fsb--;
355 error = xfs_bmapi_read(ip, start_fsb, 1, imap, &imaps, XFS_BMAPI_ENTIRE);
356 if (error)
357 return 0;
358
359 ASSERT(imaps == 1);
360 if (imap[0].br_startblock == HOLESTARTBLOCK)
361 return 0;
362 if (imap[0].br_blockcount <= (MAXEXTLEN >> 1))
Brian Fostere114b5f2013-02-19 10:24:41 -0500363 return imap[0].br_blockcount << 1;
Dave Chinnera1e16c22013-02-11 16:05:01 +1100364 return XFS_B_TO_FSB(mp, offset);
365}
366
Brian Foster76a42022013-03-18 10:51:47 -0400367STATIC bool
368xfs_quota_need_throttle(
369 struct xfs_inode *ip,
370 int type,
371 xfs_fsblock_t alloc_blocks)
372{
373 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
374
375 if (!dq || !xfs_this_quota_on(ip->i_mount, type))
376 return false;
377
378 /* no hi watermark, no throttle */
379 if (!dq->q_prealloc_hi_wmark)
380 return false;
381
382 /* under the lo watermark, no throttle */
383 if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark)
384 return false;
385
386 return true;
387}
388
389STATIC void
390xfs_quota_calc_throttle(
391 struct xfs_inode *ip,
392 int type,
393 xfs_fsblock_t *qblocks,
Brian Fosterf0740512014-07-24 19:56:08 +1000394 int *qshift,
395 int64_t *qfreesp)
Brian Foster76a42022013-03-18 10:51:47 -0400396{
397 int64_t freesp;
398 int shift = 0;
399 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
400
Eric Sandeen5cca3f62014-10-02 09:27:09 +1000401 /* no dq, or over hi wmark, squash the prealloc completely */
402 if (!dq || dq->q_res_bcount >= dq->q_prealloc_hi_wmark) {
Brian Foster76a42022013-03-18 10:51:47 -0400403 *qblocks = 0;
Brian Fosterf0740512014-07-24 19:56:08 +1000404 *qfreesp = 0;
Brian Foster76a42022013-03-18 10:51:47 -0400405 return;
406 }
407
408 freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount;
409 if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
410 shift = 2;
411 if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
412 shift += 2;
413 if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
414 shift += 2;
415 }
416
Brian Fosterf0740512014-07-24 19:56:08 +1000417 if (freesp < *qfreesp)
418 *qfreesp = freesp;
419
Brian Foster76a42022013-03-18 10:51:47 -0400420 /* only overwrite the throttle values if we are more aggressive */
421 if ((freesp >> shift) < (*qblocks >> *qshift)) {
422 *qblocks = freesp;
423 *qshift = shift;
424 }
425}
426
Dave Chinnera1e16c22013-02-11 16:05:01 +1100427/*
Dave Chinner055388a2011-01-04 11:35:03 +1100428 * If we don't have a user specified preallocation size, dynamically increase
429 * the preallocation size as the size of the file grows. Cap the maximum size
430 * at a single extent or less if the filesystem is near full. The closer the
431 * filesystem is to full, the smaller the maximum prealocation.
432 */
433STATIC xfs_fsblock_t
434xfs_iomap_prealloc_size(
435 struct xfs_mount *mp,
Dave Chinnera1e16c22013-02-11 16:05:01 +1100436 struct xfs_inode *ip,
437 xfs_off_t offset,
438 struct xfs_bmbt_irec *imap,
439 int nimaps)
Dave Chinner055388a2011-01-04 11:35:03 +1100440{
441 xfs_fsblock_t alloc_blocks = 0;
Brian Foster3c58b5f2013-03-18 10:51:43 -0400442 int shift = 0;
443 int64_t freesp;
Brian Foster76a42022013-03-18 10:51:47 -0400444 xfs_fsblock_t qblocks;
445 int qshift = 0;
Dave Chinner055388a2011-01-04 11:35:03 +1100446
Dave Chinnera1e16c22013-02-11 16:05:01 +1100447 alloc_blocks = xfs_iomap_eof_prealloc_initial_size(mp, ip, offset,
448 imap, nimaps);
Brian Foster3c58b5f2013-03-18 10:51:43 -0400449 if (!alloc_blocks)
450 goto check_writeio;
Brian Foster76a42022013-03-18 10:51:47 -0400451 qblocks = alloc_blocks;
Dave Chinner055388a2011-01-04 11:35:03 +1100452
Brian Fosterc9bdbdc2013-03-18 10:51:44 -0400453 /*
454 * MAXEXTLEN is not a power of two value but we round the prealloc down
455 * to the nearest power of two value after throttling. To prevent the
456 * round down from unconditionally reducing the maximum supported prealloc
457 * size, we round up first, apply appropriate throttling, round down and
458 * cap the value to MAXEXTLEN.
459 */
460 alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
461 alloc_blocks);
Dave Chinner055388a2011-01-04 11:35:03 +1100462
Dave Chinner0d485ad2015-02-23 21:22:03 +1100463 freesp = percpu_counter_read_positive(&mp->m_fdblocks);
Brian Foster3c58b5f2013-03-18 10:51:43 -0400464 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 /*
Brian Fosterf0740512014-07-24 19:56:08 +1000477 * Check each quota to cap the prealloc size, provide a shift value to
478 * throttle with and adjust amount of available space.
Brian Foster76a42022013-03-18 10:51:47 -0400479 */
480 if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks))
Brian Fosterf0740512014-07-24 19:56:08 +1000481 xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift,
482 &freesp);
Brian Foster76a42022013-03-18 10:51:47 -0400483 if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks))
Brian Fosterf0740512014-07-24 19:56:08 +1000484 xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift,
485 &freesp);
Brian Foster76a42022013-03-18 10:51:47 -0400486 if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks))
Brian Fosterf0740512014-07-24 19:56:08 +1000487 xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift,
488 &freesp);
Brian Foster76a42022013-03-18 10:51:47 -0400489
490 /*
491 * The final prealloc size is set to the minimum of free space available
492 * in each of the quotas and the overall filesystem.
493 *
494 * The shift throttle value is set to the maximum value as determined by
495 * the global low free space values and per-quota low free space values.
496 */
497 alloc_blocks = MIN(alloc_blocks, qblocks);
498 shift = MAX(shift, qshift);
499
Brian Foster3c58b5f2013-03-18 10:51:43 -0400500 if (shift)
501 alloc_blocks >>= shift;
Brian Fosterc9bdbdc2013-03-18 10:51:44 -0400502 /*
503 * rounddown_pow_of_two() returns an undefined result if we pass in
504 * alloc_blocks = 0.
505 */
506 if (alloc_blocks)
507 alloc_blocks = rounddown_pow_of_two(alloc_blocks);
508 if (alloc_blocks > MAXEXTLEN)
509 alloc_blocks = MAXEXTLEN;
Dave Chinner055388a2011-01-04 11:35:03 +1100510
Brian Foster3c58b5f2013-03-18 10:51:43 -0400511 /*
512 * If we are still trying to allocate more space than is
513 * available, squash the prealloc hard. This can happen if we
514 * have a large file on a small filesystem and the above
515 * lowspace thresholds are smaller than MAXEXTLEN.
516 */
517 while (alloc_blocks && alloc_blocks >= freesp)
518 alloc_blocks >>= 4;
519
520check_writeio:
Dave Chinner055388a2011-01-04 11:35:03 +1100521 if (alloc_blocks < mp->m_writeio_blocks)
522 alloc_blocks = mp->m_writeio_blocks;
523
Brian Foster19cb7e32013-03-18 10:51:48 -0400524 trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
525 mp->m_writeio_blocks);
526
Dave Chinner055388a2011-01-04 11:35:03 +1100527 return alloc_blocks;
528}
529
Christoph Hellwiga206c812010-12-10 08:42:20 +0000530int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531xfs_iomap_write_delay(
532 xfs_inode_t *ip,
Nathan Scottf403b7f2005-05-05 13:33:40 -0700533 xfs_off_t offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 size_t count,
Christoph Hellwig405f8042010-12-10 08:42:19 +0000535 xfs_bmbt_irec_t *ret_imap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536{
537 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 xfs_fileoff_t offset_fsb;
539 xfs_fileoff_t last_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100540 xfs_off_t aligned_offset;
541 xfs_fileoff_t ioalign;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100542 xfs_extlen_t extsz;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 int nimaps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
Dave Chinner9aa05002012-10-08 21:56:04 +1100545 int prealloc;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100546 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000548 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
550 /*
551 * Make sure that the dquots are there. This doesn't hold
552 * the ilock across a disk read.
553 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200554 error = xfs_qm_dqattach_locked(ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 if (error)
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000556 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
David Chinner957d0eb2007-06-18 16:50:37 +1000558 extsz = xfs_get_extsz_hint(ip);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100559 offset_fsb = XFS_B_TO_FSBT(mp, offset);
560
Lachlan McIlroy9f6c92b2008-12-22 17:56:49 +1100561 error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
Christoph Hellwig405f8042010-12-10 08:42:19 +0000562 imap, XFS_WRITE_IMAPS, &prealloc);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100563 if (error)
564 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565
Dave Chinner8de2bf92009-04-06 18:49:12 +0200566retry:
Nathan Scottdd9f4382006-01-11 15:28:28 +1100567 if (prealloc) {
Dave Chinnera1e16c22013-02-11 16:05:01 +1100568 xfs_fsblock_t alloc_blocks;
569
570 alloc_blocks = xfs_iomap_prealloc_size(mp, ip, offset, imap,
571 XFS_WRITE_IMAPS);
Dave Chinner055388a2011-01-04 11:35:03 +1100572
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
574 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
Dave Chinner055388a2011-01-04 11:35:03 +1100575 last_fsb = ioalign + alloc_blocks;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100576 } else {
577 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 }
Nathan Scottdd9f4382006-01-11 15:28:28 +1100579
580 if (prealloc || extsz) {
Lachlan McIlroy9f6c92b2008-12-22 17:56:49 +1100581 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100582 if (error)
583 return error;
584 }
585
Dave Chinner3ed91162012-04-29 22:43:19 +1000586 /*
587 * Make sure preallocation does not create extents beyond the range we
588 * actually support in this filesystem.
589 */
Dave Chinnerd2c28192012-06-08 15:44:53 +1000590 if (last_fsb > XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes))
591 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
Dave Chinner3ed91162012-04-29 22:43:19 +1000592
593 ASSERT(last_fsb > offset_fsb);
594
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 nimaps = XFS_WRITE_IMAPS;
Christoph Hellwig44032802011-09-18 20:40:48 +0000596 error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb,
597 imap, &nimaps, XFS_BMAPI_ENTIRE);
Dave Chinner055388a2011-01-04 11:35:03 +1100598 switch (error) {
599 case 0:
Dave Chinner24513372014-06-25 14:58:08 +1000600 case -ENOSPC:
601 case -EDQUOT:
Dave Chinner055388a2011-01-04 11:35:03 +1100602 break;
603 default:
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000604 return error;
Dave Chinner055388a2011-01-04 11:35:03 +1100605 }
Nathan Scottdd9f4382006-01-11 15:28:28 +1100606
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 /*
Dave Chinner9aa05002012-10-08 21:56:04 +1100608 * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. Retry
Dave Chinner055388a2011-01-04 11:35:03 +1100609 * without EOF preallocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 */
611 if (nimaps == 0) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000612 trace_xfs_delalloc_enospc(ip, offset, count);
Dave Chinner9aa05002012-10-08 21:56:04 +1100613 if (prealloc) {
614 prealloc = 0;
615 error = 0;
616 goto retry;
Dave Chinner055388a2011-01-04 11:35:03 +1100617 }
Dave Chinner24513372014-06-25 14:58:08 +1000618 return error ? error : -ENOSPC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 }
620
David Chinner86c4d622008-04-29 12:53:21 +1000621 if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
Dave Chinner6d4a8ec2011-03-07 10:06:35 +1100622 return xfs_alert_fsblock_zero(ip, &imap[0]);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100623
Brian Foster27b52862012-11-06 09:50:38 -0500624 /*
625 * Tag the inode as speculatively preallocated so we can reclaim this
626 * space on demand, if necessary.
627 */
628 if (prealloc)
629 xfs_inode_set_eofblocks_tag(ip);
630
Nathan Scottdd9f4382006-01-11 15:28:28 +1100631 *ret_imap = imap[0];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 return 0;
633}
634
635/*
636 * Pass in a delayed allocate extent, convert it to real extents;
637 * return to the caller the extent we create which maps on top of
638 * the originating callers request.
639 *
640 * Called without a lock on the inode.
David Chinnere4143a12007-11-23 16:29:11 +1100641 *
642 * We no longer bother to look at the incoming map - all we have to
643 * guarantee is that whatever we allocate fills the required range.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 */
Christoph Hellwiga206c812010-12-10 08:42:20 +0000645int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646xfs_iomap_write_allocate(
647 xfs_inode_t *ip,
Nathan Scottf403b7f2005-05-05 13:33:40 -0700648 xfs_off_t offset,
Christoph Hellwig405f8042010-12-10 08:42:19 +0000649 xfs_bmbt_irec_t *imap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650{
651 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 xfs_fileoff_t offset_fsb, last_block;
653 xfs_fileoff_t end_fsb, map_start_fsb;
654 xfs_fsblock_t first_block;
655 xfs_bmap_free_t free_list;
656 xfs_filblks_t count_fsb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 xfs_trans_t *tp;
David Chinnere4143a12007-11-23 16:29:11 +1100658 int nimaps, committed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 int error = 0;
660 int nres;
661
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 /*
663 * Make sure that the dquots are there.
664 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200665 error = xfs_qm_dqattach(ip, 0);
666 if (error)
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000667 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
Nathan Scott24e17b52005-05-05 13:33:20 -0700669 offset_fsb = XFS_B_TO_FSBT(mp, offset);
Christoph Hellwig30704512010-06-24 11:42:19 +1000670 count_fsb = imap->br_blockcount;
671 map_start_fsb = imap->br_startoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672
673 XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
674
675 while (count_fsb != 0) {
676 /*
677 * Set up a transaction with which to allocate the
678 * backing store for the file. Do allocations in a
679 * loop until we get some space in the range we are
680 * interested in. The other space that might be allocated
681 * is in the delayed allocation extent on which we sit
682 * but before our buffer starts.
683 */
684
685 nimaps = 0;
686 while (nimaps == 0) {
687 tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
David Chinner84e1e992007-06-18 16:50:27 +1000688 tp->t_flags |= XFS_TRANS_RESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
Jie Liu3d3c8b52013-08-12 20:49:59 +1000690 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
691 nres, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 if (error) {
Christoph Hellwig4906e212015-06-04 13:47:56 +1000693 xfs_trans_cancel(tp);
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000694 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 }
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));
Eric Sandeen7fb2cd42014-04-14 18:58:05 +1000734 error = xfs_bmap_last_offset(ip, &last_block,
David Chinner7c9ef852008-04-10 12:21:59 +1000735 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) {
Dave Chinner24513372014-06-25 14:58:08 +1000743 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 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 Chinnercf11da92014-07-15 07:08:24 +1000753 count_fsb, 0,
Dave Chinner24558812012-10-05 11:06:58 +1000754 &first_block, 1,
Dave Chinnerc0dc7822011-09-18 20:40:52 +0000755 imap, &nimaps, &free_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 if (error)
757 goto trans_cancel;
758
Eric Sandeenf7c99b62007-02-10 18:37:16 +1100759 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 if (error)
761 goto trans_cancel;
762
Christoph Hellwig70393312015-06-04 13:48:08 +1000763 error = xfs_trans_commit(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 if (error)
765 goto error0;
766
767 xfs_iunlock(ip, XFS_ILOCK_EXCL);
768 }
769
770 /*
771 * See if we were able to allocate an extent that
772 * covers at least part of the callers request
773 */
Christoph Hellwig30704512010-06-24 11:42:19 +1000774 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
Dave Chinner6d4a8ec2011-03-07 10:06:35 +1100775 return xfs_alert_fsblock_zero(ip, imap);
David Chinner86c4d622008-04-29 12:53:21 +1000776
Christoph Hellwig30704512010-06-24 11:42:19 +1000777 if ((offset_fsb >= imap->br_startoff) &&
778 (offset_fsb < (imap->br_startoff +
779 imap->br_blockcount))) {
David Chinnere4143a12007-11-23 16:29:11 +1100780 XFS_STATS_INC(xs_xstrat_quick);
781 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 }
783
David Chinnere4143a12007-11-23 16:29:11 +1100784 /*
785 * So far we have not mapped the requested part of the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 * file, just surrounding data, try again.
787 */
Christoph Hellwig30704512010-06-24 11:42:19 +1000788 count_fsb -= imap->br_blockcount;
789 map_start_fsb = imap->br_startoff + imap->br_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 }
791
792trans_cancel:
793 xfs_bmap_cancel(&free_list);
Christoph Hellwig4906e212015-06-04 13:47:56 +1000794 xfs_trans_cancel(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795error0:
796 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000797 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798}
799
800int
801xfs_iomap_write_unwritten(
802 xfs_inode_t *ip,
Nathan Scottf403b7f2005-05-05 13:33:40 -0700803 xfs_off_t offset,
Christoph Hellwigd32057f2015-01-09 10:48:12 +1100804 xfs_off_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805{
806 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 xfs_fileoff_t offset_fsb;
808 xfs_filblks_t count_fsb;
809 xfs_filblks_t numblks_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100810 xfs_fsblock_t firstfsb;
811 int nimaps;
812 xfs_trans_t *tp;
813 xfs_bmbt_irec_t imap;
814 xfs_bmap_free_t free_list;
Christoph Hellwig84803fb2012-02-29 09:53:50 +0000815 xfs_fsize_t i_size;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100816 uint resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 int committed;
818 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000820 trace_xfs_unwritten_convert(ip, offset, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821
822 offset_fsb = XFS_B_TO_FSBT(mp, offset);
823 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
824 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
825
Lachlan McIlroy4ddd8bb2008-06-27 13:32:53 +1000826 /*
827 * Reserve enough blocks in this transaction for two complete extent
828 * btree splits. We may be converting the middle part of an unwritten
829 * extent and in this case we will insert two new extents in the btree
830 * each of which could cause a full split.
831 *
832 * This reservation amount will be used in the first call to
833 * xfs_bmbt_split() to select an AG with enough space to satisfy the
834 * rest of the operation.
835 */
Nathan Scottdd9f4382006-01-11 15:28:28 +1100836 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837
Nathan Scottdd9f4382006-01-11 15:28:28 +1100838 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 /*
840 * set up a transaction to convert the range of extents
841 * from unwritten to real. Do allocations in a loop until
842 * we have covered the range passed in.
Christoph Hellwig80641dc2009-10-19 04:00:03 +0000843 *
844 * Note that we open code the transaction allocation here
845 * to pass KM_NOFS--we can't risk to recursing back into
846 * the filesystem here as we might be asked to write out
847 * the same inode that we complete here and might deadlock
848 * on the iolock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 */
Jan Karad9457dc2012-06-12 16:20:39 +0200850 sb_start_intwrite(mp->m_super);
Christoph Hellwig80641dc2009-10-19 04:00:03 +0000851 tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS);
Jan Karad9457dc2012-06-12 16:20:39 +0200852 tp->t_flags |= XFS_TRANS_RESERVE | XFS_TRANS_FREEZE_PROT;
Jie Liu3d3c8b52013-08-12 20:49:59 +1000853 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
854 resblks, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 if (error) {
Christoph Hellwig4906e212015-06-04 13:47:56 +1000856 xfs_trans_cancel(tp);
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000857 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 }
859
860 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigddc34152011-09-19 15:00:54 +0000861 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
863 /*
864 * Modify the unwritten extent state of the buffer.
865 */
Eric Sandeen9d87c312009-01-14 23:22:07 -0600866 xfs_bmap_init(&free_list, &firstfsb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 nimaps = 1;
Dave Chinnerc0dc7822011-09-18 20:40:52 +0000868 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
869 XFS_BMAPI_CONVERT, &firstfsb,
Christoph Hellwigb4e91812010-06-23 18:11:15 +1000870 1, &imap, &nimaps, &free_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 if (error)
872 goto error_on_bmapi_transaction;
873
Christoph Hellwig84803fb2012-02-29 09:53:50 +0000874 /*
875 * Log the updated inode size as we go. We have to be careful
876 * to only log it up to the actual write offset if it is
877 * halfway into a block.
878 */
879 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
880 if (i_size > offset + count)
881 i_size = offset + count;
882
883 i_size = xfs_new_eof(ip, i_size);
884 if (i_size) {
885 ip->i_d.di_size = i_size;
886 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
887 }
888
889 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 if (error)
891 goto error_on_bmapi_transaction;
892
Christoph Hellwig70393312015-06-04 13:48:08 +1000893 error = xfs_trans_commit(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 xfs_iunlock(ip, XFS_ILOCK_EXCL);
895 if (error)
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000896 return error;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100897
David Chinner86c4d622008-04-29 12:53:21 +1000898 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
Dave Chinner6d4a8ec2011-03-07 10:06:35 +1100899 return xfs_alert_fsblock_zero(ip, &imap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900
901 if ((numblks_fsb = imap.br_blockcount) == 0) {
902 /*
903 * The numblks_fsb value should always get
904 * smaller, otherwise the loop is stuck.
905 */
906 ASSERT(imap.br_blockcount);
907 break;
908 }
909 offset_fsb += numblks_fsb;
910 count_fsb -= numblks_fsb;
911 } while (count_fsb > 0);
912
913 return 0;
914
915error_on_bmapi_transaction:
916 xfs_bmap_cancel(&free_list);
Christoph Hellwig4906e212015-06-04 13:47:56 +1000917 xfs_trans_cancel(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000919 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920}