blob: 9afa282aa937b473aad0d90fc6f2a0bfb659dd29 [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"
Nathan Scotta844f452005-11-02 14:38:42 +110020#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110022#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_trans.h"
24#include "xfs_sb.h"
25#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_quota.h"
28#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110030#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_dinode.h"
33#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110034#include "xfs_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_bmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_rtalloc.h"
37#include "xfs_error.h"
38#include "xfs_itable.h"
39#include "xfs_rw.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include "xfs_attr.h"
41#include "xfs_buf_item.h"
42#include "xfs_trans_space.h"
43#include "xfs_utils.h"
44#include "xfs_iomap.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000045#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48#define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \
49 << mp->m_writeio_log)
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP
51
Linus Torvalds1da177e2005-04-16 15:20:36 -070052STATIC int
Nathan Scottdd9f4382006-01-11 15:28:28 +110053xfs_iomap_eof_align_last_fsb(
54 xfs_mount_t *mp,
Lachlan McIlroy541d7d32007-10-11 17:34:33 +100055 xfs_inode_t *ip,
Nathan Scottdd9f4382006-01-11 15:28:28 +110056 xfs_extlen_t extsize,
57 xfs_fileoff_t *last_fsb)
58{
59 xfs_fileoff_t new_last_fsb = 0;
60 xfs_extlen_t align;
61 int eof, error;
62
Eric Sandeen71ddabb2007-11-23 16:29:42 +110063 if (XFS_IS_REALTIME_INODE(ip))
Nathan Scottdd9f4382006-01-11 15:28:28 +110064 ;
65 /*
66 * If mounted with the "-o swalloc" option, roundup the allocation
67 * request to a stripe width boundary if the file size is >=
68 * stripe width and we are allocating past the allocation eof.
69 */
70 else if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC) &&
Lachlan McIlroy9f6c92b2008-12-22 17:56:49 +110071 (ip->i_size >= XFS_FSB_TO_B(mp, mp->m_swidth)))
Nathan Scottdd9f4382006-01-11 15:28:28 +110072 new_last_fsb = roundup_64(*last_fsb, mp->m_swidth);
73 /*
74 * Roundup the allocation request to a stripe unit (m_dalign) boundary
75 * if the file size is >= stripe unit size, and we are allocating past
76 * the allocation eof.
77 */
Lachlan McIlroy9f6c92b2008-12-22 17:56:49 +110078 else if (mp->m_dalign && (ip->i_size >= XFS_FSB_TO_B(mp, mp->m_dalign)))
Nathan Scottdd9f4382006-01-11 15:28:28 +110079 new_last_fsb = roundup_64(*last_fsb, mp->m_dalign);
80
81 /*
82 * Always round up the allocation request to an extent boundary
83 * (when file on a real-time subvolume or has di_extsize hint).
84 */
85 if (extsize) {
86 if (new_last_fsb)
87 align = roundup_64(new_last_fsb, extsize);
88 else
89 align = extsize;
90 new_last_fsb = roundup_64(*last_fsb, align);
91 }
92
93 if (new_last_fsb) {
Lachlan McIlroy541d7d32007-10-11 17:34:33 +100094 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
Nathan Scottdd9f4382006-01-11 15:28:28 +110095 if (error)
96 return error;
97 if (eof)
98 *last_fsb = new_last_fsb;
99 }
100 return 0;
101}
102
103STATIC int
Dave Chinner6d4a8ec2011-03-07 10:06:35 +1100104xfs_alert_fsblock_zero(
Nathan Scott572d95f2006-09-28 11:03:20 +1000105 xfs_inode_t *ip,
106 xfs_bmbt_irec_t *imap)
107{
Dave Chinner6a19d932011-03-07 10:02:35 +1100108 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
Nathan Scott572d95f2006-09-28 11:03:20 +1000109 "Access to block zero in inode %llu "
110 "start_block: %llx start_off: %llx "
111 "blkcnt: %llx extent-state: %x\n",
112 (unsigned long long)ip->i_ino,
113 (unsigned long long)imap->br_startblock,
114 (unsigned long long)imap->br_startoff,
115 (unsigned long long)imap->br_blockcount,
116 imap->br_state);
117 return EFSCORRUPTED;
118}
119
Christoph Hellwiga206c812010-12-10 08:42:20 +0000120int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121xfs_iomap_write_direct(
122 xfs_inode_t *ip,
Nathan Scottf403b7f2005-05-05 13:33:40 -0700123 xfs_off_t offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 size_t count,
Christoph Hellwig30704512010-06-24 11:42:19 +1000125 xfs_bmbt_irec_t *imap,
Christoph Hellwig405f8042010-12-10 08:42:19 +0000126 int nmaps)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127{
128 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 xfs_fileoff_t offset_fsb;
130 xfs_fileoff_t last_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100131 xfs_filblks_t count_fsb, resaligned;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 xfs_fsblock_t firstfsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100133 xfs_extlen_t extsz, temp;
Eric Sandeen0116d932005-11-02 15:00:01 +1100134 int nimaps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 int bmapi_flag;
Nathan Scott06d10dd2005-06-21 15:48:47 +1000136 int quota_flag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 int rt;
138 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 xfs_bmap_free_t free_list;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100140 uint qblocks, resblks, resrtextents;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 int committed;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100142 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
144 /*
145 * Make sure that the dquots are there. This doesn't hold
146 * the ilock across a disk read.
147 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200148 error = xfs_qm_dqattach_locked(ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 if (error)
150 return XFS_ERROR(error);
151
Nathan Scottdd9f4382006-01-11 15:28:28 +1100152 rt = XFS_IS_REALTIME_INODE(ip);
David Chinner957d0eb2007-06-18 16:50:37 +1000153 extsz = xfs_get_extsz_hint(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
David Chinner957d0eb2007-06-18 16:50:37 +1000155 offset_fsb = XFS_B_TO_FSBT(mp, offset);
156 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
Lachlan McIlroy9f6c92b2008-12-22 17:56:49 +1100157 if ((offset + count) > ip->i_size) {
158 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100159 if (error)
160 goto error_out;
161 } else {
Christoph Hellwig405f8042010-12-10 08:42:19 +0000162 if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
Nathan Scottdd9f4382006-01-11 15:28:28 +1100163 last_fsb = MIN(last_fsb, (xfs_fileoff_t)
Christoph Hellwig30704512010-06-24 11:42:19 +1000164 imap->br_blockcount +
165 imap->br_startoff);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100166 }
167 count_fsb = last_fsb - offset_fsb;
168 ASSERT(count_fsb > 0);
169
170 resaligned = count_fsb;
171 if (unlikely(extsz)) {
172 if ((temp = do_mod(offset_fsb, extsz)))
173 resaligned += temp;
174 if ((temp = do_mod(resaligned, extsz)))
175 resaligned += extsz - temp;
176 }
177
178 if (unlikely(rt)) {
179 resrtextents = qblocks = resaligned;
180 resrtextents /= mp->m_sb.sb_rextsize;
David Chinner84e1e992007-06-18 16:50:27 +1000181 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
182 quota_flag = XFS_QMOPT_RES_RTBLKS;
183 } else {
184 resrtextents = 0;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100185 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
David Chinner84e1e992007-06-18 16:50:27 +1000186 quota_flag = XFS_QMOPT_RES_REGBLKS;
187 }
Nathan Scottdd9f4382006-01-11 15:28:28 +1100188
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 /*
Nathan Scott06d10dd2005-06-21 15:48:47 +1000190 * Allocate and setup the transaction
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 */
192 xfs_iunlock(ip, XFS_ILOCK_EXCL);
193 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 error = xfs_trans_reserve(tp, resblks,
Nathan Scottd52b44d2005-09-02 16:41:32 +1000195 XFS_WRITE_LOG_RES(mp), resrtextents,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 XFS_TRANS_PERM_LOG_RES,
197 XFS_WRITE_LOG_COUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 /*
Nathan Scott06d10dd2005-06-21 15:48:47 +1000199 * Check for running out of space, note: need lock to return
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 */
201 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 xfs_trans_cancel(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 xfs_ilock(ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 if (error)
Nathan Scott06d10dd2005-06-21 15:48:47 +1000205 goto error_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
Christoph Hellwig7d095252009-06-08 15:33:32 +0200207 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100208 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 goto error1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
Christoph Hellwigddc34152011-09-19 15:00:54 +0000211 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212
Dave Chinnerc0dc7822011-09-18 20:40:52 +0000213 bmapi_flag = 0;
Christoph Hellwig405f8042010-12-10 08:42:19 +0000214 if (offset < ip->i_size || extsz)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 bmapi_flag |= XFS_BMAPI_PREALLOC;
216
217 /*
Christoph Hellwig30704512010-06-24 11:42:19 +1000218 * From this point onwards we overwrite the imap pointer that the
219 * caller gave to us.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 */
Eric Sandeen9d87c312009-01-14 23:22:07 -0600221 xfs_bmap_init(&free_list, &firstfsb);
Nathan Scott06d10dd2005-06-21 15:48:47 +1000222 nimaps = 1;
Dave Chinnerc0dc7822011-09-18 20:40:52 +0000223 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flag,
224 &firstfsb, 0, imap, &nimaps, &free_list);
Nathan Scott06d10dd2005-06-21 15:48:47 +1000225 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 goto error0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227
228 /*
Nathan Scott06d10dd2005-06-21 15:48:47 +1000229 * Complete the transaction
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +1100231 error = xfs_bmap_finish(&tp, &free_list, &committed);
Nathan Scott06d10dd2005-06-21 15:48:47 +1000232 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 goto error0;
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000234 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Nathan Scott06d10dd2005-06-21 15:48:47 +1000235 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 goto error_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237
Nathan Scott06d10dd2005-06-21 15:48:47 +1000238 /*
239 * Copy any maps to caller's array and return any error.
240 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 if (nimaps == 0) {
Nathan Scott572d95f2006-09-28 11:03:20 +1000242 error = ENOSPC;
243 goto error_out;
244 }
245
Christoph Hellwig30704512010-06-24 11:42:19 +1000246 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) {
Dave Chinner6d4a8ec2011-03-07 10:06:35 +1100247 error = xfs_alert_fsblock_zero(ip, imap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 goto error_out;
249 }
250
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 return 0;
252
Nathan Scott06d10dd2005-06-21 15:48:47 +1000253error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 xfs_bmap_cancel(&free_list);
Christoph Hellwig7d095252009-06-08 15:33:32 +0200255 xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
Nathan Scott06d10dd2005-06-21 15:48:47 +1000257error1: /* Just cancel transaction */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
260error_out:
261 return XFS_ERROR(error);
262}
263
Nathan Scottdd9f4382006-01-11 15:28:28 +1100264/*
Dave Chinner8de2bf92009-04-06 18:49:12 +0200265 * If the caller is doing a write at the end of the file, then extend the
266 * allocation out to the file system's write iosize. We clean up any extra
267 * space left over when the file is closed in xfs_inactive().
Dave Chinner055388a2011-01-04 11:35:03 +1100268 *
269 * If we find we already have delalloc preallocation beyond EOF, don't do more
270 * preallocation as it it not needed.
Nathan Scottdd9f4382006-01-11 15:28:28 +1100271 */
272STATIC int
273xfs_iomap_eof_want_preallocate(
274 xfs_mount_t *mp,
Lachlan McIlroy541d7d32007-10-11 17:34:33 +1000275 xfs_inode_t *ip,
Nathan Scottdd9f4382006-01-11 15:28:28 +1100276 xfs_off_t offset,
277 size_t count,
Nathan Scottdd9f4382006-01-11 15:28:28 +1100278 xfs_bmbt_irec_t *imap,
279 int nimaps,
280 int *prealloc)
281{
282 xfs_fileoff_t start_fsb;
283 xfs_filblks_t count_fsb;
284 xfs_fsblock_t firstblock;
285 int n, error, imaps;
Dave Chinner055388a2011-01-04 11:35:03 +1100286 int found_delalloc = 0;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100287
288 *prealloc = 0;
Dave Chinner8de2bf92009-04-06 18:49:12 +0200289 if ((offset + count) <= ip->i_size)
Nathan Scottdd9f4382006-01-11 15:28:28 +1100290 return 0;
291
292 /*
293 * If there are any real blocks past eof, then don't
294 * do any speculative allocation.
295 */
296 start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
297 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
298 while (count_fsb > 0) {
299 imaps = nimaps;
Nathan Scott3ddb8fa2006-01-11 15:33:02 +1100300 firstblock = NULLFSBLOCK;
Dave Chinner5c8ed202011-09-18 20:40:45 +0000301 error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps,
302 0);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100303 if (error)
304 return error;
305 for (n = 0; n < imaps; n++) {
306 if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
307 (imap[n].br_startblock != DELAYSTARTBLOCK))
308 return 0;
309 start_fsb += imap[n].br_blockcount;
310 count_fsb -= imap[n].br_blockcount;
Dave Chinner055388a2011-01-04 11:35:03 +1100311
312 if (imap[n].br_startblock == DELAYSTARTBLOCK)
313 found_delalloc = 1;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100314 }
315 }
Dave Chinner055388a2011-01-04 11:35:03 +1100316 if (!found_delalloc)
317 *prealloc = 1;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100318 return 0;
319}
320
Dave Chinner055388a2011-01-04 11:35:03 +1100321/*
322 * If we don't have a user specified preallocation size, dynamically increase
323 * the preallocation size as the size of the file grows. Cap the maximum size
324 * at a single extent or less if the filesystem is near full. The closer the
325 * filesystem is to full, the smaller the maximum prealocation.
326 */
327STATIC xfs_fsblock_t
328xfs_iomap_prealloc_size(
329 struct xfs_mount *mp,
330 struct xfs_inode *ip)
331{
332 xfs_fsblock_t alloc_blocks = 0;
333
334 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
335 int shift = 0;
336 int64_t freesp;
337
Dave Chinnerb8fc82632011-01-27 12:14:12 +1100338 /*
339 * rounddown_pow_of_two() returns an undefined result
340 * if we pass in alloc_blocks = 0. Hence the "+ 1" to
341 * ensure we always pass in a non-zero value.
342 */
343 alloc_blocks = XFS_B_TO_FSB(mp, ip->i_size) + 1;
Dave Chinner055388a2011-01-04 11:35:03 +1100344 alloc_blocks = XFS_FILEOFF_MIN(MAXEXTLEN,
345 rounddown_pow_of_two(alloc_blocks));
346
347 xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
348 freesp = mp->m_sb.sb_fdblocks;
349 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
350 shift = 2;
351 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
352 shift++;
353 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
354 shift++;
355 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
356 shift++;
357 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
358 shift++;
359 }
360 if (shift)
361 alloc_blocks >>= shift;
362 }
363
364 if (alloc_blocks < mp->m_writeio_blocks)
365 alloc_blocks = mp->m_writeio_blocks;
366
367 return alloc_blocks;
368}
369
Christoph Hellwiga206c812010-12-10 08:42:20 +0000370int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371xfs_iomap_write_delay(
372 xfs_inode_t *ip,
Nathan Scottf403b7f2005-05-05 13:33:40 -0700373 xfs_off_t offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 size_t count,
Christoph Hellwig405f8042010-12-10 08:42:19 +0000375 xfs_bmbt_irec_t *ret_imap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376{
377 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 xfs_fileoff_t offset_fsb;
379 xfs_fileoff_t last_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100380 xfs_off_t aligned_offset;
381 xfs_fileoff_t ioalign;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100382 xfs_extlen_t extsz;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 int nimaps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
Dave Chinner8de2bf92009-04-06 18:49:12 +0200385 int prealloc, flushed = 0;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100386 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000388 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
390 /*
391 * Make sure that the dquots are there. This doesn't hold
392 * the ilock across a disk read.
393 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200394 error = xfs_qm_dqattach_locked(ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 if (error)
396 return XFS_ERROR(error);
397
David Chinner957d0eb2007-06-18 16:50:37 +1000398 extsz = xfs_get_extsz_hint(ip);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100399 offset_fsb = XFS_B_TO_FSBT(mp, offset);
400
Dave Chinner055388a2011-01-04 11:35:03 +1100401
Lachlan McIlroy9f6c92b2008-12-22 17:56:49 +1100402 error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
Christoph Hellwig405f8042010-12-10 08:42:19 +0000403 imap, XFS_WRITE_IMAPS, &prealloc);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100404 if (error)
405 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
Dave Chinner8de2bf92009-04-06 18:49:12 +0200407retry:
Nathan Scottdd9f4382006-01-11 15:28:28 +1100408 if (prealloc) {
Dave Chinner055388a2011-01-04 11:35:03 +1100409 xfs_fsblock_t alloc_blocks = xfs_iomap_prealloc_size(mp, ip);
410
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
412 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
Dave Chinner055388a2011-01-04 11:35:03 +1100413 last_fsb = ioalign + alloc_blocks;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100414 } else {
415 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 }
Nathan Scottdd9f4382006-01-11 15:28:28 +1100417
418 if (prealloc || extsz) {
Lachlan McIlroy9f6c92b2008-12-22 17:56:49 +1100419 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100420 if (error)
421 return error;
422 }
423
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 nimaps = XFS_WRITE_IMAPS;
Christoph Hellwig44032802011-09-18 20:40:48 +0000425 error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb,
426 imap, &nimaps, XFS_BMAPI_ENTIRE);
Dave Chinner055388a2011-01-04 11:35:03 +1100427 switch (error) {
428 case 0:
429 case ENOSPC:
430 case EDQUOT:
431 break;
432 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 return XFS_ERROR(error);
Dave Chinner055388a2011-01-04 11:35:03 +1100434 }
Nathan Scottdd9f4382006-01-11 15:28:28 +1100435
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 /*
Dave Chinner055388a2011-01-04 11:35:03 +1100437 * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. For
438 * ENOSPC, * flush all other inodes with delalloc blocks to free up
439 * some of the excess reserved metadata space. For both cases, retry
440 * without EOF preallocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 */
442 if (nimaps == 0) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000443 trace_xfs_delalloc_enospc(ip, offset, count);
Dave Chinner8de2bf92009-04-06 18:49:12 +0200444 if (flushed)
Dave Chinner055388a2011-01-04 11:35:03 +1100445 return XFS_ERROR(error ? error : ENOSPC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446
Dave Chinner055388a2011-01-04 11:35:03 +1100447 if (error == ENOSPC) {
448 xfs_iunlock(ip, XFS_ILOCK_EXCL);
449 xfs_flush_inodes(ip);
450 xfs_ilock(ip, XFS_ILOCK_EXCL);
451 }
Dave Chinner8de2bf92009-04-06 18:49:12 +0200452
453 flushed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 error = 0;
Dave Chinner8de2bf92009-04-06 18:49:12 +0200455 prealloc = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 goto retry;
457 }
458
David Chinner86c4d622008-04-29 12:53:21 +1000459 if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
Dave Chinner6d4a8ec2011-03-07 10:06:35 +1100460 return xfs_alert_fsblock_zero(ip, &imap[0]);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100461
462 *ret_imap = imap[0];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 return 0;
464}
465
466/*
467 * Pass in a delayed allocate extent, convert it to real extents;
468 * return to the caller the extent we create which maps on top of
469 * the originating callers request.
470 *
471 * Called without a lock on the inode.
David Chinnere4143a12007-11-23 16:29:11 +1100472 *
473 * We no longer bother to look at the incoming map - all we have to
474 * guarantee is that whatever we allocate fills the required range.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 */
Christoph Hellwiga206c812010-12-10 08:42:20 +0000476int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477xfs_iomap_write_allocate(
478 xfs_inode_t *ip,
Nathan Scottf403b7f2005-05-05 13:33:40 -0700479 xfs_off_t offset,
Nathan Scott24e17b52005-05-05 13:33:20 -0700480 size_t count,
Christoph Hellwig405f8042010-12-10 08:42:19 +0000481 xfs_bmbt_irec_t *imap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482{
483 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 xfs_fileoff_t offset_fsb, last_block;
485 xfs_fileoff_t end_fsb, map_start_fsb;
486 xfs_fsblock_t first_block;
487 xfs_bmap_free_t free_list;
488 xfs_filblks_t count_fsb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 xfs_trans_t *tp;
David Chinnere4143a12007-11-23 16:29:11 +1100490 int nimaps, committed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 int error = 0;
492 int nres;
493
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 /*
495 * Make sure that the dquots are there.
496 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200497 error = xfs_qm_dqattach(ip, 0);
498 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 return XFS_ERROR(error);
500
Nathan Scott24e17b52005-05-05 13:33:20 -0700501 offset_fsb = XFS_B_TO_FSBT(mp, offset);
Christoph Hellwig30704512010-06-24 11:42:19 +1000502 count_fsb = imap->br_blockcount;
503 map_start_fsb = imap->br_startoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
505 XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
506
507 while (count_fsb != 0) {
508 /*
509 * Set up a transaction with which to allocate the
510 * backing store for the file. Do allocations in a
511 * loop until we get some space in the range we are
512 * interested in. The other space that might be allocated
513 * is in the delayed allocation extent on which we sit
514 * but before our buffer starts.
515 */
516
517 nimaps = 0;
518 while (nimaps == 0) {
519 tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
David Chinner84e1e992007-06-18 16:50:27 +1000520 tp->t_flags |= XFS_TRANS_RESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
522 error = xfs_trans_reserve(tp, nres,
523 XFS_WRITE_LOG_RES(mp),
524 0, XFS_TRANS_PERM_LOG_RES,
525 XFS_WRITE_LOG_COUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 if (error) {
527 xfs_trans_cancel(tp, 0);
528 return XFS_ERROR(error);
529 }
530 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigddc34152011-09-19 15:00:54 +0000531 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532
Eric Sandeen9d87c312009-01-14 23:22:07 -0600533 xfs_bmap_init(&free_list, &first_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 /*
David Chinnere4143a12007-11-23 16:29:11 +1100536 * it is possible that the extents have changed since
537 * we did the read call as we dropped the ilock for a
538 * while. We have to be careful about truncates or hole
539 * punchs here - we are not allowed to allocate
540 * non-delalloc blocks here.
541 *
542 * The only protection against truncation is the pages
543 * for the range we are being asked to convert are
544 * locked and hence a truncate will block on them
545 * first.
546 *
547 * As a result, if we go beyond the range we really
548 * need and hit an delalloc extent boundary followed by
549 * a hole while we have excess blocks in the map, we
550 * will fill the hole incorrectly and overrun the
551 * transaction reservation.
552 *
553 * Using a single map prevents this as we are forced to
554 * check each map we look for overlap with the desired
555 * range and abort as soon as we find it. Also, given
556 * that we only return a single map, having one beyond
557 * what we can return is probably a bit silly.
558 *
559 * We also need to check that we don't go beyond EOF;
560 * this is a truncate optimisation as a truncate sets
561 * the new file size before block on the pages we
562 * currently have locked under writeback. Because they
563 * are about to be tossed, we don't need to write them
564 * back....
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 */
David Chinnere4143a12007-11-23 16:29:11 +1100566 nimaps = 1;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000567 end_fsb = XFS_B_TO_FSB(mp, ip->i_size);
David Chinner7c9ef852008-04-10 12:21:59 +1000568 error = xfs_bmap_last_offset(NULL, ip, &last_block,
569 XFS_DATA_FORK);
570 if (error)
571 goto trans_cancel;
572
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
574 if ((map_start_fsb + count_fsb) > last_block) {
575 count_fsb = last_block - map_start_fsb;
576 if (count_fsb == 0) {
577 error = EAGAIN;
578 goto trans_cancel;
579 }
580 }
581
Christoph Hellwig30704512010-06-24 11:42:19 +1000582 /*
Christoph Hellwig30704512010-06-24 11:42:19 +1000583 * From this point onwards we overwrite the imap
584 * pointer that the caller gave to us.
585 */
Dave Chinnerc0dc7822011-09-18 20:40:52 +0000586 error = xfs_bmapi_write(tp, ip, map_start_fsb,
587 count_fsb, 0, &first_block, 1,
588 imap, &nimaps, &free_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 if (error)
590 goto trans_cancel;
591
Eric Sandeenf7c99b62007-02-10 18:37:16 +1100592 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 if (error)
594 goto trans_cancel;
595
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000596 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 if (error)
598 goto error0;
599
600 xfs_iunlock(ip, XFS_ILOCK_EXCL);
601 }
602
603 /*
604 * See if we were able to allocate an extent that
605 * covers at least part of the callers request
606 */
Christoph Hellwig30704512010-06-24 11:42:19 +1000607 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
Dave Chinner6d4a8ec2011-03-07 10:06:35 +1100608 return xfs_alert_fsblock_zero(ip, imap);
David Chinner86c4d622008-04-29 12:53:21 +1000609
Christoph Hellwig30704512010-06-24 11:42:19 +1000610 if ((offset_fsb >= imap->br_startoff) &&
611 (offset_fsb < (imap->br_startoff +
612 imap->br_blockcount))) {
David Chinnere4143a12007-11-23 16:29:11 +1100613 XFS_STATS_INC(xs_xstrat_quick);
614 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 }
616
David Chinnere4143a12007-11-23 16:29:11 +1100617 /*
618 * So far we have not mapped the requested part of the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 * file, just surrounding data, try again.
620 */
Christoph Hellwig30704512010-06-24 11:42:19 +1000621 count_fsb -= imap->br_blockcount;
622 map_start_fsb = imap->br_startoff + imap->br_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 }
624
625trans_cancel:
626 xfs_bmap_cancel(&free_list);
627 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
628error0:
629 xfs_iunlock(ip, XFS_ILOCK_EXCL);
630 return XFS_ERROR(error);
631}
632
633int
634xfs_iomap_write_unwritten(
635 xfs_inode_t *ip,
Nathan Scottf403b7f2005-05-05 13:33:40 -0700636 xfs_off_t offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 size_t count)
638{
639 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 xfs_fileoff_t offset_fsb;
641 xfs_filblks_t count_fsb;
642 xfs_filblks_t numblks_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100643 xfs_fsblock_t firstfsb;
644 int nimaps;
645 xfs_trans_t *tp;
646 xfs_bmbt_irec_t imap;
647 xfs_bmap_free_t free_list;
648 uint resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 int committed;
650 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000652 trace_xfs_unwritten_convert(ip, offset, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
654 offset_fsb = XFS_B_TO_FSBT(mp, offset);
655 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
656 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
657
Lachlan McIlroy4ddd8bb2008-06-27 13:32:53 +1000658 /*
659 * Reserve enough blocks in this transaction for two complete extent
660 * btree splits. We may be converting the middle part of an unwritten
661 * extent and in this case we will insert two new extents in the btree
662 * each of which could cause a full split.
663 *
664 * This reservation amount will be used in the first call to
665 * xfs_bmbt_split() to select an AG with enough space to satisfy the
666 * rest of the operation.
667 */
Nathan Scottdd9f4382006-01-11 15:28:28 +1100668 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669
Nathan Scottdd9f4382006-01-11 15:28:28 +1100670 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 /*
672 * set up a transaction to convert the range of extents
673 * from unwritten to real. Do allocations in a loop until
674 * we have covered the range passed in.
Christoph Hellwig80641dc2009-10-19 04:00:03 +0000675 *
676 * Note that we open code the transaction allocation here
677 * to pass KM_NOFS--we can't risk to recursing back into
678 * the filesystem here as we might be asked to write out
679 * the same inode that we complete here and might deadlock
680 * on the iolock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 */
Christoph Hellwig80641dc2009-10-19 04:00:03 +0000682 xfs_wait_for_freeze(mp, SB_FREEZE_TRANS);
683 tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS);
David Chinner84e1e992007-06-18 16:50:27 +1000684 tp->t_flags |= XFS_TRANS_RESERVE;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100685 error = xfs_trans_reserve(tp, resblks,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 XFS_WRITE_LOG_RES(mp), 0,
687 XFS_TRANS_PERM_LOG_RES,
688 XFS_WRITE_LOG_COUNT);
689 if (error) {
690 xfs_trans_cancel(tp, 0);
Nathan Scott572d95f2006-09-28 11:03:20 +1000691 return XFS_ERROR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 }
693
694 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigddc34152011-09-19 15:00:54 +0000695 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696
697 /*
698 * Modify the unwritten extent state of the buffer.
699 */
Eric Sandeen9d87c312009-01-14 23:22:07 -0600700 xfs_bmap_init(&free_list, &firstfsb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 nimaps = 1;
Dave Chinnerc0dc7822011-09-18 20:40:52 +0000702 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
703 XFS_BMAPI_CONVERT, &firstfsb,
Christoph Hellwigb4e91812010-06-23 18:11:15 +1000704 1, &imap, &nimaps, &free_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 if (error)
706 goto error_on_bmapi_transaction;
707
Eric Sandeenf7c99b62007-02-10 18:37:16 +1100708 error = xfs_bmap_finish(&(tp), &(free_list), &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 if (error)
710 goto error_on_bmapi_transaction;
711
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000712 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 xfs_iunlock(ip, XFS_ILOCK_EXCL);
714 if (error)
Nathan Scott572d95f2006-09-28 11:03:20 +1000715 return XFS_ERROR(error);
Nathan Scottdd9f4382006-01-11 15:28:28 +1100716
David Chinner86c4d622008-04-29 12:53:21 +1000717 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
Dave Chinner6d4a8ec2011-03-07 10:06:35 +1100718 return xfs_alert_fsblock_zero(ip, &imap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719
720 if ((numblks_fsb = imap.br_blockcount) == 0) {
721 /*
722 * The numblks_fsb value should always get
723 * smaller, otherwise the loop is stuck.
724 */
725 ASSERT(imap.br_blockcount);
726 break;
727 }
728 offset_fsb += numblks_fsb;
729 count_fsb -= numblks_fsb;
730 } while (count_fsb > 0);
731
732 return 0;
733
734error_on_bmapi_transaction:
735 xfs_bmap_cancel(&free_list);
736 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
737 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 return XFS_ERROR(error);
739}