blob: 092d652bc03df3153b6bcf427ed972ffe480fd86 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * 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"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Christoph Hellwig4fb6e8a2014-11-28 14:25:04 +110020#include "xfs_format.h"
Dave Chinner239880e2013-10-23 10:50:10 +110021#include "xfs_log_format.h"
22#include "xfs_trans_resv.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_sb.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_mount.h"
Dave Chinner239880e2013-10-23 10:50:10 +110026#include "xfs_trans.h"
Nathan Scotta844f452005-11-02 14:38:42 +110027#include "xfs_buf_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_trans_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_error.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000030#include "xfs_trace.h"
Dave Chinner239880e2013-10-23 10:50:10 +110031#include "xfs_log.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
33
34kmem_zone_t *xfs_buf_item_zone;
35
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +100036static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip)
37{
38 return container_of(lip, struct xfs_buf_log_item, bli_item);
39}
40
Dave Chinnerc90821a2010-12-03 17:00:52 +110041STATIC void xfs_buf_do_callbacks(struct xfs_buf *bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Dave Chinner166d1362013-08-12 20:50:04 +100043static inline int
44xfs_buf_log_format_size(
45 struct xfs_buf_log_format *blfp)
46{
47 return offsetof(struct xfs_buf_log_format, blf_data_map) +
48 (blfp->blf_map_size * sizeof(blfp->blf_data_map[0]));
49}
50
Linus Torvalds1da177e2005-04-16 15:20:36 -070051/*
52 * This returns the number of log iovecs needed to log the
53 * given buf log item.
54 *
55 * It calculates this as 1 iovec for the buf log format structure
56 * and 1 for each stretch of non-contiguous chunks to be logged.
57 * Contiguous chunks are logged in a single iovec.
58 *
59 * If the XFS_BLI_STALE flag has been set, then log nothing.
60 */
Dave Chinner166d1362013-08-12 20:50:04 +100061STATIC void
Dave Chinner372cc85e2012-06-22 18:50:12 +100062xfs_buf_item_size_segment(
63 struct xfs_buf_log_item *bip,
Dave Chinner166d1362013-08-12 20:50:04 +100064 struct xfs_buf_log_format *blfp,
65 int *nvecs,
66 int *nbytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -070067{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +100068 struct xfs_buf *bp = bip->bli_buf;
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +100069 int next_bit;
70 int last_bit;
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
Dave Chinner372cc85e2012-06-22 18:50:12 +100072 last_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
73 if (last_bit == -1)
Dave Chinner166d1362013-08-12 20:50:04 +100074 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
Dave Chinner372cc85e2012-06-22 18:50:12 +100076 /*
77 * initial count for a dirty buffer is 2 vectors - the format structure
78 * and the first dirty region.
79 */
Dave Chinner166d1362013-08-12 20:50:04 +100080 *nvecs += 2;
81 *nbytes += xfs_buf_log_format_size(blfp) + XFS_BLF_CHUNK;
Dave Chinner372cc85e2012-06-22 18:50:12 +100082
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 while (last_bit != -1) {
84 /*
85 * This takes the bit number to start looking from and
86 * returns the next set bit from there. It returns -1
87 * if there are no more bits set or the start bit is
88 * beyond the end of the bitmap.
89 */
Dave Chinner372cc85e2012-06-22 18:50:12 +100090 next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
91 last_bit + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 /*
93 * If we run out of bits, leave the loop,
94 * else if we find a new set of bits bump the number of vecs,
95 * else keep scanning the current set of bits.
96 */
97 if (next_bit == -1) {
Dave Chinner372cc85e2012-06-22 18:50:12 +100098 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 } else if (next_bit != last_bit + 1) {
100 last_bit = next_bit;
Dave Chinner166d1362013-08-12 20:50:04 +1000101 (*nvecs)++;
Dave Chinnerc1155412010-05-07 11:05:19 +1000102 } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) !=
103 (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) +
104 XFS_BLF_CHUNK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 last_bit = next_bit;
Dave Chinner166d1362013-08-12 20:50:04 +1000106 (*nvecs)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 } else {
108 last_bit++;
109 }
Dave Chinner166d1362013-08-12 20:50:04 +1000110 *nbytes += XFS_BLF_CHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 }
Dave Chinner372cc85e2012-06-22 18:50:12 +1000112}
113
114/*
115 * This returns the number of log iovecs needed to log the given buf log item.
116 *
117 * It calculates this as 1 iovec for the buf log format structure and 1 for each
118 * stretch of non-contiguous chunks to be logged. Contiguous chunks are logged
119 * in a single iovec.
120 *
121 * Discontiguous buffers need a format structure per region that that is being
122 * logged. This makes the changes in the buffer appear to log recovery as though
123 * they came from separate buffers, just like would occur if multiple buffers
124 * were used instead of a single discontiguous buffer. This enables
125 * discontiguous buffers to be in-memory constructs, completely transparent to
126 * what ends up on disk.
127 *
128 * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log
129 * format structures.
130 */
Dave Chinner166d1362013-08-12 20:50:04 +1000131STATIC void
Dave Chinner372cc85e2012-06-22 18:50:12 +1000132xfs_buf_item_size(
Dave Chinner166d1362013-08-12 20:50:04 +1000133 struct xfs_log_item *lip,
134 int *nvecs,
135 int *nbytes)
Dave Chinner372cc85e2012-06-22 18:50:12 +1000136{
137 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
Dave Chinner372cc85e2012-06-22 18:50:12 +1000138 int i;
139
140 ASSERT(atomic_read(&bip->bli_refcount) > 0);
141 if (bip->bli_flags & XFS_BLI_STALE) {
142 /*
143 * The buffer is stale, so all we need to log
144 * is the buf log format structure with the
145 * cancel flag in it.
146 */
147 trace_xfs_buf_item_size_stale(bip);
Mark Tinguelyb9438172012-12-04 17:18:03 -0600148 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
Dave Chinner166d1362013-08-12 20:50:04 +1000149 *nvecs += bip->bli_format_count;
150 for (i = 0; i < bip->bli_format_count; i++) {
151 *nbytes += xfs_buf_log_format_size(&bip->bli_formats[i]);
152 }
153 return;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000154 }
155
156 ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
157
Dave Chinner5f6bed72013-06-27 16:04:52 +1000158 if (bip->bli_flags & XFS_BLI_ORDERED) {
159 /*
160 * The buffer has been logged just to order it.
161 * It is not being included in the transaction
162 * commit, so no vectors are used at all.
163 */
164 trace_xfs_buf_item_size_ordered(bip);
Dave Chinner166d1362013-08-12 20:50:04 +1000165 *nvecs = XFS_LOG_VEC_ORDERED;
166 return;
Dave Chinner5f6bed72013-06-27 16:04:52 +1000167 }
168
Dave Chinner372cc85e2012-06-22 18:50:12 +1000169 /*
170 * the vector count is based on the number of buffer vectors we have
171 * dirty bits in. This will only be greater than one when we have a
172 * compound buffer with more than one segment dirty. Hence for compound
173 * buffers we need to track which segment the dirty bits correspond to,
174 * and when we move from one segment to the next increment the vector
175 * count for the extra buf log format structure that will need to be
176 * written.
177 */
Dave Chinner372cc85e2012-06-22 18:50:12 +1000178 for (i = 0; i < bip->bli_format_count; i++) {
Dave Chinner166d1362013-08-12 20:50:04 +1000179 xfs_buf_item_size_segment(bip, &bip->bli_formats[i],
180 nvecs, nbytes);
Dave Chinner372cc85e2012-06-22 18:50:12 +1000181 }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000182 trace_xfs_buf_item_size(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183}
184
Christoph Hellwig12343512013-12-13 11:00:43 +1100185static inline void
Christoph Hellwig7aeb7222013-12-13 11:00:43 +1100186xfs_buf_item_copy_iovec(
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100187 struct xfs_log_vec *lv,
Christoph Hellwig12343512013-12-13 11:00:43 +1100188 struct xfs_log_iovec **vecp,
Christoph Hellwig7aeb7222013-12-13 11:00:43 +1100189 struct xfs_buf *bp,
190 uint offset,
191 int first_bit,
192 uint nbits)
193{
194 offset += first_bit * XFS_BLF_CHUNK;
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100195 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BCHUNK,
Christoph Hellwig12343512013-12-13 11:00:43 +1100196 xfs_buf_offset(bp, offset),
197 nbits * XFS_BLF_CHUNK);
Christoph Hellwig7aeb7222013-12-13 11:00:43 +1100198}
199
200static inline bool
201xfs_buf_item_straddle(
202 struct xfs_buf *bp,
203 uint offset,
204 int next_bit,
205 int last_bit)
206{
207 return xfs_buf_offset(bp, offset + (next_bit << XFS_BLF_SHIFT)) !=
208 (xfs_buf_offset(bp, offset + (last_bit << XFS_BLF_SHIFT)) +
209 XFS_BLF_CHUNK);
210}
211
Christoph Hellwig12343512013-12-13 11:00:43 +1100212static void
Dave Chinner372cc85e2012-06-22 18:50:12 +1000213xfs_buf_item_format_segment(
214 struct xfs_buf_log_item *bip,
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100215 struct xfs_log_vec *lv,
Christoph Hellwig12343512013-12-13 11:00:43 +1100216 struct xfs_log_iovec **vecp,
Dave Chinner372cc85e2012-06-22 18:50:12 +1000217 uint offset,
218 struct xfs_buf_log_format *blfp)
219{
220 struct xfs_buf *bp = bip->bli_buf;
221 uint base_size;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000222 int first_bit;
223 int last_bit;
224 int next_bit;
225 uint nbits;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000226
227 /* copy the flags across from the base format item */
Mark Tinguelyb9438172012-12-04 17:18:03 -0600228 blfp->blf_flags = bip->__bli_format.blf_flags;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000229
230 /*
231 * Base size is the actual size of the ondisk structure - it reflects
232 * the actual size of the dirty bitmap rather than the size of the in
233 * memory structure.
234 */
Dave Chinner166d1362013-08-12 20:50:04 +1000235 base_size = xfs_buf_log_format_size(blfp);
Mark Tinguely820a5542012-12-04 17:18:04 -0600236
Mark Tinguely820a5542012-12-04 17:18:04 -0600237 first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
238 if (!(bip->bli_flags & XFS_BLI_STALE) && first_bit == -1) {
239 /*
240 * If the map is not be dirty in the transaction, mark
241 * the size as zero and do not advance the vector pointer.
242 */
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100243 return;
Mark Tinguely820a5542012-12-04 17:18:04 -0600244 }
245
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100246 blfp = xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BFORMAT, blfp, base_size);
247 blfp->blf_size = 1;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000248
249 if (bip->bli_flags & XFS_BLI_STALE) {
250 /*
251 * The buffer is stale, so all we need to log
252 * is the buf log format structure with the
253 * cancel flag in it.
254 */
255 trace_xfs_buf_item_format_stale(bip);
256 ASSERT(blfp->blf_flags & XFS_BLF_CANCEL);
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100257 return;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000258 }
259
Dave Chinner5f6bed72013-06-27 16:04:52 +1000260
Dave Chinner372cc85e2012-06-22 18:50:12 +1000261 /*
262 * Fill in an iovec for each set of contiguous chunks.
263 */
Dave Chinner372cc85e2012-06-22 18:50:12 +1000264 last_bit = first_bit;
265 nbits = 1;
266 for (;;) {
267 /*
268 * This takes the bit number to start looking from and
269 * returns the next set bit from there. It returns -1
270 * if there are no more bits set or the start bit is
271 * beyond the end of the bitmap.
272 */
273 next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
274 (uint)last_bit + 1);
275 /*
Christoph Hellwig7aeb7222013-12-13 11:00:43 +1100276 * If we run out of bits fill in the last iovec and get out of
277 * the loop. Else if we start a new set of bits then fill in
278 * the iovec for the series we were looking at and start
279 * counting the bits in the new one. Else we're still in the
280 * same set of bits so just keep counting and scanning.
Dave Chinner372cc85e2012-06-22 18:50:12 +1000281 */
282 if (next_bit == -1) {
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100283 xfs_buf_item_copy_iovec(lv, vecp, bp, offset,
Christoph Hellwig7aeb7222013-12-13 11:00:43 +1100284 first_bit, nbits);
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100285 blfp->blf_size++;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000286 break;
Christoph Hellwig7aeb7222013-12-13 11:00:43 +1100287 } else if (next_bit != last_bit + 1 ||
288 xfs_buf_item_straddle(bp, offset, next_bit, last_bit)) {
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100289 xfs_buf_item_copy_iovec(lv, vecp, bp, offset,
Christoph Hellwig12343512013-12-13 11:00:43 +1100290 first_bit, nbits);
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100291 blfp->blf_size++;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000292 first_bit = next_bit;
293 last_bit = next_bit;
294 nbits = 1;
295 } else {
296 last_bit++;
297 nbits++;
298 }
299 }
Dave Chinner372cc85e2012-06-22 18:50:12 +1000300}
301
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302/*
303 * This is called to fill in the vector of log iovecs for the
304 * given log buf item. It fills the first entry with a buf log
305 * format structure, and the rest point to contiguous chunks
306 * within the buffer.
307 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000308STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309xfs_buf_item_format(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000310 struct xfs_log_item *lip,
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100311 struct xfs_log_vec *lv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000313 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
Dave Chinner372cc85e2012-06-22 18:50:12 +1000314 struct xfs_buf *bp = bip->bli_buf;
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100315 struct xfs_log_iovec *vecp = NULL;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000316 uint offset = 0;
317 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
319 ASSERT(atomic_read(&bip->bli_refcount) > 0);
320 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
321 (bip->bli_flags & XFS_BLI_STALE));
Dave Chinner0d612fb2015-01-22 09:29:05 +1100322 ASSERT((bip->bli_flags & XFS_BLI_STALE) ||
323 (xfs_blft_from_flags(&bip->__bli_format) > XFS_BLFT_UNKNOWN_BUF
324 && xfs_blft_from_flags(&bip->__bli_format) < XFS_BLFT_MAX_BUF));
325
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
327 /*
Dave Chinnerccf7c232010-05-20 23:19:42 +1000328 * If it is an inode buffer, transfer the in-memory state to the
Dave Chinnerddf6ad02013-06-27 16:04:56 +1000329 * format flags and clear the in-memory state.
330 *
331 * For buffer based inode allocation, we do not transfer
Dave Chinnerccf7c232010-05-20 23:19:42 +1000332 * this state if the inode buffer allocation has not yet been committed
333 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
334 * correct replay of the inode allocation.
Dave Chinnerddf6ad02013-06-27 16:04:56 +1000335 *
336 * For icreate item based inode allocation, the buffers aren't written
337 * to the journal during allocation, and hence we should always tag the
338 * buffer as an inode buffer so that the correct unlinked list replay
339 * occurs during recovery.
Dave Chinnerccf7c232010-05-20 23:19:42 +1000340 */
341 if (bip->bli_flags & XFS_BLI_INODE_BUF) {
Dave Chinnerddf6ad02013-06-27 16:04:56 +1000342 if (xfs_sb_version_hascrc(&lip->li_mountp->m_sb) ||
343 !((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000344 xfs_log_item_in_current_chkpt(lip)))
Mark Tinguelyb9438172012-12-04 17:18:03 -0600345 bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF;
Dave Chinnerccf7c232010-05-20 23:19:42 +1000346 bip->bli_flags &= ~XFS_BLI_INODE_BUF;
347 }
348
Dave Chinner5f6bed72013-06-27 16:04:52 +1000349 if ((bip->bli_flags & (XFS_BLI_ORDERED|XFS_BLI_STALE)) ==
350 XFS_BLI_ORDERED) {
351 /*
352 * The buffer has been logged just to order it. It is not being
353 * included in the transaction commit, so don't format it.
354 */
355 trace_xfs_buf_item_format_ordered(bip);
356 return;
357 }
358
Dave Chinner372cc85e2012-06-22 18:50:12 +1000359 for (i = 0; i < bip->bli_format_count; i++) {
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100360 xfs_buf_item_format_segment(bip, lv, &vecp, offset,
Christoph Hellwig12343512013-12-13 11:00:43 +1100361 &bip->bli_formats[i]);
Dave Chinner372cc85e2012-06-22 18:50:12 +1000362 offset += bp->b_maps[i].bm_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 }
364
365 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 * Check to make sure everything is consistent.
367 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000368 trace_xfs_buf_item_format(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369}
370
371/*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000372 * This is called to pin the buffer associated with the buf log item in memory
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000373 * so it cannot be written out.
Dave Chinner64fc35d2010-05-07 11:04:34 +1000374 *
375 * We also always take a reference to the buffer log item here so that the bli
376 * is held while the item is pinned in memory. This means that we can
377 * unconditionally drop the reference count a transaction holds when the
378 * transaction is completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000380STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381xfs_buf_item_pin(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000382 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000384 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 ASSERT(atomic_read(&bip->bli_refcount) > 0);
387 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
Dave Chinner5f6bed72013-06-27 16:04:52 +1000388 (bip->bli_flags & XFS_BLI_ORDERED) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 (bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000390
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000391 trace_xfs_buf_item_pin(bip);
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000392
393 atomic_inc(&bip->bli_refcount);
394 atomic_inc(&bip->bli_buf->b_pin_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395}
396
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397/*
398 * This is called to unpin the buffer associated with the buf log
399 * item which was previously pinned with a call to xfs_buf_item_pin().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 *
401 * Also drop the reference to the buf item for the current transaction.
402 * If the XFS_BLI_STALE flag is set and we are the last reference,
403 * then free up the buf log item and unlock the buffer.
Christoph Hellwig9412e312010-06-23 18:11:15 +1000404 *
405 * If the remove flag is set we are called from uncommit in the
406 * forced-shutdown path. If that is true and the reference count on
407 * the log item is going to drop to zero we need to free the item's
408 * descriptor in the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000410STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411xfs_buf_item_unpin(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000412 struct xfs_log_item *lip,
Christoph Hellwig9412e312010-06-23 18:11:15 +1000413 int remove)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000415 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000416 xfs_buf_t *bp = bip->bli_buf;
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000417 struct xfs_ail *ailp = lip->li_ailp;
Dave Chinner8e123852010-03-08 11:26:03 +1100418 int stale = bip->bli_flags & XFS_BLI_STALE;
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000419 int freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200421 ASSERT(bp->b_fspriv == bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 ASSERT(atomic_read(&bip->bli_refcount) > 0);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000423
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000424 trace_xfs_buf_item_unpin(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
426 freed = atomic_dec_and_test(&bip->bli_refcount);
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000427
428 if (atomic_dec_and_test(&bp->b_pin_count))
429 wake_up_all(&bp->b_waiters);
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000430
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 if (freed && stale) {
432 ASSERT(bip->bli_flags & XFS_BLI_STALE);
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200433 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 ASSERT(XFS_BUF_ISSTALE(bp));
Mark Tinguelyb9438172012-12-04 17:18:03 -0600435 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000436
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000437 trace_xfs_buf_item_unpin_stale(bip);
438
Christoph Hellwig9412e312010-06-23 18:11:15 +1000439 if (remove) {
440 /*
Dave Chinnere34a3142011-01-27 12:13:35 +1100441 * If we are in a transaction context, we have to
442 * remove the log item from the transaction as we are
443 * about to release our reference to the buffer. If we
444 * don't, the unlock that occurs later in
445 * xfs_trans_uncommit() will try to reference the
Christoph Hellwig9412e312010-06-23 18:11:15 +1000446 * buffer which we no longer have a hold on.
447 */
Dave Chinnere34a3142011-01-27 12:13:35 +1100448 if (lip->li_desc)
449 xfs_trans_del_item(lip);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000450
451 /*
452 * Since the transaction no longer refers to the buffer,
453 * the buffer should no longer refer to the transaction.
454 */
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200455 bp->b_transp = NULL;
Christoph Hellwig9412e312010-06-23 18:11:15 +1000456 }
457
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 /*
459 * If we get called here because of an IO error, we may
David Chinner783a2f62008-10-30 17:39:58 +1100460 * or may not have the item on the AIL. xfs_trans_ail_delete()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 * will take care of that situation.
David Chinner783a2f62008-10-30 17:39:58 +1100462 * xfs_trans_ail_delete() drops the AIL lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 */
464 if (bip->bli_flags & XFS_BLI_STALE_INODE) {
Dave Chinnerc90821a2010-12-03 17:00:52 +1100465 xfs_buf_do_callbacks(bp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200466 bp->b_fspriv = NULL;
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200467 bp->b_iodone = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 } else {
David Chinner783a2f62008-10-30 17:39:58 +1100469 spin_lock(&ailp->xa_lock);
Dave Chinner04913fd2012-04-23 15:58:41 +1000470 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_LOG_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 xfs_buf_item_relse(bp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200472 ASSERT(bp->b_fspriv == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 }
474 xfs_buf_relse(bp);
Christoph Hellwig960c60a2012-04-23 15:58:38 +1000475 } else if (freed && remove) {
Dave Chinner137fff02012-11-02 14:23:12 +1100476 /*
477 * There are currently two references to the buffer - the active
478 * LRU reference and the buf log item. What we are about to do
479 * here - simulate a failed IO completion - requires 3
480 * references.
481 *
482 * The LRU reference is removed by the xfs_buf_stale() call. The
483 * buf item reference is removed by the xfs_buf_iodone()
484 * callback that is run by xfs_buf_do_callbacks() during ioend
485 * processing (via the bp->b_iodone callback), and then finally
486 * the ioend processing will drop the IO reference if the buffer
487 * is marked XBF_ASYNC.
488 *
489 * Hence we need to take an additional reference here so that IO
490 * completion processing doesn't free the buffer prematurely.
491 */
Christoph Hellwig960c60a2012-04-23 15:58:38 +1000492 xfs_buf_lock(bp);
Dave Chinner137fff02012-11-02 14:23:12 +1100493 xfs_buf_hold(bp);
494 bp->b_flags |= XBF_ASYNC;
Dave Chinner24513372014-06-25 14:58:08 +1000495 xfs_buf_ioerror(bp, -EIO);
Christoph Hellwig960c60a2012-04-23 15:58:38 +1000496 XFS_BUF_UNDONE(bp);
497 xfs_buf_stale(bp);
Dave Chinnere8aaba92014-10-02 09:04:22 +1000498 xfs_buf_ioend(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 }
500}
501
Dave Chinnerac8809f2013-12-12 16:34:38 +1100502/*
503 * Buffer IO error rate limiting. Limit it to no more than 10 messages per 30
504 * seconds so as to not spam logs too much on repeated detection of the same
505 * buffer being bad..
506 */
507
Dave Chinner02cc1872014-09-23 16:15:45 +1000508static DEFINE_RATELIMIT_STATE(xfs_buf_write_fail_rl_state, 30 * HZ, 10);
Dave Chinnerac8809f2013-12-12 16:34:38 +1100509
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000510STATIC uint
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000511xfs_buf_item_push(
512 struct xfs_log_item *lip,
513 struct list_head *buffer_list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000515 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
516 struct xfs_buf *bp = bip->bli_buf;
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000517 uint rval = XFS_ITEM_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
Chandra Seetharaman811e64c2011-07-22 23:40:27 +0000519 if (xfs_buf_ispinned(bp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 return XFS_ITEM_PINNED;
Brian Foster5337fe92013-02-11 10:08:21 -0500521 if (!xfs_buf_trylock(bp)) {
522 /*
523 * If we have just raced with a buffer being pinned and it has
524 * been marked stale, we could end up stalling until someone else
525 * issues a log force to unpin the stale buffer. Check for the
526 * race condition here so xfsaild recognizes the buffer is pinned
527 * and queues a log force to move it along.
528 */
529 if (xfs_buf_ispinned(bp))
530 return XFS_ITEM_PINNED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 return XFS_ITEM_LOCKED;
Brian Foster5337fe92013-02-11 10:08:21 -0500532 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000535
536 trace_xfs_buf_item_push(bip);
537
Dave Chinnerac8809f2013-12-12 16:34:38 +1100538 /* has a previous flush failed due to IO errors? */
539 if ((bp->b_flags & XBF_WRITE_FAIL) &&
Eric Sandeenfdadf262015-02-24 10:14:04 +1100540 ___ratelimit(&xfs_buf_write_fail_rl_state, "XFS: Failing async write")) {
Dave Chinnerac8809f2013-12-12 16:34:38 +1100541 xfs_warn(bp->b_target->bt_mount,
Eric Sandeenfdadf262015-02-24 10:14:04 +1100542"Failing async write on buffer block 0x%llx. Retrying async write.",
Dave Chinnerac8809f2013-12-12 16:34:38 +1100543 (long long)bp->b_bn);
544 }
545
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000546 if (!xfs_buf_delwri_queue(bp, buffer_list))
547 rval = XFS_ITEM_FLUSHING;
548 xfs_buf_unlock(bp);
549 return rval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550}
551
552/*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000553 * Release the buffer associated with the buf log item. If there is no dirty
554 * logged data associated with the buffer recorded in the buf log item, then
555 * free the buf log item and remove the reference to it in the buffer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 *
Dave Chinner64fc35d2010-05-07 11:04:34 +1000557 * This call ignores the recursion count. It is only called when the buffer
558 * should REALLY be unlocked, regardless of the recursion count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 *
Dave Chinner64fc35d2010-05-07 11:04:34 +1000560 * We unconditionally drop the transaction's reference to the log item. If the
561 * item was logged, then another reference was taken when it was pinned, so we
562 * can safely drop the transaction reference now. This also allows us to avoid
563 * potential races with the unpin code freeing the bli by not referencing the
564 * bli after we've dropped the reference count.
565 *
566 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
567 * if necessary but do not unlock the buffer. This is for support of
568 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
569 * free the item.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000571STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572xfs_buf_item_unlock(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000573 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000575 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
576 struct xfs_buf *bp = bip->bli_buf;
Dave Chinner5f6bed72013-06-27 16:04:52 +1000577 bool clean;
578 bool aborted;
579 int flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
Dave Chinner64fc35d2010-05-07 11:04:34 +1000581 /* Clear the buffer's association with this transaction. */
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200582 bp->b_transp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
584 /*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000585 * If this is a transaction abort, don't return early. Instead, allow
586 * the brelse to happen. Normally it would be done for stale
587 * (cancelled) buffers at unpin time, but we'll never go through the
588 * pin/unpin cycle if we abort inside commit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 */
Dave Chinner5f6bed72013-06-27 16:04:52 +1000590 aborted = (lip->li_flags & XFS_LI_ABORTED) ? true : false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 /*
Dave Chinner5f6bed72013-06-27 16:04:52 +1000592 * Before possibly freeing the buf item, copy the per-transaction state
593 * so we can reference it safely later after clearing it from the
594 * buffer log item.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 */
Dave Chinner5f6bed72013-06-27 16:04:52 +1000596 flags = bip->bli_flags;
597 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD | XFS_BLI_ORDERED);
Dave Chinner64fc35d2010-05-07 11:04:34 +1000598
599 /*
600 * If the buf item is marked stale, then don't do anything. We'll
601 * unlock the buffer and free the buf item when the buffer is unpinned
602 * for the last time.
603 */
Dave Chinner5f6bed72013-06-27 16:04:52 +1000604 if (flags & XFS_BLI_STALE) {
Dave Chinner64fc35d2010-05-07 11:04:34 +1000605 trace_xfs_buf_item_unlock_stale(bip);
Mark Tinguelyb9438172012-12-04 17:18:03 -0600606 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
Dave Chinner64fc35d2010-05-07 11:04:34 +1000607 if (!aborted) {
608 atomic_dec(&bip->bli_refcount);
609 return;
610 }
611 }
612
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000613 trace_xfs_buf_item_unlock(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
615 /*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000616 * If the buf item isn't tracking any data, free it, otherwise drop the
Dave Chinner3b190342013-01-21 23:53:55 +1100617 * reference we hold to it. If we are aborting the transaction, this may
618 * be the only reference to the buf item, so we free it anyway
619 * regardless of whether it is dirty or not. A dirty abort implies a
620 * shutdown, anyway.
Dave Chinner5f6bed72013-06-27 16:04:52 +1000621 *
622 * Ordered buffers are dirty but may have no recorded changes, so ensure
623 * we only release clean items here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 */
Dave Chinner5f6bed72013-06-27 16:04:52 +1000625 clean = (flags & XFS_BLI_DIRTY) ? false : true;
626 if (clean) {
627 int i;
628 for (i = 0; i < bip->bli_format_count; i++) {
629 if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map,
630 bip->bli_formats[i].blf_map_size)) {
631 clean = false;
632 break;
633 }
Mark Tinguelyc883d0c2012-12-04 17:18:05 -0600634 }
635 }
Dave Chinner46f9d2e2013-09-03 21:47:37 +1000636
637 /*
638 * Clean buffers, by definition, cannot be in the AIL. However, aborted
639 * buffers may be dirty and hence in the AIL. Therefore if we are
640 * aborting a buffer and we've just taken the last refernce away, we
641 * have to check if it is in the AIL before freeing it. We need to free
642 * it in this case, because an aborted transaction has already shut the
643 * filesystem down and this is the last chance we will have to do so.
644 */
645 if (atomic_dec_and_test(&bip->bli_refcount)) {
646 if (clean)
647 xfs_buf_item_relse(bp);
648 else if (aborted) {
649 ASSERT(XFS_FORCED_SHUTDOWN(lip->li_mountp));
650 if (lip->li_flags & XFS_LI_IN_AIL) {
Dave Chinner48852352013-09-24 16:01:13 +1000651 spin_lock(&lip->li_ailp->xa_lock);
Dave Chinner46f9d2e2013-09-03 21:47:37 +1000652 xfs_trans_ail_delete(lip->li_ailp, lip,
653 SHUTDOWN_LOG_IO_ERROR);
654 }
Dave Chinner3b190342013-01-21 23:53:55 +1100655 xfs_buf_item_relse(bp);
656 }
Dave Chinner46f9d2e2013-09-03 21:47:37 +1000657 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658
Dave Chinner5f6bed72013-06-27 16:04:52 +1000659 if (!(flags & XFS_BLI_HOLD))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661}
662
663/*
664 * This is called to find out where the oldest active copy of the
665 * buf log item in the on disk log resides now that the last log
666 * write of it completed at the given lsn.
667 * We always re-log all the dirty data in a buffer, so usually the
668 * latest copy in the on disk log is the only one that matters. For
669 * those cases we simply return the given lsn.
670 *
671 * The one exception to this is for buffers full of newly allocated
672 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
673 * flag set, indicating that only the di_next_unlinked fields from the
674 * inodes in the buffers will be replayed during recovery. If the
675 * original newly allocated inode images have not yet been flushed
676 * when the buffer is so relogged, then we need to make sure that we
677 * keep the old images in the 'active' portion of the log. We do this
678 * by returning the original lsn of that transaction here rather than
679 * the current one.
680 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000681STATIC xfs_lsn_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682xfs_buf_item_committed(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000683 struct xfs_log_item *lip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 xfs_lsn_t lsn)
685{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000686 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
687
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000688 trace_xfs_buf_item_committed(bip);
689
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000690 if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0)
691 return lip->li_lsn;
692 return lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693}
694
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000695STATIC void
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000696xfs_buf_item_committing(
697 struct xfs_log_item *lip,
698 xfs_lsn_t commit_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699{
700}
701
702/*
703 * This is the ops vector shared by all buf log items.
704 */
Christoph Hellwig272e42b2011-10-28 09:54:24 +0000705static const struct xfs_item_ops xfs_buf_item_ops = {
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000706 .iop_size = xfs_buf_item_size,
707 .iop_format = xfs_buf_item_format,
708 .iop_pin = xfs_buf_item_pin,
709 .iop_unpin = xfs_buf_item_unpin,
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000710 .iop_unlock = xfs_buf_item_unlock,
711 .iop_committed = xfs_buf_item_committed,
712 .iop_push = xfs_buf_item_push,
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000713 .iop_committing = xfs_buf_item_committing
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714};
715
Dave Chinner372cc85e2012-06-22 18:50:12 +1000716STATIC int
717xfs_buf_item_get_format(
718 struct xfs_buf_log_item *bip,
719 int count)
720{
721 ASSERT(bip->bli_formats == NULL);
722 bip->bli_format_count = count;
723
724 if (count == 1) {
Mark Tinguelyb9438172012-12-04 17:18:03 -0600725 bip->bli_formats = &bip->__bli_format;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000726 return 0;
727 }
728
729 bip->bli_formats = kmem_zalloc(count * sizeof(struct xfs_buf_log_format),
730 KM_SLEEP);
731 if (!bip->bli_formats)
Dave Chinner24513372014-06-25 14:58:08 +1000732 return -ENOMEM;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000733 return 0;
734}
735
736STATIC void
737xfs_buf_item_free_format(
738 struct xfs_buf_log_item *bip)
739{
Mark Tinguelyb9438172012-12-04 17:18:03 -0600740 if (bip->bli_formats != &bip->__bli_format) {
Dave Chinner372cc85e2012-06-22 18:50:12 +1000741 kmem_free(bip->bli_formats);
742 bip->bli_formats = NULL;
743 }
744}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745
746/*
747 * Allocate a new buf log item to go with the given buffer.
748 * Set the buffer's b_fsprivate field to point to the new
749 * buf log item. If there are other item's attached to the
750 * buffer (see xfs_buf_attach_iodone() below), then put the
751 * buf log item at the front.
752 */
753void
754xfs_buf_item_init(
755 xfs_buf_t *bp,
756 xfs_mount_t *mp)
757{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200758 xfs_log_item_t *lip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 xfs_buf_log_item_t *bip;
760 int chunks;
761 int map_size;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000762 int error;
763 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
765 /*
766 * Check to see if there is already a buf log item for
767 * this buffer. If there is, it is guaranteed to be
768 * the first. If we do already have one, there is
769 * nothing to do here so return.
770 */
Dave Chinnerebad8612010-09-22 10:47:20 +1000771 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200772 if (lip != NULL && lip->li_type == XFS_LI_BUF)
773 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
Dave Chinner372cc85e2012-06-22 18:50:12 +1000775 bip = kmem_zone_zalloc(xfs_buf_item_zone, KM_SLEEP);
Dave Chinner43f5efc2010-03-23 10:10:00 +1100776 xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 bip->bli_buf = bp;
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000778 xfs_buf_hold(bp);
Dave Chinner372cc85e2012-06-22 18:50:12 +1000779
780 /*
781 * chunks is the number of XFS_BLF_CHUNK size pieces the buffer
782 * can be divided into. Make sure not to truncate any pieces.
783 * map_size is the size of the bitmap needed to describe the
784 * chunks of the buffer.
785 *
786 * Discontiguous buffer support follows the layout of the underlying
787 * buffer. This makes the implementation as simple as possible.
788 */
789 error = xfs_buf_item_get_format(bip, bp->b_map_count);
790 ASSERT(error == 0);
791
792 for (i = 0; i < bip->bli_format_count; i++) {
793 chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len),
794 XFS_BLF_CHUNK);
795 map_size = DIV_ROUND_UP(chunks, NBWORD);
796
797 bip->bli_formats[i].blf_type = XFS_LI_BUF;
798 bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn;
799 bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len;
800 bip->bli_formats[i].blf_map_size = map_size;
801 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 /*
804 * Put the buf item into the list of items attached to the
805 * buffer at the front.
806 */
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200807 if (bp->b_fspriv)
808 bip->bli_item.li_bio_list = bp->b_fspriv;
809 bp->b_fspriv = bip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810}
811
812
813/*
814 * Mark bytes first through last inclusive as dirty in the buf
815 * item's bitmap.
816 */
Dave Chinner632b89e2013-10-29 22:11:58 +1100817static void
Dave Chinner372cc85e2012-06-22 18:50:12 +1000818xfs_buf_item_log_segment(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 uint first,
Dave Chinner372cc85e2012-06-22 18:50:12 +1000820 uint last,
821 uint *map)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822{
823 uint first_bit;
824 uint last_bit;
825 uint bits_to_set;
826 uint bits_set;
827 uint word_num;
828 uint *wordp;
829 uint bit;
830 uint end_bit;
831 uint mask;
832
833 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 * Convert byte offsets to bit numbers.
835 */
Dave Chinnerc1155412010-05-07 11:05:19 +1000836 first_bit = first >> XFS_BLF_SHIFT;
837 last_bit = last >> XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
839 /*
840 * Calculate the total number of bits to be set.
841 */
842 bits_to_set = last_bit - first_bit + 1;
843
844 /*
845 * Get a pointer to the first word in the bitmap
846 * to set a bit in.
847 */
848 word_num = first_bit >> BIT_TO_WORD_SHIFT;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000849 wordp = &map[word_num];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850
851 /*
852 * Calculate the starting bit in the first word.
853 */
854 bit = first_bit & (uint)(NBWORD - 1);
855
856 /*
857 * First set any bits in the first word of our range.
858 * If it starts at bit 0 of the word, it will be
859 * set below rather than here. That is what the variable
860 * bit tells us. The variable bits_set tracks the number
861 * of bits that have been set so far. End_bit is the number
862 * of the last bit to be set in this word plus one.
863 */
864 if (bit) {
865 end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
866 mask = ((1 << (end_bit - bit)) - 1) << bit;
867 *wordp |= mask;
868 wordp++;
869 bits_set = end_bit - bit;
870 } else {
871 bits_set = 0;
872 }
873
874 /*
875 * Now set bits a whole word at a time that are between
876 * first_bit and last_bit.
877 */
878 while ((bits_to_set - bits_set) >= NBWORD) {
879 *wordp |= 0xffffffff;
880 bits_set += NBWORD;
881 wordp++;
882 }
883
884 /*
885 * Finally, set any bits left to be set in one last partial word.
886 */
887 end_bit = bits_to_set - bits_set;
888 if (end_bit) {
889 mask = (1 << end_bit) - 1;
890 *wordp |= mask;
891 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892}
893
Dave Chinner372cc85e2012-06-22 18:50:12 +1000894/*
895 * Mark bytes first through last inclusive as dirty in the buf
896 * item's bitmap.
897 */
898void
899xfs_buf_item_log(
900 xfs_buf_log_item_t *bip,
901 uint first,
902 uint last)
903{
904 int i;
905 uint start;
906 uint end;
907 struct xfs_buf *bp = bip->bli_buf;
908
909 /*
Dave Chinner372cc85e2012-06-22 18:50:12 +1000910 * walk each buffer segment and mark them dirty appropriately.
911 */
912 start = 0;
913 for (i = 0; i < bip->bli_format_count; i++) {
914 if (start > last)
915 break;
916 end = start + BBTOB(bp->b_maps[i].bm_len);
917 if (first > end) {
918 start += BBTOB(bp->b_maps[i].bm_len);
919 continue;
920 }
921 if (first < start)
922 first = start;
923 if (end > last)
924 end = last;
925
Eric Sandeen72b06362014-04-14 19:01:34 +1000926 xfs_buf_item_log_segment(first, end,
Dave Chinner372cc85e2012-06-22 18:50:12 +1000927 &bip->bli_formats[i].blf_data_map[0]);
928
929 start += bp->b_maps[i].bm_len;
930 }
931}
932
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
934/*
Dave Chinner5f6bed72013-06-27 16:04:52 +1000935 * Return 1 if the buffer has been logged or ordered in a transaction (at any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 * point, not just the current transaction) and 0 if not.
937 */
938uint
939xfs_buf_item_dirty(
940 xfs_buf_log_item_t *bip)
941{
942 return (bip->bli_flags & XFS_BLI_DIRTY);
943}
944
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000945STATIC void
946xfs_buf_item_free(
947 xfs_buf_log_item_t *bip)
948{
Dave Chinner372cc85e2012-06-22 18:50:12 +1000949 xfs_buf_item_free_format(bip);
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000950 kmem_zone_free(xfs_buf_item_zone, bip);
951}
952
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953/*
954 * This is called when the buf log item is no longer needed. It should
955 * free the buf log item associated with the given buffer and clear
956 * the buffer's pointer to the buf log item. If there are no more
957 * items in the list, clear the b_iodone field of the buffer (see
958 * xfs_buf_attach_iodone() below).
959 */
960void
961xfs_buf_item_relse(
962 xfs_buf_t *bp)
963{
Dave Chinner5f6bed72013-06-27 16:04:52 +1000964 xfs_buf_log_item_t *bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000966 trace_xfs_buf_item_relse(bp, _RET_IP_);
Dave Chinner5f6bed72013-06-27 16:04:52 +1000967 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000968
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200969 bp->b_fspriv = bip->bli_item.li_bio_list;
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200970 if (bp->b_fspriv == NULL)
971 bp->b_iodone = NULL;
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200972
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000973 xfs_buf_rele(bp);
974 xfs_buf_item_free(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975}
976
977
978/*
979 * Add the given log item with its callback to the list of callbacks
980 * to be called when the buffer's I/O completes. If it is not set
981 * already, set the buffer's b_iodone() routine to be
982 * xfs_buf_iodone_callbacks() and link the log item into the list of
983 * items rooted at b_fsprivate. Items are always added as the second
984 * entry in the list if there is a first, because the buf item code
985 * assumes that the buf log item is first.
986 */
987void
988xfs_buf_attach_iodone(
989 xfs_buf_t *bp,
990 void (*cb)(xfs_buf_t *, xfs_log_item_t *),
991 xfs_log_item_t *lip)
992{
993 xfs_log_item_t *head_lip;
994
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200995 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996
997 lip->li_cb = cb;
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200998 head_lip = bp->b_fspriv;
999 if (head_lip) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 lip->li_bio_list = head_lip->li_bio_list;
1001 head_lip->li_bio_list = lip;
1002 } else {
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001003 bp->b_fspriv = lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 }
1005
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02001006 ASSERT(bp->b_iodone == NULL ||
1007 bp->b_iodone == xfs_buf_iodone_callbacks);
1008 bp->b_iodone = xfs_buf_iodone_callbacks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009}
1010
Dave Chinnerc90821a2010-12-03 17:00:52 +11001011/*
1012 * We can have many callbacks on a buffer. Running the callbacks individually
1013 * can cause a lot of contention on the AIL lock, so we allow for a single
1014 * callback to be able to scan the remaining lip->li_bio_list for other items
1015 * of the same type and callback to be processed in the first call.
1016 *
1017 * As a result, the loop walking the callback list below will also modify the
1018 * list. it removes the first item from the list and then runs the callback.
1019 * The loop then restarts from the new head of the list. This allows the
1020 * callback to scan and modify the list attached to the buffer and we don't
1021 * have to care about maintaining a next item pointer.
1022 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023STATIC void
1024xfs_buf_do_callbacks(
Dave Chinnerc90821a2010-12-03 17:00:52 +11001025 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026{
Dave Chinnerc90821a2010-12-03 17:00:52 +11001027 struct xfs_log_item *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001029 while ((lip = bp->b_fspriv) != NULL) {
1030 bp->b_fspriv = lip->li_bio_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 ASSERT(lip->li_cb != NULL);
1032 /*
1033 * Clear the next pointer so we don't have any
1034 * confusion if the item is added to another buf.
1035 * Don't touch the log item after calling its
1036 * callback, because it could have freed itself.
1037 */
1038 lip->li_bio_list = NULL;
1039 lip->li_cb(bp, lip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 }
1041}
1042
1043/*
1044 * This is the iodone() function for buffers which have had callbacks
1045 * attached to them by xfs_buf_attach_iodone(). It should remove each
1046 * log item from the buffer's list and call the callback of each in turn.
1047 * When done, the buffer's fsprivate field is set to NULL and the buffer
1048 * is unlocked with a call to iodone().
1049 */
1050void
1051xfs_buf_iodone_callbacks(
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001052 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053{
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001054 struct xfs_log_item *lip = bp->b_fspriv;
1055 struct xfs_mount *mp = lip->li_mountp;
1056 static ulong lasttime;
1057 static xfs_buftarg_t *lasttarg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058
Dave Chinner36de9552014-06-06 16:02:12 +10001059 if (likely(!bp->b_error))
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001060 goto do_callbacks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001062 /*
1063 * If we've already decided to shutdown the filesystem because of
1064 * I/O errors, there's no point in giving this a retry.
1065 */
1066 if (XFS_FORCED_SHUTDOWN(mp)) {
Christoph Hellwigc867cb62011-10-10 16:52:46 +00001067 xfs_buf_stale(bp);
Christoph Hellwigc867cb62011-10-10 16:52:46 +00001068 XFS_BUF_DONE(bp);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001069 trace_xfs_buf_item_iodone(bp, _RET_IP_);
1070 goto do_callbacks;
1071 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072
Chandra Seetharaman49074c02011-07-22 23:40:40 +00001073 if (bp->b_target != lasttarg ||
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001074 time_after(jiffies, (lasttime + 5*HZ))) {
1075 lasttime = jiffies;
Christoph Hellwigb38505b2011-10-10 16:52:50 +00001076 xfs_buf_ioerror_alert(bp, __func__);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001077 }
Chandra Seetharaman49074c02011-07-22 23:40:40 +00001078 lasttarg = bp->b_target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001080 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001081 * If the write was asynchronous then no one will be looking for the
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001082 * error. Clear the error state and write the buffer out again.
1083 *
Christoph Hellwig43ff2122012-04-23 15:58:39 +10001084 * XXX: This helps against transient write errors, but we need to find
1085 * a way to shut the filesystem down if the writes keep failing.
1086 *
1087 * In practice we'll shut the filesystem down soon as non-transient
Dave Chinner595bff72014-10-02 09:05:14 +10001088 * errors tend to affect the whole device and a failing log write
Christoph Hellwig43ff2122012-04-23 15:58:39 +10001089 * will make us give up. But we really ought to do better here.
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001090 */
1091 if (XFS_BUF_ISASYNC(bp)) {
Christoph Hellwig43ff2122012-04-23 15:58:39 +10001092 ASSERT(bp->b_iodone != NULL);
1093
1094 trace_xfs_buf_item_iodone_async(bp, _RET_IP_);
1095
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00001096 xfs_buf_ioerror(bp, 0); /* errno of 0 unsets the flag */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
Dave Chinnerac8809f2013-12-12 16:34:38 +11001098 if (!(bp->b_flags & (XBF_STALE|XBF_WRITE_FAIL))) {
1099 bp->b_flags |= XBF_WRITE | XBF_ASYNC |
1100 XBF_DONE | XBF_WRITE_FAIL;
Dave Chinner595bff72014-10-02 09:05:14 +10001101 xfs_buf_submit(bp);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10001102 } else {
1103 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 }
Christoph Hellwig43ff2122012-04-23 15:58:39 +10001105
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 return;
1107 }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001108
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001109 /*
1110 * If the write of the buffer was synchronous, we want to make
1111 * sure to return the error to the caller of xfs_bwrite().
1112 */
Christoph Hellwigc867cb62011-10-10 16:52:46 +00001113 xfs_buf_stale(bp);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001114 XFS_BUF_DONE(bp);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001115
1116 trace_xfs_buf_error_relse(bp, _RET_IP_);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001117
1118do_callbacks:
Dave Chinnerc90821a2010-12-03 17:00:52 +11001119 xfs_buf_do_callbacks(bp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001120 bp->b_fspriv = NULL;
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02001121 bp->b_iodone = NULL;
Dave Chinnere8aaba92014-10-02 09:04:22 +10001122 xfs_buf_ioend(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123}
1124
1125/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 * This is the iodone() function for buffers which have been
1127 * logged. It is called when they are eventually flushed out.
1128 * It should remove the buf item from the AIL, and free the buf item.
1129 * It is called by xfs_buf_iodone_callbacks() above which will take
1130 * care of cleaning up the buffer itself.
1131 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132void
1133xfs_buf_iodone(
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001134 struct xfs_buf *bp,
1135 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136{
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001137 struct xfs_ail *ailp = lip->li_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001139 ASSERT(BUF_ITEM(lip)->bli_buf == bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +10001141 xfs_buf_rele(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142
1143 /*
1144 * If we are forcibly shutting down, this may well be
1145 * off the AIL already. That's because we simulate the
1146 * log-committed callbacks to unpin these buffers. Or we may never
1147 * have put this item on AIL because of the transaction was
David Chinner783a2f62008-10-30 17:39:58 +11001148 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 *
1150 * Either way, AIL is useless if we're forcing a shutdown.
1151 */
David Chinnerfc1829f2008-10-30 17:39:46 +11001152 spin_lock(&ailp->xa_lock);
Dave Chinner04913fd2012-04-23 15:58:41 +10001153 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001154 xfs_buf_item_free(BUF_ITEM(lip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155}