blob: 4654338b03fc5a160247fbc9d0a52ffb26b79cf9 [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"
Dave Chinner239880e2013-10-23 10:50:10 +110020#include "xfs_log_format.h"
21#include "xfs_trans_resv.h"
Nathan Scotta844f452005-11-02 14:38:42 +110022#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_sb.h"
David Chinnerda353b02007-08-28 14:00:13 +100024#include "xfs_ag.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));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
323 /*
Dave Chinnerccf7c232010-05-20 23:19:42 +1000324 * If it is an inode buffer, transfer the in-memory state to the
Dave Chinnerddf6ad02013-06-27 16:04:56 +1000325 * format flags and clear the in-memory state.
326 *
327 * For buffer based inode allocation, we do not transfer
Dave Chinnerccf7c232010-05-20 23:19:42 +1000328 * this state if the inode buffer allocation has not yet been committed
329 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
330 * correct replay of the inode allocation.
Dave Chinnerddf6ad02013-06-27 16:04:56 +1000331 *
332 * For icreate item based inode allocation, the buffers aren't written
333 * to the journal during allocation, and hence we should always tag the
334 * buffer as an inode buffer so that the correct unlinked list replay
335 * occurs during recovery.
Dave Chinnerccf7c232010-05-20 23:19:42 +1000336 */
337 if (bip->bli_flags & XFS_BLI_INODE_BUF) {
Dave Chinnerddf6ad02013-06-27 16:04:56 +1000338 if (xfs_sb_version_hascrc(&lip->li_mountp->m_sb) ||
339 !((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000340 xfs_log_item_in_current_chkpt(lip)))
Mark Tinguelyb9438172012-12-04 17:18:03 -0600341 bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF;
Dave Chinnerccf7c232010-05-20 23:19:42 +1000342 bip->bli_flags &= ~XFS_BLI_INODE_BUF;
343 }
344
Dave Chinner5f6bed72013-06-27 16:04:52 +1000345 if ((bip->bli_flags & (XFS_BLI_ORDERED|XFS_BLI_STALE)) ==
346 XFS_BLI_ORDERED) {
347 /*
348 * The buffer has been logged just to order it. It is not being
349 * included in the transaction commit, so don't format it.
350 */
351 trace_xfs_buf_item_format_ordered(bip);
352 return;
353 }
354
Dave Chinner372cc85e2012-06-22 18:50:12 +1000355 for (i = 0; i < bip->bli_format_count; i++) {
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100356 xfs_buf_item_format_segment(bip, lv, &vecp, offset,
Christoph Hellwig12343512013-12-13 11:00:43 +1100357 &bip->bli_formats[i]);
Dave Chinner372cc85e2012-06-22 18:50:12 +1000358 offset += bp->b_maps[i].bm_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 }
360
361 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 * Check to make sure everything is consistent.
363 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000364 trace_xfs_buf_item_format(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365}
366
367/*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000368 * This is called to pin the buffer associated with the buf log item in memory
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000369 * so it cannot be written out.
Dave Chinner64fc35d2010-05-07 11:04:34 +1000370 *
371 * We also always take a reference to the buffer log item here so that the bli
372 * is held while the item is pinned in memory. This means that we can
373 * unconditionally drop the reference count a transaction holds when the
374 * transaction is completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000376STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377xfs_buf_item_pin(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000378 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000380 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 ASSERT(atomic_read(&bip->bli_refcount) > 0);
383 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
Dave Chinner5f6bed72013-06-27 16:04:52 +1000384 (bip->bli_flags & XFS_BLI_ORDERED) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 (bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000386
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000387 trace_xfs_buf_item_pin(bip);
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000388
389 atomic_inc(&bip->bli_refcount);
390 atomic_inc(&bip->bli_buf->b_pin_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391}
392
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393/*
394 * This is called to unpin the buffer associated with the buf log
395 * item which was previously pinned with a call to xfs_buf_item_pin().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 *
397 * Also drop the reference to the buf item for the current transaction.
398 * If the XFS_BLI_STALE flag is set and we are the last reference,
399 * then free up the buf log item and unlock the buffer.
Christoph Hellwig9412e312010-06-23 18:11:15 +1000400 *
401 * If the remove flag is set we are called from uncommit in the
402 * forced-shutdown path. If that is true and the reference count on
403 * the log item is going to drop to zero we need to free the item's
404 * descriptor in the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000406STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407xfs_buf_item_unpin(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000408 struct xfs_log_item *lip,
Christoph Hellwig9412e312010-06-23 18:11:15 +1000409 int remove)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000411 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000412 xfs_buf_t *bp = bip->bli_buf;
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000413 struct xfs_ail *ailp = lip->li_ailp;
Dave Chinner8e123852010-03-08 11:26:03 +1100414 int stale = bip->bli_flags & XFS_BLI_STALE;
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000415 int freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200417 ASSERT(bp->b_fspriv == bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 ASSERT(atomic_read(&bip->bli_refcount) > 0);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000419
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000420 trace_xfs_buf_item_unpin(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
422 freed = atomic_dec_and_test(&bip->bli_refcount);
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000423
424 if (atomic_dec_and_test(&bp->b_pin_count))
425 wake_up_all(&bp->b_waiters);
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000426
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 if (freed && stale) {
428 ASSERT(bip->bli_flags & XFS_BLI_STALE);
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200429 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 ASSERT(XFS_BUF_ISSTALE(bp));
Mark Tinguelyb9438172012-12-04 17:18:03 -0600431 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000432
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000433 trace_xfs_buf_item_unpin_stale(bip);
434
Christoph Hellwig9412e312010-06-23 18:11:15 +1000435 if (remove) {
436 /*
Dave Chinnere34a3142011-01-27 12:13:35 +1100437 * If we are in a transaction context, we have to
438 * remove the log item from the transaction as we are
439 * about to release our reference to the buffer. If we
440 * don't, the unlock that occurs later in
441 * xfs_trans_uncommit() will try to reference the
Christoph Hellwig9412e312010-06-23 18:11:15 +1000442 * buffer which we no longer have a hold on.
443 */
Dave Chinnere34a3142011-01-27 12:13:35 +1100444 if (lip->li_desc)
445 xfs_trans_del_item(lip);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000446
447 /*
448 * Since the transaction no longer refers to the buffer,
449 * the buffer should no longer refer to the transaction.
450 */
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200451 bp->b_transp = NULL;
Christoph Hellwig9412e312010-06-23 18:11:15 +1000452 }
453
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 /*
455 * If we get called here because of an IO error, we may
David Chinner783a2f62008-10-30 17:39:58 +1100456 * or may not have the item on the AIL. xfs_trans_ail_delete()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 * will take care of that situation.
David Chinner783a2f62008-10-30 17:39:58 +1100458 * xfs_trans_ail_delete() drops the AIL lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 */
460 if (bip->bli_flags & XFS_BLI_STALE_INODE) {
Dave Chinnerc90821a2010-12-03 17:00:52 +1100461 xfs_buf_do_callbacks(bp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200462 bp->b_fspriv = NULL;
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200463 bp->b_iodone = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 } else {
David Chinner783a2f62008-10-30 17:39:58 +1100465 spin_lock(&ailp->xa_lock);
Dave Chinner04913fd2012-04-23 15:58:41 +1000466 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_LOG_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 xfs_buf_item_relse(bp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200468 ASSERT(bp->b_fspriv == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 }
470 xfs_buf_relse(bp);
Christoph Hellwig960c60a2012-04-23 15:58:38 +1000471 } else if (freed && remove) {
Dave Chinner137fff02012-11-02 14:23:12 +1100472 /*
473 * There are currently two references to the buffer - the active
474 * LRU reference and the buf log item. What we are about to do
475 * here - simulate a failed IO completion - requires 3
476 * references.
477 *
478 * The LRU reference is removed by the xfs_buf_stale() call. The
479 * buf item reference is removed by the xfs_buf_iodone()
480 * callback that is run by xfs_buf_do_callbacks() during ioend
481 * processing (via the bp->b_iodone callback), and then finally
482 * the ioend processing will drop the IO reference if the buffer
483 * is marked XBF_ASYNC.
484 *
485 * Hence we need to take an additional reference here so that IO
486 * completion processing doesn't free the buffer prematurely.
487 */
Christoph Hellwig960c60a2012-04-23 15:58:38 +1000488 xfs_buf_lock(bp);
Dave Chinner137fff02012-11-02 14:23:12 +1100489 xfs_buf_hold(bp);
490 bp->b_flags |= XBF_ASYNC;
Christoph Hellwig960c60a2012-04-23 15:58:38 +1000491 xfs_buf_ioerror(bp, EIO);
492 XFS_BUF_UNDONE(bp);
493 xfs_buf_stale(bp);
494 xfs_buf_ioend(bp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 }
496}
497
Dave Chinnerac8809f2013-12-12 16:34:38 +1100498/*
499 * Buffer IO error rate limiting. Limit it to no more than 10 messages per 30
500 * seconds so as to not spam logs too much on repeated detection of the same
501 * buffer being bad..
502 */
503
504DEFINE_RATELIMIT_STATE(xfs_buf_write_fail_rl_state, 30 * HZ, 10);
505
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000506STATIC uint
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000507xfs_buf_item_push(
508 struct xfs_log_item *lip,
509 struct list_head *buffer_list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000511 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
512 struct xfs_buf *bp = bip->bli_buf;
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000513 uint rval = XFS_ITEM_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
Chandra Seetharaman811e64c2011-07-22 23:40:27 +0000515 if (xfs_buf_ispinned(bp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 return XFS_ITEM_PINNED;
Brian Foster5337fe92013-02-11 10:08:21 -0500517 if (!xfs_buf_trylock(bp)) {
518 /*
519 * If we have just raced with a buffer being pinned and it has
520 * been marked stale, we could end up stalling until someone else
521 * issues a log force to unpin the stale buffer. Check for the
522 * race condition here so xfsaild recognizes the buffer is pinned
523 * and queues a log force to move it along.
524 */
525 if (xfs_buf_ispinned(bp))
526 return XFS_ITEM_PINNED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 return XFS_ITEM_LOCKED;
Brian Foster5337fe92013-02-11 10:08:21 -0500528 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000531
532 trace_xfs_buf_item_push(bip);
533
Dave Chinnerac8809f2013-12-12 16:34:38 +1100534 /* has a previous flush failed due to IO errors? */
535 if ((bp->b_flags & XBF_WRITE_FAIL) &&
536 ___ratelimit(&xfs_buf_write_fail_rl_state, "XFS:")) {
537 xfs_warn(bp->b_target->bt_mount,
538"Detected failing async write on buffer block 0x%llx. Retrying async write.\n",
539 (long long)bp->b_bn);
540 }
541
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000542 if (!xfs_buf_delwri_queue(bp, buffer_list))
543 rval = XFS_ITEM_FLUSHING;
544 xfs_buf_unlock(bp);
545 return rval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546}
547
548/*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000549 * Release the buffer associated with the buf log item. If there is no dirty
550 * logged data associated with the buffer recorded in the buf log item, then
551 * free the buf log item and remove the reference to it in the buffer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 *
Dave Chinner64fc35d2010-05-07 11:04:34 +1000553 * This call ignores the recursion count. It is only called when the buffer
554 * should REALLY be unlocked, regardless of the recursion count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 *
Dave Chinner64fc35d2010-05-07 11:04:34 +1000556 * We unconditionally drop the transaction's reference to the log item. If the
557 * item was logged, then another reference was taken when it was pinned, so we
558 * can safely drop the transaction reference now. This also allows us to avoid
559 * potential races with the unpin code freeing the bli by not referencing the
560 * bli after we've dropped the reference count.
561 *
562 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
563 * if necessary but do not unlock the buffer. This is for support of
564 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
565 * free the item.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000567STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568xfs_buf_item_unlock(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000569 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000571 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
572 struct xfs_buf *bp = bip->bli_buf;
Dave Chinner5f6bed72013-06-27 16:04:52 +1000573 bool clean;
574 bool aborted;
575 int flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
Dave Chinner64fc35d2010-05-07 11:04:34 +1000577 /* Clear the buffer's association with this transaction. */
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200578 bp->b_transp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
580 /*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000581 * If this is a transaction abort, don't return early. Instead, allow
582 * the brelse to happen. Normally it would be done for stale
583 * (cancelled) buffers at unpin time, but we'll never go through the
584 * pin/unpin cycle if we abort inside commit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 */
Dave Chinner5f6bed72013-06-27 16:04:52 +1000586 aborted = (lip->li_flags & XFS_LI_ABORTED) ? true : false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 /*
Dave Chinner5f6bed72013-06-27 16:04:52 +1000588 * Before possibly freeing the buf item, copy the per-transaction state
589 * so we can reference it safely later after clearing it from the
590 * buffer log item.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 */
Dave Chinner5f6bed72013-06-27 16:04:52 +1000592 flags = bip->bli_flags;
593 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD | XFS_BLI_ORDERED);
Dave Chinner64fc35d2010-05-07 11:04:34 +1000594
595 /*
596 * If the buf item is marked stale, then don't do anything. We'll
597 * unlock the buffer and free the buf item when the buffer is unpinned
598 * for the last time.
599 */
Dave Chinner5f6bed72013-06-27 16:04:52 +1000600 if (flags & XFS_BLI_STALE) {
Dave Chinner64fc35d2010-05-07 11:04:34 +1000601 trace_xfs_buf_item_unlock_stale(bip);
Mark Tinguelyb9438172012-12-04 17:18:03 -0600602 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
Dave Chinner64fc35d2010-05-07 11:04:34 +1000603 if (!aborted) {
604 atomic_dec(&bip->bli_refcount);
605 return;
606 }
607 }
608
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000609 trace_xfs_buf_item_unlock(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610
611 /*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000612 * If the buf item isn't tracking any data, free it, otherwise drop the
Dave Chinner3b190342013-01-21 23:53:55 +1100613 * reference we hold to it. If we are aborting the transaction, this may
614 * be the only reference to the buf item, so we free it anyway
615 * regardless of whether it is dirty or not. A dirty abort implies a
616 * shutdown, anyway.
Dave Chinner5f6bed72013-06-27 16:04:52 +1000617 *
618 * Ordered buffers are dirty but may have no recorded changes, so ensure
619 * we only release clean items here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 */
Dave Chinner5f6bed72013-06-27 16:04:52 +1000621 clean = (flags & XFS_BLI_DIRTY) ? false : true;
622 if (clean) {
623 int i;
624 for (i = 0; i < bip->bli_format_count; i++) {
625 if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map,
626 bip->bli_formats[i].blf_map_size)) {
627 clean = false;
628 break;
629 }
Mark Tinguelyc883d0c2012-12-04 17:18:05 -0600630 }
631 }
Dave Chinner46f9d2e2013-09-03 21:47:37 +1000632
633 /*
634 * Clean buffers, by definition, cannot be in the AIL. However, aborted
635 * buffers may be dirty and hence in the AIL. Therefore if we are
636 * aborting a buffer and we've just taken the last refernce away, we
637 * have to check if it is in the AIL before freeing it. We need to free
638 * it in this case, because an aborted transaction has already shut the
639 * filesystem down and this is the last chance we will have to do so.
640 */
641 if (atomic_dec_and_test(&bip->bli_refcount)) {
642 if (clean)
643 xfs_buf_item_relse(bp);
644 else if (aborted) {
645 ASSERT(XFS_FORCED_SHUTDOWN(lip->li_mountp));
646 if (lip->li_flags & XFS_LI_IN_AIL) {
Dave Chinner48852352013-09-24 16:01:13 +1000647 spin_lock(&lip->li_ailp->xa_lock);
Dave Chinner46f9d2e2013-09-03 21:47:37 +1000648 xfs_trans_ail_delete(lip->li_ailp, lip,
649 SHUTDOWN_LOG_IO_ERROR);
650 }
Dave Chinner3b190342013-01-21 23:53:55 +1100651 xfs_buf_item_relse(bp);
652 }
Dave Chinner46f9d2e2013-09-03 21:47:37 +1000653 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654
Dave Chinner5f6bed72013-06-27 16:04:52 +1000655 if (!(flags & XFS_BLI_HOLD))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657}
658
659/*
660 * This is called to find out where the oldest active copy of the
661 * buf log item in the on disk log resides now that the last log
662 * write of it completed at the given lsn.
663 * We always re-log all the dirty data in a buffer, so usually the
664 * latest copy in the on disk log is the only one that matters. For
665 * those cases we simply return the given lsn.
666 *
667 * The one exception to this is for buffers full of newly allocated
668 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
669 * flag set, indicating that only the di_next_unlinked fields from the
670 * inodes in the buffers will be replayed during recovery. If the
671 * original newly allocated inode images have not yet been flushed
672 * when the buffer is so relogged, then we need to make sure that we
673 * keep the old images in the 'active' portion of the log. We do this
674 * by returning the original lsn of that transaction here rather than
675 * the current one.
676 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000677STATIC xfs_lsn_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678xfs_buf_item_committed(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000679 struct xfs_log_item *lip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 xfs_lsn_t lsn)
681{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000682 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
683
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000684 trace_xfs_buf_item_committed(bip);
685
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000686 if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0)
687 return lip->li_lsn;
688 return lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689}
690
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000691STATIC void
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000692xfs_buf_item_committing(
693 struct xfs_log_item *lip,
694 xfs_lsn_t commit_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695{
696}
697
698/*
699 * This is the ops vector shared by all buf log items.
700 */
Christoph Hellwig272e42b2011-10-28 09:54:24 +0000701static const struct xfs_item_ops xfs_buf_item_ops = {
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000702 .iop_size = xfs_buf_item_size,
703 .iop_format = xfs_buf_item_format,
704 .iop_pin = xfs_buf_item_pin,
705 .iop_unpin = xfs_buf_item_unpin,
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000706 .iop_unlock = xfs_buf_item_unlock,
707 .iop_committed = xfs_buf_item_committed,
708 .iop_push = xfs_buf_item_push,
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000709 .iop_committing = xfs_buf_item_committing
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710};
711
Dave Chinner372cc85e2012-06-22 18:50:12 +1000712STATIC int
713xfs_buf_item_get_format(
714 struct xfs_buf_log_item *bip,
715 int count)
716{
717 ASSERT(bip->bli_formats == NULL);
718 bip->bli_format_count = count;
719
720 if (count == 1) {
Mark Tinguelyb9438172012-12-04 17:18:03 -0600721 bip->bli_formats = &bip->__bli_format;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000722 return 0;
723 }
724
725 bip->bli_formats = kmem_zalloc(count * sizeof(struct xfs_buf_log_format),
726 KM_SLEEP);
727 if (!bip->bli_formats)
728 return ENOMEM;
729 return 0;
730}
731
732STATIC void
733xfs_buf_item_free_format(
734 struct xfs_buf_log_item *bip)
735{
Mark Tinguelyb9438172012-12-04 17:18:03 -0600736 if (bip->bli_formats != &bip->__bli_format) {
Dave Chinner372cc85e2012-06-22 18:50:12 +1000737 kmem_free(bip->bli_formats);
738 bip->bli_formats = NULL;
739 }
740}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741
742/*
743 * Allocate a new buf log item to go with the given buffer.
744 * Set the buffer's b_fsprivate field to point to the new
745 * buf log item. If there are other item's attached to the
746 * buffer (see xfs_buf_attach_iodone() below), then put the
747 * buf log item at the front.
748 */
749void
750xfs_buf_item_init(
751 xfs_buf_t *bp,
752 xfs_mount_t *mp)
753{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200754 xfs_log_item_t *lip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 xfs_buf_log_item_t *bip;
756 int chunks;
757 int map_size;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000758 int error;
759 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
761 /*
762 * Check to see if there is already a buf log item for
763 * this buffer. If there is, it is guaranteed to be
764 * the first. If we do already have one, there is
765 * nothing to do here so return.
766 */
Dave Chinnerebad8612010-09-22 10:47:20 +1000767 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200768 if (lip != NULL && lip->li_type == XFS_LI_BUF)
769 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
Dave Chinner372cc85e2012-06-22 18:50:12 +1000771 bip = kmem_zone_zalloc(xfs_buf_item_zone, KM_SLEEP);
Dave Chinner43f5efc2010-03-23 10:10:00 +1100772 xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 bip->bli_buf = bp;
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000774 xfs_buf_hold(bp);
Dave Chinner372cc85e2012-06-22 18:50:12 +1000775
776 /*
777 * chunks is the number of XFS_BLF_CHUNK size pieces the buffer
778 * can be divided into. Make sure not to truncate any pieces.
779 * map_size is the size of the bitmap needed to describe the
780 * chunks of the buffer.
781 *
782 * Discontiguous buffer support follows the layout of the underlying
783 * buffer. This makes the implementation as simple as possible.
784 */
785 error = xfs_buf_item_get_format(bip, bp->b_map_count);
786 ASSERT(error == 0);
787
788 for (i = 0; i < bip->bli_format_count; i++) {
789 chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len),
790 XFS_BLF_CHUNK);
791 map_size = DIV_ROUND_UP(chunks, NBWORD);
792
793 bip->bli_formats[i].blf_type = XFS_LI_BUF;
794 bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn;
795 bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len;
796 bip->bli_formats[i].blf_map_size = map_size;
797 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 /*
800 * Put the buf item into the list of items attached to the
801 * buffer at the front.
802 */
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200803 if (bp->b_fspriv)
804 bip->bli_item.li_bio_list = bp->b_fspriv;
805 bp->b_fspriv = bip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806}
807
808
809/*
810 * Mark bytes first through last inclusive as dirty in the buf
811 * item's bitmap.
812 */
Dave Chinner632b89e2013-10-29 22:11:58 +1100813static void
Dave Chinner372cc85e2012-06-22 18:50:12 +1000814xfs_buf_item_log_segment(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 uint first,
Dave Chinner372cc85e2012-06-22 18:50:12 +1000816 uint last,
817 uint *map)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818{
819 uint first_bit;
820 uint last_bit;
821 uint bits_to_set;
822 uint bits_set;
823 uint word_num;
824 uint *wordp;
825 uint bit;
826 uint end_bit;
827 uint mask;
828
829 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 * Convert byte offsets to bit numbers.
831 */
Dave Chinnerc1155412010-05-07 11:05:19 +1000832 first_bit = first >> XFS_BLF_SHIFT;
833 last_bit = last >> XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834
835 /*
836 * Calculate the total number of bits to be set.
837 */
838 bits_to_set = last_bit - first_bit + 1;
839
840 /*
841 * Get a pointer to the first word in the bitmap
842 * to set a bit in.
843 */
844 word_num = first_bit >> BIT_TO_WORD_SHIFT;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000845 wordp = &map[word_num];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
847 /*
848 * Calculate the starting bit in the first word.
849 */
850 bit = first_bit & (uint)(NBWORD - 1);
851
852 /*
853 * First set any bits in the first word of our range.
854 * If it starts at bit 0 of the word, it will be
855 * set below rather than here. That is what the variable
856 * bit tells us. The variable bits_set tracks the number
857 * of bits that have been set so far. End_bit is the number
858 * of the last bit to be set in this word plus one.
859 */
860 if (bit) {
861 end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
862 mask = ((1 << (end_bit - bit)) - 1) << bit;
863 *wordp |= mask;
864 wordp++;
865 bits_set = end_bit - bit;
866 } else {
867 bits_set = 0;
868 }
869
870 /*
871 * Now set bits a whole word at a time that are between
872 * first_bit and last_bit.
873 */
874 while ((bits_to_set - bits_set) >= NBWORD) {
875 *wordp |= 0xffffffff;
876 bits_set += NBWORD;
877 wordp++;
878 }
879
880 /*
881 * Finally, set any bits left to be set in one last partial word.
882 */
883 end_bit = bits_to_set - bits_set;
884 if (end_bit) {
885 mask = (1 << end_bit) - 1;
886 *wordp |= mask;
887 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888}
889
Dave Chinner372cc85e2012-06-22 18:50:12 +1000890/*
891 * Mark bytes first through last inclusive as dirty in the buf
892 * item's bitmap.
893 */
894void
895xfs_buf_item_log(
896 xfs_buf_log_item_t *bip,
897 uint first,
898 uint last)
899{
900 int i;
901 uint start;
902 uint end;
903 struct xfs_buf *bp = bip->bli_buf;
904
905 /*
Dave Chinner372cc85e2012-06-22 18:50:12 +1000906 * walk each buffer segment and mark them dirty appropriately.
907 */
908 start = 0;
909 for (i = 0; i < bip->bli_format_count; i++) {
910 if (start > last)
911 break;
912 end = start + BBTOB(bp->b_maps[i].bm_len);
913 if (first > end) {
914 start += BBTOB(bp->b_maps[i].bm_len);
915 continue;
916 }
917 if (first < start)
918 first = start;
919 if (end > last)
920 end = last;
921
Eric Sandeen72b06362014-04-14 19:01:34 +1000922 xfs_buf_item_log_segment(first, end,
Dave Chinner372cc85e2012-06-22 18:50:12 +1000923 &bip->bli_formats[i].blf_data_map[0]);
924
925 start += bp->b_maps[i].bm_len;
926 }
927}
928
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929
930/*
Dave Chinner5f6bed72013-06-27 16:04:52 +1000931 * Return 1 if the buffer has been logged or ordered in a transaction (at any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 * point, not just the current transaction) and 0 if not.
933 */
934uint
935xfs_buf_item_dirty(
936 xfs_buf_log_item_t *bip)
937{
938 return (bip->bli_flags & XFS_BLI_DIRTY);
939}
940
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000941STATIC void
942xfs_buf_item_free(
943 xfs_buf_log_item_t *bip)
944{
Dave Chinner372cc85e2012-06-22 18:50:12 +1000945 xfs_buf_item_free_format(bip);
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000946 kmem_zone_free(xfs_buf_item_zone, bip);
947}
948
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949/*
950 * This is called when the buf log item is no longer needed. It should
951 * free the buf log item associated with the given buffer and clear
952 * the buffer's pointer to the buf log item. If there are no more
953 * items in the list, clear the b_iodone field of the buffer (see
954 * xfs_buf_attach_iodone() below).
955 */
956void
957xfs_buf_item_relse(
958 xfs_buf_t *bp)
959{
Dave Chinner5f6bed72013-06-27 16:04:52 +1000960 xfs_buf_log_item_t *bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000962 trace_xfs_buf_item_relse(bp, _RET_IP_);
Dave Chinner5f6bed72013-06-27 16:04:52 +1000963 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000964
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200965 bp->b_fspriv = bip->bli_item.li_bio_list;
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200966 if (bp->b_fspriv == NULL)
967 bp->b_iodone = NULL;
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200968
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000969 xfs_buf_rele(bp);
970 xfs_buf_item_free(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971}
972
973
974/*
975 * Add the given log item with its callback to the list of callbacks
976 * to be called when the buffer's I/O completes. If it is not set
977 * already, set the buffer's b_iodone() routine to be
978 * xfs_buf_iodone_callbacks() and link the log item into the list of
979 * items rooted at b_fsprivate. Items are always added as the second
980 * entry in the list if there is a first, because the buf item code
981 * assumes that the buf log item is first.
982 */
983void
984xfs_buf_attach_iodone(
985 xfs_buf_t *bp,
986 void (*cb)(xfs_buf_t *, xfs_log_item_t *),
987 xfs_log_item_t *lip)
988{
989 xfs_log_item_t *head_lip;
990
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200991 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
993 lip->li_cb = cb;
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200994 head_lip = bp->b_fspriv;
995 if (head_lip) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 lip->li_bio_list = head_lip->li_bio_list;
997 head_lip->li_bio_list = lip;
998 } else {
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200999 bp->b_fspriv = lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 }
1001
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02001002 ASSERT(bp->b_iodone == NULL ||
1003 bp->b_iodone == xfs_buf_iodone_callbacks);
1004 bp->b_iodone = xfs_buf_iodone_callbacks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005}
1006
Dave Chinnerc90821a2010-12-03 17:00:52 +11001007/*
1008 * We can have many callbacks on a buffer. Running the callbacks individually
1009 * can cause a lot of contention on the AIL lock, so we allow for a single
1010 * callback to be able to scan the remaining lip->li_bio_list for other items
1011 * of the same type and callback to be processed in the first call.
1012 *
1013 * As a result, the loop walking the callback list below will also modify the
1014 * list. it removes the first item from the list and then runs the callback.
1015 * The loop then restarts from the new head of the list. This allows the
1016 * callback to scan and modify the list attached to the buffer and we don't
1017 * have to care about maintaining a next item pointer.
1018 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019STATIC void
1020xfs_buf_do_callbacks(
Dave Chinnerc90821a2010-12-03 17:00:52 +11001021 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022{
Dave Chinnerc90821a2010-12-03 17:00:52 +11001023 struct xfs_log_item *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001025 while ((lip = bp->b_fspriv) != NULL) {
1026 bp->b_fspriv = lip->li_bio_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 ASSERT(lip->li_cb != NULL);
1028 /*
1029 * Clear the next pointer so we don't have any
1030 * confusion if the item is added to another buf.
1031 * Don't touch the log item after calling its
1032 * callback, because it could have freed itself.
1033 */
1034 lip->li_bio_list = NULL;
1035 lip->li_cb(bp, lip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 }
1037}
1038
1039/*
1040 * This is the iodone() function for buffers which have had callbacks
1041 * attached to them by xfs_buf_attach_iodone(). It should remove each
1042 * log item from the buffer's list and call the callback of each in turn.
1043 * When done, the buffer's fsprivate field is set to NULL and the buffer
1044 * is unlocked with a call to iodone().
1045 */
1046void
1047xfs_buf_iodone_callbacks(
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001048 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049{
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001050 struct xfs_log_item *lip = bp->b_fspriv;
1051 struct xfs_mount *mp = lip->li_mountp;
1052 static ulong lasttime;
1053 static xfs_buftarg_t *lasttarg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
Dave Chinner36de9552014-06-06 16:02:12 +10001055 if (likely(!bp->b_error))
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001056 goto do_callbacks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001058 /*
1059 * If we've already decided to shutdown the filesystem because of
1060 * I/O errors, there's no point in giving this a retry.
1061 */
1062 if (XFS_FORCED_SHUTDOWN(mp)) {
Christoph Hellwigc867cb62011-10-10 16:52:46 +00001063 xfs_buf_stale(bp);
Christoph Hellwigc867cb62011-10-10 16:52:46 +00001064 XFS_BUF_DONE(bp);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001065 trace_xfs_buf_item_iodone(bp, _RET_IP_);
1066 goto do_callbacks;
1067 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068
Chandra Seetharaman49074c02011-07-22 23:40:40 +00001069 if (bp->b_target != lasttarg ||
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001070 time_after(jiffies, (lasttime + 5*HZ))) {
1071 lasttime = jiffies;
Christoph Hellwigb38505b2011-10-10 16:52:50 +00001072 xfs_buf_ioerror_alert(bp, __func__);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001073 }
Chandra Seetharaman49074c02011-07-22 23:40:40 +00001074 lasttarg = bp->b_target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001076 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001077 * If the write was asynchronous then no one will be looking for the
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001078 * error. Clear the error state and write the buffer out again.
1079 *
Christoph Hellwig43ff2122012-04-23 15:58:39 +10001080 * XXX: This helps against transient write errors, but we need to find
1081 * a way to shut the filesystem down if the writes keep failing.
1082 *
1083 * In practice we'll shut the filesystem down soon as non-transient
1084 * erorrs tend to affect the whole device and a failing log write
1085 * will make us give up. But we really ought to do better here.
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001086 */
1087 if (XFS_BUF_ISASYNC(bp)) {
Christoph Hellwig43ff2122012-04-23 15:58:39 +10001088 ASSERT(bp->b_iodone != NULL);
1089
1090 trace_xfs_buf_item_iodone_async(bp, _RET_IP_);
1091
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00001092 xfs_buf_ioerror(bp, 0); /* errno of 0 unsets the flag */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093
Dave Chinnerac8809f2013-12-12 16:34:38 +11001094 if (!(bp->b_flags & (XBF_STALE|XBF_WRITE_FAIL))) {
1095 bp->b_flags |= XBF_WRITE | XBF_ASYNC |
1096 XBF_DONE | XBF_WRITE_FAIL;
Christoph Hellwiga2dcf5d2012-07-13 02:24:10 -04001097 xfs_buf_iorequest(bp);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10001098 } else {
1099 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 }
Christoph Hellwig43ff2122012-04-23 15:58:39 +10001101
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 return;
1103 }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001104
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001105 /*
1106 * If the write of the buffer was synchronous, we want to make
1107 * sure to return the error to the caller of xfs_bwrite().
1108 */
Christoph Hellwigc867cb62011-10-10 16:52:46 +00001109 xfs_buf_stale(bp);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001110 XFS_BUF_DONE(bp);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001111
1112 trace_xfs_buf_error_relse(bp, _RET_IP_);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001113
1114do_callbacks:
Dave Chinnerc90821a2010-12-03 17:00:52 +11001115 xfs_buf_do_callbacks(bp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001116 bp->b_fspriv = NULL;
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02001117 bp->b_iodone = NULL;
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00001118 xfs_buf_ioend(bp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119}
1120
1121/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 * This is the iodone() function for buffers which have been
1123 * logged. It is called when they are eventually flushed out.
1124 * It should remove the buf item from the AIL, and free the buf item.
1125 * It is called by xfs_buf_iodone_callbacks() above which will take
1126 * care of cleaning up the buffer itself.
1127 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128void
1129xfs_buf_iodone(
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001130 struct xfs_buf *bp,
1131 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132{
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001133 struct xfs_ail *ailp = lip->li_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001135 ASSERT(BUF_ITEM(lip)->bli_buf == bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +10001137 xfs_buf_rele(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
1139 /*
1140 * If we are forcibly shutting down, this may well be
1141 * off the AIL already. That's because we simulate the
1142 * log-committed callbacks to unpin these buffers. Or we may never
1143 * have put this item on AIL because of the transaction was
David Chinner783a2f62008-10-30 17:39:58 +11001144 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 *
1146 * Either way, AIL is useless if we're forcing a shutdown.
1147 */
David Chinnerfc1829f2008-10-30 17:39:46 +11001148 spin_lock(&ailp->xa_lock);
Dave Chinner04913fd2012-04-23 15:58:41 +10001149 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001150 xfs_buf_item_free(BUF_ITEM(lip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151}