blob: 1b09d7a280dfa6f1e1abcf247f27989da5fd6fc4 [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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_sb.h"
David Chinnerda353b02007-08-28 14:00:13 +100026#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_mount.h"
Nathan Scotta844f452005-11-02 14:38:42 +110028#include "xfs_buf_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_trans_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_error.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000031#include "xfs_trace.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
41
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#ifdef XFS_TRANS_DEBUG
43/*
44 * This function uses an alternate strategy for tracking the bytes
45 * that the user requests to be logged. This can then be used
46 * in conjunction with the bli_orig array in the buf log item to
47 * catch bugs in our callers' code.
48 *
49 * We also double check the bits set in xfs_buf_item_log using a
50 * simple algorithm to check that every byte is accounted for.
51 */
52STATIC void
53xfs_buf_item_log_debug(
54 xfs_buf_log_item_t *bip,
55 uint first,
56 uint last)
57{
58 uint x;
59 uint byte;
60 uint nbytes;
61 uint chunk_num;
62 uint word_num;
63 uint bit_num;
64 uint bit_set;
65 uint *wordp;
66
67 ASSERT(bip->bli_logged != NULL);
68 byte = first;
69 nbytes = last - first + 1;
70 bfset(bip->bli_logged, first, nbytes);
71 for (x = 0; x < nbytes; x++) {
Dave Chinnerc1155412010-05-07 11:05:19 +100072 chunk_num = byte >> XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 word_num = chunk_num >> BIT_TO_WORD_SHIFT;
74 bit_num = chunk_num & (NBWORD - 1);
75 wordp = &(bip->bli_format.blf_data_map[word_num]);
76 bit_set = *wordp & (1 << bit_num);
77 ASSERT(bit_set);
78 byte++;
79 }
80}
81
82/*
83 * This function is called when we flush something into a buffer without
84 * logging it. This happens for things like inodes which are logged
85 * separately from the buffer.
86 */
87void
88xfs_buf_item_flush_log_debug(
89 xfs_buf_t *bp,
90 uint first,
91 uint last)
92{
93 xfs_buf_log_item_t *bip;
94 uint nbytes;
95
96 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
97 if ((bip == NULL) || (bip->bli_item.li_type != XFS_LI_BUF)) {
98 return;
99 }
100
101 ASSERT(bip->bli_logged != NULL);
102 nbytes = last - first + 1;
103 bfset(bip->bli_logged, first, nbytes);
104}
105
106/*
Nathan Scottc41564b2006-03-29 08:55:14 +1000107 * This function is called to verify that our callers have logged
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 * all the bytes that they changed.
109 *
110 * It does this by comparing the original copy of the buffer stored in
111 * the buf log item's bli_orig array to the current copy of the buffer
Nathan Scottc41564b2006-03-29 08:55:14 +1000112 * and ensuring that all bytes which mismatch are set in the bli_logged
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 * array of the buf log item.
114 */
115STATIC void
116xfs_buf_item_log_check(
117 xfs_buf_log_item_t *bip)
118{
119 char *orig;
120 char *buffer;
121 int x;
122 xfs_buf_t *bp;
123
124 ASSERT(bip->bli_orig != NULL);
125 ASSERT(bip->bli_logged != NULL);
126
127 bp = bip->bli_buf;
128 ASSERT(XFS_BUF_COUNT(bp) > 0);
129 ASSERT(XFS_BUF_PTR(bp) != NULL);
130 orig = bip->bli_orig;
131 buffer = XFS_BUF_PTR(bp);
132 for (x = 0; x < XFS_BUF_COUNT(bp); x++) {
133 if (orig[x] != buffer[x] && !btst(bip->bli_logged, x))
134 cmn_err(CE_PANIC,
135 "xfs_buf_item_log_check bip %x buffer %x orig %x index %d",
136 bip, bp, orig, x);
137 }
138}
139#else
140#define xfs_buf_item_log_debug(x,y,z)
141#define xfs_buf_item_log_check(x)
142#endif
143
144STATIC void xfs_buf_error_relse(xfs_buf_t *bp);
145STATIC void xfs_buf_do_callbacks(xfs_buf_t *bp, xfs_log_item_t *lip);
146
147/*
148 * This returns the number of log iovecs needed to log the
149 * given buf log item.
150 *
151 * It calculates this as 1 iovec for the buf log format structure
152 * and 1 for each stretch of non-contiguous chunks to be logged.
153 * Contiguous chunks are logged in a single iovec.
154 *
155 * If the XFS_BLI_STALE flag has been set, then log nothing.
156 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000157STATIC uint
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158xfs_buf_item_size(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000159 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000161 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
162 struct xfs_buf *bp = bip->bli_buf;
163 uint nvecs;
164 int next_bit;
165 int last_bit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
167 ASSERT(atomic_read(&bip->bli_refcount) > 0);
168 if (bip->bli_flags & XFS_BLI_STALE) {
169 /*
170 * The buffer is stale, so all we need to log
171 * is the buf log format structure with the
172 * cancel flag in it.
173 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000174 trace_xfs_buf_item_size_stale(bip);
Dave Chinnerc1155412010-05-07 11:05:19 +1000175 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 return 1;
177 }
178
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
180 nvecs = 1;
181 last_bit = xfs_next_bit(bip->bli_format.blf_data_map,
182 bip->bli_format.blf_map_size, 0);
183 ASSERT(last_bit != -1);
184 nvecs++;
185 while (last_bit != -1) {
186 /*
187 * This takes the bit number to start looking from and
188 * returns the next set bit from there. It returns -1
189 * if there are no more bits set or the start bit is
190 * beyond the end of the bitmap.
191 */
192 next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
193 bip->bli_format.blf_map_size,
194 last_bit + 1);
195 /*
196 * If we run out of bits, leave the loop,
197 * else if we find a new set of bits bump the number of vecs,
198 * else keep scanning the current set of bits.
199 */
200 if (next_bit == -1) {
201 last_bit = -1;
202 } else if (next_bit != last_bit + 1) {
203 last_bit = next_bit;
204 nvecs++;
Dave Chinnerc1155412010-05-07 11:05:19 +1000205 } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) !=
206 (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) +
207 XFS_BLF_CHUNK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 last_bit = next_bit;
209 nvecs++;
210 } else {
211 last_bit++;
212 }
213 }
214
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000215 trace_xfs_buf_item_size(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 return nvecs;
217}
218
219/*
220 * This is called to fill in the vector of log iovecs for the
221 * given log buf item. It fills the first entry with a buf log
222 * format structure, and the rest point to contiguous chunks
223 * within the buffer.
224 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000225STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226xfs_buf_item_format(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000227 struct xfs_log_item *lip,
228 struct xfs_log_iovec *vecp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000230 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
231 struct xfs_buf *bp = bip->bli_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 uint base_size;
233 uint nvecs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 int first_bit;
235 int last_bit;
236 int next_bit;
237 uint nbits;
238 uint buffer_offset;
239
240 ASSERT(atomic_read(&bip->bli_refcount) > 0);
241 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
242 (bip->bli_flags & XFS_BLI_STALE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
244 /*
245 * The size of the base structure is the size of the
246 * declared structure plus the space for the extra words
247 * of the bitmap. We subtract one from the map size, because
248 * the first element of the bitmap is accounted for in the
249 * size of the base structure.
250 */
251 base_size =
252 (uint)(sizeof(xfs_buf_log_format_t) +
253 ((bip->bli_format.blf_map_size - 1) * sizeof(uint)));
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +1000254 vecp->i_addr = &bip->bli_format;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 vecp->i_len = base_size;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000256 vecp->i_type = XLOG_REG_TYPE_BFORMAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 vecp++;
258 nvecs = 1;
259
Dave Chinnerccf7c232010-05-20 23:19:42 +1000260 /*
261 * If it is an inode buffer, transfer the in-memory state to the
262 * format flags and clear the in-memory state. We do not transfer
263 * this state if the inode buffer allocation has not yet been committed
264 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
265 * correct replay of the inode allocation.
266 */
267 if (bip->bli_flags & XFS_BLI_INODE_BUF) {
268 if (!((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000269 xfs_log_item_in_current_chkpt(lip)))
Dave Chinnerccf7c232010-05-20 23:19:42 +1000270 bip->bli_format.blf_flags |= XFS_BLF_INODE_BUF;
271 bip->bli_flags &= ~XFS_BLI_INODE_BUF;
272 }
273
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 if (bip->bli_flags & XFS_BLI_STALE) {
275 /*
276 * The buffer is stale, so all we need to log
277 * is the buf log format structure with the
278 * cancel flag in it.
279 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000280 trace_xfs_buf_item_format_stale(bip);
Dave Chinnerc1155412010-05-07 11:05:19 +1000281 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 bip->bli_format.blf_size = nvecs;
283 return;
284 }
285
286 /*
287 * Fill in an iovec for each set of contiguous chunks.
288 */
289 first_bit = xfs_next_bit(bip->bli_format.blf_data_map,
290 bip->bli_format.blf_map_size, 0);
291 ASSERT(first_bit != -1);
292 last_bit = first_bit;
293 nbits = 1;
294 for (;;) {
295 /*
296 * This takes the bit number to start looking from and
297 * returns the next set bit from there. It returns -1
298 * if there are no more bits set or the start bit is
299 * beyond the end of the bitmap.
300 */
301 next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
302 bip->bli_format.blf_map_size,
303 (uint)last_bit + 1);
304 /*
305 * If we run out of bits fill in the last iovec and get
306 * out of the loop.
307 * Else if we start a new set of bits then fill in the
308 * iovec for the series we were looking at and start
309 * counting the bits in the new one.
310 * Else we're still in the same set of bits so just
311 * keep counting and scanning.
312 */
313 if (next_bit == -1) {
Dave Chinnerc1155412010-05-07 11:05:19 +1000314 buffer_offset = first_bit * XFS_BLF_CHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
Dave Chinnerc1155412010-05-07 11:05:19 +1000316 vecp->i_len = nbits * XFS_BLF_CHUNK;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000317 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 nvecs++;
319 break;
320 } else if (next_bit != last_bit + 1) {
Dave Chinnerc1155412010-05-07 11:05:19 +1000321 buffer_offset = first_bit * XFS_BLF_CHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
Dave Chinnerc1155412010-05-07 11:05:19 +1000323 vecp->i_len = nbits * XFS_BLF_CHUNK;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000324 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 nvecs++;
326 vecp++;
327 first_bit = next_bit;
328 last_bit = next_bit;
329 nbits = 1;
Dave Chinnerc1155412010-05-07 11:05:19 +1000330 } else if (xfs_buf_offset(bp, next_bit << XFS_BLF_SHIFT) !=
331 (xfs_buf_offset(bp, last_bit << XFS_BLF_SHIFT) +
332 XFS_BLF_CHUNK)) {
333 buffer_offset = first_bit * XFS_BLF_CHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
Dave Chinnerc1155412010-05-07 11:05:19 +1000335 vecp->i_len = nbits * XFS_BLF_CHUNK;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000336 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337/* You would think we need to bump the nvecs here too, but we do not
338 * this number is used by recovery, and it gets confused by the boundary
339 * split here
340 * nvecs++;
341 */
342 vecp++;
343 first_bit = next_bit;
344 last_bit = next_bit;
345 nbits = 1;
346 } else {
347 last_bit++;
348 nbits++;
349 }
350 }
351 bip->bli_format.blf_size = nvecs;
352
353 /*
354 * Check to make sure everything is consistent.
355 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000356 trace_xfs_buf_item_format(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 xfs_buf_item_log_check(bip);
358}
359
360/*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000361 * This is called to pin the buffer associated with the buf log item in memory
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000362 * so it cannot be written out.
Dave Chinner64fc35d2010-05-07 11:04:34 +1000363 *
364 * We also always take a reference to the buffer log item here so that the bli
365 * is held while the item is pinned in memory. This means that we can
366 * unconditionally drop the reference count a transaction holds when the
367 * transaction is completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000369STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370xfs_buf_item_pin(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000371 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000373 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000375 ASSERT(XFS_BUF_ISBUSY(bip->bli_buf));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 ASSERT(atomic_read(&bip->bli_refcount) > 0);
377 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
378 (bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000379
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000380 trace_xfs_buf_item_pin(bip);
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000381
382 atomic_inc(&bip->bli_refcount);
383 atomic_inc(&bip->bli_buf->b_pin_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384}
385
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386/*
387 * This is called to unpin the buffer associated with the buf log
388 * item which was previously pinned with a call to xfs_buf_item_pin().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 *
390 * Also drop the reference to the buf item for the current transaction.
391 * If the XFS_BLI_STALE flag is set and we are the last reference,
392 * then free up the buf log item and unlock the buffer.
Christoph Hellwig9412e312010-06-23 18:11:15 +1000393 *
394 * If the remove flag is set we are called from uncommit in the
395 * forced-shutdown path. If that is true and the reference count on
396 * the log item is going to drop to zero we need to free the item's
397 * descriptor in the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000399STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400xfs_buf_item_unpin(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000401 struct xfs_log_item *lip,
Christoph Hellwig9412e312010-06-23 18:11:15 +1000402 int remove)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000404 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000405 xfs_buf_t *bp = bip->bli_buf;
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000406 struct xfs_ail *ailp = lip->li_ailp;
Dave Chinner8e123852010-03-08 11:26:03 +1100407 int stale = bip->bli_flags & XFS_BLI_STALE;
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000408 int freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 ASSERT(XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *) == bip);
411 ASSERT(atomic_read(&bip->bli_refcount) > 0);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000412
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000413 trace_xfs_buf_item_unpin(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
415 freed = atomic_dec_and_test(&bip->bli_refcount);
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000416
417 if (atomic_dec_and_test(&bp->b_pin_count))
418 wake_up_all(&bp->b_waiters);
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000419
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 if (freed && stale) {
421 ASSERT(bip->bli_flags & XFS_BLI_STALE);
422 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
423 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
424 ASSERT(XFS_BUF_ISSTALE(bp));
Dave Chinnerc1155412010-05-07 11:05:19 +1000425 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000426
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000427 trace_xfs_buf_item_unpin_stale(bip);
428
Christoph Hellwig9412e312010-06-23 18:11:15 +1000429 if (remove) {
430 /*
431 * We have to remove the log item from the transaction
432 * as we are about to release our reference to the
433 * buffer. If we don't, the unlock that occurs later
434 * in xfs_trans_uncommit() will ry to reference the
435 * buffer which we no longer have a hold on.
436 */
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000437 xfs_trans_del_item(lip);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000438
439 /*
440 * Since the transaction no longer refers to the buffer,
441 * the buffer should no longer refer to the transaction.
442 */
443 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
444 }
445
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 /*
447 * If we get called here because of an IO error, we may
David Chinner783a2f62008-10-30 17:39:58 +1100448 * or may not have the item on the AIL. xfs_trans_ail_delete()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 * will take care of that situation.
David Chinner783a2f62008-10-30 17:39:58 +1100450 * xfs_trans_ail_delete() drops the AIL lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 */
452 if (bip->bli_flags & XFS_BLI_STALE_INODE) {
453 xfs_buf_do_callbacks(bp, (xfs_log_item_t *)bip);
454 XFS_BUF_SET_FSPRIVATE(bp, NULL);
455 XFS_BUF_CLR_IODONE_FUNC(bp);
456 } else {
David Chinner783a2f62008-10-30 17:39:58 +1100457 spin_lock(&ailp->xa_lock);
458 xfs_trans_ail_delete(ailp, (xfs_log_item_t *)bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 xfs_buf_item_relse(bp);
460 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL);
461 }
462 xfs_buf_relse(bp);
463 }
464}
465
466/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 * This is called to attempt to lock the buffer associated with this
468 * buf log item. Don't sleep on the buffer lock. If we can't get
Dave Chinnerd808f612010-02-02 10:13:42 +1100469 * the lock right away, return 0. If we can get the lock, take a
470 * reference to the buffer. If this is a delayed write buffer that
471 * needs AIL help to be written back, invoke the pushbuf routine
472 * rather than the normal success path.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000474STATIC uint
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475xfs_buf_item_trylock(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000476 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000478 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
479 struct xfs_buf *bp = bip->bli_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
Dave Chinnerd808f612010-02-02 10:13:42 +1100481 if (XFS_BUF_ISPINNED(bp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 return XFS_ITEM_PINNED;
Dave Chinnerd808f612010-02-02 10:13:42 +1100483 if (!XFS_BUF_CPSEMA(bp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 return XFS_ITEM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485
Dave Chinnerd808f612010-02-02 10:13:42 +1100486 /* take a reference to the buffer. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 XFS_BUF_HOLD(bp);
488
489 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000490 trace_xfs_buf_item_trylock(bip);
Dave Chinnerd808f612010-02-02 10:13:42 +1100491 if (XFS_BUF_ISDELAYWRITE(bp))
492 return XFS_ITEM_PUSHBUF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 return XFS_ITEM_SUCCESS;
494}
495
496/*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000497 * Release the buffer associated with the buf log item. If there is no dirty
498 * logged data associated with the buffer recorded in the buf log item, then
499 * free the buf log item and remove the reference to it in the buffer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 *
Dave Chinner64fc35d2010-05-07 11:04:34 +1000501 * This call ignores the recursion count. It is only called when the buffer
502 * should REALLY be unlocked, regardless of the recursion count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 *
Dave Chinner64fc35d2010-05-07 11:04:34 +1000504 * We unconditionally drop the transaction's reference to the log item. If the
505 * item was logged, then another reference was taken when it was pinned, so we
506 * can safely drop the transaction reference now. This also allows us to avoid
507 * potential races with the unpin code freeing the bli by not referencing the
508 * bli after we've dropped the reference count.
509 *
510 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
511 * if necessary but do not unlock the buffer. This is for support of
512 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
513 * free the item.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000515STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516xfs_buf_item_unlock(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000517 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000519 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
520 struct xfs_buf *bp = bip->bli_buf;
521 int aborted;
522 uint hold;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523
Dave Chinner64fc35d2010-05-07 11:04:34 +1000524 /* Clear the buffer's association with this transaction. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
526
527 /*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000528 * If this is a transaction abort, don't return early. Instead, allow
529 * the brelse to happen. Normally it would be done for stale
530 * (cancelled) buffers at unpin time, but we'll never go through the
531 * pin/unpin cycle if we abort inside commit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 */
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000533 aborted = (lip->li_flags & XFS_LI_ABORTED) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
535 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 * Before possibly freeing the buf item, determine if we should
537 * release the buffer at the end of this routine.
538 */
539 hold = bip->bli_flags & XFS_BLI_HOLD;
Dave Chinner64fc35d2010-05-07 11:04:34 +1000540
541 /* Clear the per transaction state. */
542 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD);
543
544 /*
545 * If the buf item is marked stale, then don't do anything. We'll
546 * unlock the buffer and free the buf item when the buffer is unpinned
547 * for the last time.
548 */
549 if (bip->bli_flags & XFS_BLI_STALE) {
550 trace_xfs_buf_item_unlock_stale(bip);
Dave Chinnerc1155412010-05-07 11:05:19 +1000551 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Dave Chinner64fc35d2010-05-07 11:04:34 +1000552 if (!aborted) {
553 atomic_dec(&bip->bli_refcount);
554 return;
555 }
556 }
557
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000558 trace_xfs_buf_item_unlock(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
560 /*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000561 * If the buf item isn't tracking any data, free it, otherwise drop the
562 * reference we hold to it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 */
Eric Sandeen24ad33f2007-06-28 16:43:30 +1000564 if (xfs_bitmap_empty(bip->bli_format.blf_data_map,
Dave Chinner64fc35d2010-05-07 11:04:34 +1000565 bip->bli_format.blf_map_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 xfs_buf_item_relse(bp);
Dave Chinner64fc35d2010-05-07 11:04:34 +1000567 else
568 atomic_dec(&bip->bli_refcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
Dave Chinner64fc35d2010-05-07 11:04:34 +1000570 if (!hold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572}
573
574/*
575 * This is called to find out where the oldest active copy of the
576 * buf log item in the on disk log resides now that the last log
577 * write of it completed at the given lsn.
578 * We always re-log all the dirty data in a buffer, so usually the
579 * latest copy in the on disk log is the only one that matters. For
580 * those cases we simply return the given lsn.
581 *
582 * The one exception to this is for buffers full of newly allocated
583 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
584 * flag set, indicating that only the di_next_unlinked fields from the
585 * inodes in the buffers will be replayed during recovery. If the
586 * original newly allocated inode images have not yet been flushed
587 * when the buffer is so relogged, then we need to make sure that we
588 * keep the old images in the 'active' portion of the log. We do this
589 * by returning the original lsn of that transaction here rather than
590 * the current one.
591 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000592STATIC xfs_lsn_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593xfs_buf_item_committed(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000594 struct xfs_log_item *lip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 xfs_lsn_t lsn)
596{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000597 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
598
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000599 trace_xfs_buf_item_committed(bip);
600
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000601 if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0)
602 return lip->li_lsn;
603 return lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604}
605
606/*
Dave Chinnerd808f612010-02-02 10:13:42 +1100607 * The buffer is locked, but is not a delayed write buffer. This happens
608 * if we race with IO completion and hence we don't want to try to write it
609 * again. Just release the buffer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000611STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612xfs_buf_item_push(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000613 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000615 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
616 struct xfs_buf *bp = bip->bli_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617
618 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000619 ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
620
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000621 trace_xfs_buf_item_push(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622
Dave Chinnerd808f612010-02-02 10:13:42 +1100623 xfs_buf_relse(bp);
624}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625
Dave Chinnerd808f612010-02-02 10:13:42 +1100626/*
627 * The buffer is locked and is a delayed write buffer. Promote the buffer
628 * in the delayed write queue as the caller knows that they must invoke
629 * the xfsbufd to get this buffer written. We have to unlock the buffer
630 * to allow the xfsbufd to write it, too.
631 */
632STATIC void
633xfs_buf_item_pushbuf(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000634 struct xfs_log_item *lip)
Dave Chinnerd808f612010-02-02 10:13:42 +1100635{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000636 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
637 struct xfs_buf *bp = bip->bli_buf;
Dave Chinnerd808f612010-02-02 10:13:42 +1100638
639 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000640 ASSERT(XFS_BUF_ISDELAYWRITE(bp));
641
Dave Chinnerd808f612010-02-02 10:13:42 +1100642 trace_xfs_buf_item_pushbuf(bip);
643
Dave Chinnerd808f612010-02-02 10:13:42 +1100644 xfs_buf_delwri_promote(bp);
645 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646}
647
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000648STATIC void
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000649xfs_buf_item_committing(
650 struct xfs_log_item *lip,
651 xfs_lsn_t commit_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652{
653}
654
655/*
656 * This is the ops vector shared by all buf log items.
657 */
David Chinner7989cb82007-02-10 18:34:56 +1100658static struct xfs_item_ops xfs_buf_item_ops = {
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000659 .iop_size = xfs_buf_item_size,
660 .iop_format = xfs_buf_item_format,
661 .iop_pin = xfs_buf_item_pin,
662 .iop_unpin = xfs_buf_item_unpin,
663 .iop_trylock = xfs_buf_item_trylock,
664 .iop_unlock = xfs_buf_item_unlock,
665 .iop_committed = xfs_buf_item_committed,
666 .iop_push = xfs_buf_item_push,
667 .iop_pushbuf = xfs_buf_item_pushbuf,
668 .iop_committing = xfs_buf_item_committing
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669};
670
671
672/*
673 * Allocate a new buf log item to go with the given buffer.
674 * Set the buffer's b_fsprivate field to point to the new
675 * buf log item. If there are other item's attached to the
676 * buffer (see xfs_buf_attach_iodone() below), then put the
677 * buf log item at the front.
678 */
679void
680xfs_buf_item_init(
681 xfs_buf_t *bp,
682 xfs_mount_t *mp)
683{
684 xfs_log_item_t *lip;
685 xfs_buf_log_item_t *bip;
686 int chunks;
687 int map_size;
688
689 /*
690 * Check to see if there is already a buf log item for
691 * this buffer. If there is, it is guaranteed to be
692 * the first. If we do already have one, there is
693 * nothing to do here so return.
694 */
Christoph Hellwig15ac08a2008-12-09 04:47:30 -0500695 if (bp->b_mount != mp)
696 bp->b_mount = mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
698 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
699 if (lip->li_type == XFS_LI_BUF) {
700 return;
701 }
702 }
703
704 /*
Dave Chinnerc1155412010-05-07 11:05:19 +1000705 * chunks is the number of XFS_BLF_CHUNK size pieces
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 * the buffer can be divided into. Make sure not to
707 * truncate any pieces. map_size is the size of the
708 * bitmap needed to describe the chunks of the buffer.
709 */
Dave Chinnerc1155412010-05-07 11:05:19 +1000710 chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLF_CHUNK - 1)) >> XFS_BLF_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT);
712
713 bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone,
714 KM_SLEEP);
Dave Chinner43f5efc2010-03-23 10:10:00 +1100715 xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 bip->bli_buf = bp;
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000717 xfs_buf_hold(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 bip->bli_format.blf_type = XFS_LI_BUF;
719 bip->bli_format.blf_blkno = (__int64_t)XFS_BUF_ADDR(bp);
720 bip->bli_format.blf_len = (ushort)BTOBB(XFS_BUF_COUNT(bp));
721 bip->bli_format.blf_map_size = map_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
723#ifdef XFS_TRANS_DEBUG
724 /*
725 * Allocate the arrays for tracking what needs to be logged
726 * and what our callers request to be logged. bli_orig
727 * holds a copy of the original, clean buffer for comparison
728 * against, and bli_logged keeps a 1 bit flag per byte in
729 * the buffer to indicate which bytes the callers have asked
730 * to have logged.
731 */
732 bip->bli_orig = (char *)kmem_alloc(XFS_BUF_COUNT(bp), KM_SLEEP);
733 memcpy(bip->bli_orig, XFS_BUF_PTR(bp), XFS_BUF_COUNT(bp));
734 bip->bli_logged = (char *)kmem_zalloc(XFS_BUF_COUNT(bp) / NBBY, KM_SLEEP);
735#endif
736
737 /*
738 * Put the buf item into the list of items attached to the
739 * buffer at the front.
740 */
741 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
742 bip->bli_item.li_bio_list =
743 XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
744 }
745 XFS_BUF_SET_FSPRIVATE(bp, bip);
746}
747
748
749/*
750 * Mark bytes first through last inclusive as dirty in the buf
751 * item's bitmap.
752 */
753void
754xfs_buf_item_log(
755 xfs_buf_log_item_t *bip,
756 uint first,
757 uint last)
758{
759 uint first_bit;
760 uint last_bit;
761 uint bits_to_set;
762 uint bits_set;
763 uint word_num;
764 uint *wordp;
765 uint bit;
766 uint end_bit;
767 uint mask;
768
769 /*
770 * Mark the item as having some dirty data for
771 * quick reference in xfs_buf_item_dirty.
772 */
773 bip->bli_flags |= XFS_BLI_DIRTY;
774
775 /*
776 * Convert byte offsets to bit numbers.
777 */
Dave Chinnerc1155412010-05-07 11:05:19 +1000778 first_bit = first >> XFS_BLF_SHIFT;
779 last_bit = last >> XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780
781 /*
782 * Calculate the total number of bits to be set.
783 */
784 bits_to_set = last_bit - first_bit + 1;
785
786 /*
787 * Get a pointer to the first word in the bitmap
788 * to set a bit in.
789 */
790 word_num = first_bit >> BIT_TO_WORD_SHIFT;
791 wordp = &(bip->bli_format.blf_data_map[word_num]);
792
793 /*
794 * Calculate the starting bit in the first word.
795 */
796 bit = first_bit & (uint)(NBWORD - 1);
797
798 /*
799 * First set any bits in the first word of our range.
800 * If it starts at bit 0 of the word, it will be
801 * set below rather than here. That is what the variable
802 * bit tells us. The variable bits_set tracks the number
803 * of bits that have been set so far. End_bit is the number
804 * of the last bit to be set in this word plus one.
805 */
806 if (bit) {
807 end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
808 mask = ((1 << (end_bit - bit)) - 1) << bit;
809 *wordp |= mask;
810 wordp++;
811 bits_set = end_bit - bit;
812 } else {
813 bits_set = 0;
814 }
815
816 /*
817 * Now set bits a whole word at a time that are between
818 * first_bit and last_bit.
819 */
820 while ((bits_to_set - bits_set) >= NBWORD) {
821 *wordp |= 0xffffffff;
822 bits_set += NBWORD;
823 wordp++;
824 }
825
826 /*
827 * Finally, set any bits left to be set in one last partial word.
828 */
829 end_bit = bits_to_set - bits_set;
830 if (end_bit) {
831 mask = (1 << end_bit) - 1;
832 *wordp |= mask;
833 }
834
835 xfs_buf_item_log_debug(bip, first, last);
836}
837
838
839/*
840 * Return 1 if the buffer has some data that has been logged (at any
841 * point, not just the current transaction) and 0 if not.
842 */
843uint
844xfs_buf_item_dirty(
845 xfs_buf_log_item_t *bip)
846{
847 return (bip->bli_flags & XFS_BLI_DIRTY);
848}
849
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000850STATIC void
851xfs_buf_item_free(
852 xfs_buf_log_item_t *bip)
853{
854#ifdef XFS_TRANS_DEBUG
855 kmem_free(bip->bli_orig);
856 kmem_free(bip->bli_logged);
857#endif /* XFS_TRANS_DEBUG */
858
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000859 kmem_zone_free(xfs_buf_item_zone, bip);
860}
861
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862/*
863 * This is called when the buf log item is no longer needed. It should
864 * free the buf log item associated with the given buffer and clear
865 * the buffer's pointer to the buf log item. If there are no more
866 * items in the list, clear the b_iodone field of the buffer (see
867 * xfs_buf_attach_iodone() below).
868 */
869void
870xfs_buf_item_relse(
871 xfs_buf_t *bp)
872{
873 xfs_buf_log_item_t *bip;
874
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000875 trace_xfs_buf_item_relse(bp, _RET_IP_);
876
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
878 XFS_BUF_SET_FSPRIVATE(bp, bip->bli_item.li_bio_list);
879 if ((XFS_BUF_FSPRIVATE(bp, void *) == NULL) &&
880 (XFS_BUF_IODONE_FUNC(bp) != NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 XFS_BUF_CLR_IODONE_FUNC(bp);
882 }
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000883 xfs_buf_rele(bp);
884 xfs_buf_item_free(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885}
886
887
888/*
889 * Add the given log item with its callback to the list of callbacks
890 * to be called when the buffer's I/O completes. If it is not set
891 * already, set the buffer's b_iodone() routine to be
892 * xfs_buf_iodone_callbacks() and link the log item into the list of
893 * items rooted at b_fsprivate. Items are always added as the second
894 * entry in the list if there is a first, because the buf item code
895 * assumes that the buf log item is first.
896 */
897void
898xfs_buf_attach_iodone(
899 xfs_buf_t *bp,
900 void (*cb)(xfs_buf_t *, xfs_log_item_t *),
901 xfs_log_item_t *lip)
902{
903 xfs_log_item_t *head_lip;
904
905 ASSERT(XFS_BUF_ISBUSY(bp));
906 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
907
908 lip->li_cb = cb;
909 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
910 head_lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
911 lip->li_bio_list = head_lip->li_bio_list;
912 head_lip->li_bio_list = lip;
913 } else {
914 XFS_BUF_SET_FSPRIVATE(bp, lip);
915 }
916
917 ASSERT((XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks) ||
918 (XFS_BUF_IODONE_FUNC(bp) == NULL));
919 XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks);
920}
921
922STATIC void
923xfs_buf_do_callbacks(
924 xfs_buf_t *bp,
925 xfs_log_item_t *lip)
926{
927 xfs_log_item_t *nlip;
928
929 while (lip != NULL) {
930 nlip = lip->li_bio_list;
931 ASSERT(lip->li_cb != NULL);
932 /*
933 * Clear the next pointer so we don't have any
934 * confusion if the item is added to another buf.
935 * Don't touch the log item after calling its
936 * callback, because it could have freed itself.
937 */
938 lip->li_bio_list = NULL;
939 lip->li_cb(bp, lip);
940 lip = nlip;
941 }
942}
943
944/*
945 * This is the iodone() function for buffers which have had callbacks
946 * attached to them by xfs_buf_attach_iodone(). It should remove each
947 * log item from the buffer's list and call the callback of each in turn.
948 * When done, the buffer's fsprivate field is set to NULL and the buffer
949 * is unlocked with a call to iodone().
950 */
951void
952xfs_buf_iodone_callbacks(
953 xfs_buf_t *bp)
954{
955 xfs_log_item_t *lip;
956 static ulong lasttime;
957 static xfs_buftarg_t *lasttarg;
958 xfs_mount_t *mp;
959
960 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
961 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
962
963 if (XFS_BUF_GETERROR(bp) != 0) {
964 /*
965 * If we've already decided to shutdown the filesystem
966 * because of IO errors, there's no point in giving this
967 * a retry.
968 */
969 mp = lip->li_mountp;
970 if (XFS_FORCED_SHUTDOWN(mp)) {
971 ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp);
972 XFS_BUF_SUPER_STALE(bp);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000973 trace_xfs_buf_item_iodone(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 xfs_buf_do_callbacks(bp, lip);
975 XFS_BUF_SET_FSPRIVATE(bp, NULL);
976 XFS_BUF_CLR_IODONE_FUNC(bp);
Lachlan McIlroy4fdc7782008-12-22 17:52:58 +1100977 xfs_biodone(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 return;
979 }
980
981 if ((XFS_BUF_TARGET(bp) != lasttarg) ||
982 (time_after(jiffies, (lasttime + 5*HZ)))) {
983 lasttime = jiffies;
Nathan Scottb6574522006-06-09 15:29:40 +1000984 cmn_err(CE_ALERT, "Device %s, XFS metadata write error"
985 " block 0x%llx in %s",
986 XFS_BUFTARG_NAME(XFS_BUF_TARGET(bp)),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 (__uint64_t)XFS_BUF_ADDR(bp), mp->m_fsname);
988 }
989 lasttarg = XFS_BUF_TARGET(bp);
990
991 if (XFS_BUF_ISASYNC(bp)) {
992 /*
993 * If the write was asynchronous then noone will be
994 * looking for the error. Clear the error state
995 * and write the buffer out again delayed write.
996 *
997 * XXXsup This is OK, so long as we catch these
998 * before we start the umount; we don't want these
999 * DELWRI metadata bufs to be hanging around.
1000 */
1001 XFS_BUF_ERROR(bp,0); /* errno of 0 unsets the flag */
1002
1003 if (!(XFS_BUF_ISSTALE(bp))) {
1004 XFS_BUF_DELAYWRITE(bp);
1005 XFS_BUF_DONE(bp);
1006 XFS_BUF_SET_START(bp);
1007 }
1008 ASSERT(XFS_BUF_IODONE_FUNC(bp));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001009 trace_xfs_buf_item_iodone_async(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 xfs_buf_relse(bp);
1011 } else {
1012 /*
1013 * If the write of the buffer was not asynchronous,
1014 * then we want to make sure to return the error
1015 * to the caller of bwrite(). Because of this we
1016 * cannot clear the B_ERROR state at this point.
1017 * Instead we install a callback function that
1018 * will be called when the buffer is released, and
1019 * that routine will clear the error state and
1020 * set the buffer to be written out again after
1021 * some delay.
1022 */
1023 /* We actually overwrite the existing b-relse
1024 function at times, but we're gonna be shutting down
1025 anyway. */
1026 XFS_BUF_SET_BRELSE_FUNC(bp,xfs_buf_error_relse);
1027 XFS_BUF_DONE(bp);
David Chinnerb4dd3302008-08-13 16:36:11 +10001028 XFS_BUF_FINISH_IOWAIT(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 }
1030 return;
1031 }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001032
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 xfs_buf_do_callbacks(bp, lip);
1034 XFS_BUF_SET_FSPRIVATE(bp, NULL);
1035 XFS_BUF_CLR_IODONE_FUNC(bp);
1036 xfs_biodone(bp);
1037}
1038
1039/*
1040 * This is a callback routine attached to a buffer which gets an error
1041 * when being written out synchronously.
1042 */
1043STATIC void
1044xfs_buf_error_relse(
1045 xfs_buf_t *bp)
1046{
1047 xfs_log_item_t *lip;
1048 xfs_mount_t *mp;
1049
1050 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
1051 mp = (xfs_mount_t *)lip->li_mountp;
1052 ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp);
1053
1054 XFS_BUF_STALE(bp);
1055 XFS_BUF_DONE(bp);
1056 XFS_BUF_UNDELAYWRITE(bp);
1057 XFS_BUF_ERROR(bp,0);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001058
1059 trace_xfs_buf_error_relse(bp, _RET_IP_);
1060
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 if (! XFS_FORCED_SHUTDOWN(mp))
Nathan Scott7d04a332006-06-09 14:58:38 +10001062 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 /*
1064 * We have to unpin the pinned buffers so do the
1065 * callbacks.
1066 */
1067 xfs_buf_do_callbacks(bp, lip);
1068 XFS_BUF_SET_FSPRIVATE(bp, NULL);
1069 XFS_BUF_CLR_IODONE_FUNC(bp);
1070 XFS_BUF_SET_BRELSE_FUNC(bp,NULL);
1071 xfs_buf_relse(bp);
1072}
1073
1074
1075/*
1076 * This is the iodone() function for buffers which have been
1077 * logged. It is called when they are eventually flushed out.
1078 * It should remove the buf item from the AIL, and free the buf item.
1079 * It is called by xfs_buf_iodone_callbacks() above which will take
1080 * care of cleaning up the buffer itself.
1081 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082void
1083xfs_buf_iodone(
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001084 struct xfs_buf *bp,
1085 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086{
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001087 struct xfs_ail *ailp = lip->li_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001089 ASSERT(BUF_ITEM(lip)->bli_buf == bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +10001091 xfs_buf_rele(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
1093 /*
1094 * If we are forcibly shutting down, this may well be
1095 * off the AIL already. That's because we simulate the
1096 * log-committed callbacks to unpin these buffers. Or we may never
1097 * have put this item on AIL because of the transaction was
David Chinner783a2f62008-10-30 17:39:58 +11001098 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 *
1100 * Either way, AIL is useless if we're forcing a shutdown.
1101 */
David Chinnerfc1829f2008-10-30 17:39:46 +11001102 spin_lock(&ailp->xa_lock);
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001103 xfs_trans_ail_delete(ailp, lip);
1104 xfs_buf_item_free(BUF_ITEM(lip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105}