blob: a8d0ed911196120a80a26bb7232f79d3a5badee1 [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"
23#include "xfs_trans.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_sb.h"
David Chinnerda353b02007-08-28 14:00:13 +100025#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_mount.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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32
33kmem_zone_t *xfs_buf_item_zone;
34
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +100035static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip)
36{
37 return container_of(lip, struct xfs_buf_log_item, bli_item);
38}
39
40
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#ifdef XFS_TRANS_DEBUG
42/*
43 * This function uses an alternate strategy for tracking the bytes
44 * that the user requests to be logged. This can then be used
45 * in conjunction with the bli_orig array in the buf log item to
46 * catch bugs in our callers' code.
47 *
48 * We also double check the bits set in xfs_buf_item_log using a
49 * simple algorithm to check that every byte is accounted for.
50 */
51STATIC void
52xfs_buf_item_log_debug(
53 xfs_buf_log_item_t *bip,
54 uint first,
55 uint last)
56{
57 uint x;
58 uint byte;
59 uint nbytes;
60 uint chunk_num;
61 uint word_num;
62 uint bit_num;
63 uint bit_set;
64 uint *wordp;
65
66 ASSERT(bip->bli_logged != NULL);
67 byte = first;
68 nbytes = last - first + 1;
69 bfset(bip->bli_logged, first, nbytes);
70 for (x = 0; x < nbytes; x++) {
Dave Chinnerc1155412010-05-07 11:05:19 +100071 chunk_num = byte >> XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 word_num = chunk_num >> BIT_TO_WORD_SHIFT;
73 bit_num = chunk_num & (NBWORD - 1);
74 wordp = &(bip->bli_format.blf_data_map[word_num]);
75 bit_set = *wordp & (1 << bit_num);
76 ASSERT(bit_set);
77 byte++;
78 }
79}
80
81/*
82 * This function is called when we flush something into a buffer without
83 * logging it. This happens for things like inodes which are logged
84 * separately from the buffer.
85 */
86void
87xfs_buf_item_flush_log_debug(
88 xfs_buf_t *bp,
89 uint first,
90 uint last)
91{
Christoph Hellwigadadbee2011-07-13 13:43:49 +020092 xfs_buf_log_item_t *bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 uint nbytes;
94
Christoph Hellwigadadbee2011-07-13 13:43:49 +020095 if (bip == NULL || (bip->bli_item.li_type != XFS_LI_BUF))
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
98 ASSERT(bip->bli_logged != NULL);
99 nbytes = last - first + 1;
100 bfset(bip->bli_logged, first, nbytes);
101}
102
103/*
Nathan Scottc41564b2006-03-29 08:55:14 +1000104 * This function is called to verify that our callers have logged
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 * all the bytes that they changed.
106 *
107 * It does this by comparing the original copy of the buffer stored in
108 * the buf log item's bli_orig array to the current copy of the buffer
Nathan Scottc41564b2006-03-29 08:55:14 +1000109 * and ensuring that all bytes which mismatch are set in the bli_logged
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 * array of the buf log item.
111 */
112STATIC void
113xfs_buf_item_log_check(
114 xfs_buf_log_item_t *bip)
115{
116 char *orig;
117 char *buffer;
118 int x;
119 xfs_buf_t *bp;
120
121 ASSERT(bip->bli_orig != NULL);
122 ASSERT(bip->bli_logged != NULL);
123
124 bp = bip->bli_buf;
Dave Chinneraa0e8832012-04-23 15:58:52 +1000125 ASSERT(bp->b_length > 0);
Chandra Seetharaman62926042011-07-22 23:40:15 +0000126 ASSERT(bp->b_addr != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 orig = bip->bli_orig;
Chandra Seetharaman62926042011-07-22 23:40:15 +0000128 buffer = bp->b_addr;
Dave Chinneraa0e8832012-04-23 15:58:52 +1000129 for (x = 0; x < BBTOB(bp->b_length); x++) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100130 if (orig[x] != buffer[x] && !btst(bip->bli_logged, x)) {
131 xfs_emerg(bp->b_mount,
132 "%s: bip %x buffer %x orig %x index %d",
133 __func__, bip, bp, orig, x);
134 ASSERT(0);
135 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 }
137}
138#else
139#define xfs_buf_item_log_debug(x,y,z)
140#define xfs_buf_item_log_check(x)
141#endif
142
Dave Chinnerc90821a2010-12-03 17:00:52 +1100143STATIC void xfs_buf_do_callbacks(struct xfs_buf *bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
145/*
146 * This returns the number of log iovecs needed to log the
147 * given buf log item.
148 *
149 * It calculates this as 1 iovec for the buf log format structure
150 * and 1 for each stretch of non-contiguous chunks to be logged.
151 * Contiguous chunks are logged in a single iovec.
152 *
153 * If the XFS_BLI_STALE flag has been set, then log nothing.
154 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000155STATIC uint
Dave Chinner372cc85e2012-06-22 18:50:12 +1000156xfs_buf_item_size_segment(
157 struct xfs_buf_log_item *bip,
158 struct xfs_buf_log_format *blfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000160 struct xfs_buf *bp = bip->bli_buf;
161 uint nvecs;
162 int next_bit;
163 int last_bit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
Dave Chinner372cc85e2012-06-22 18:50:12 +1000165 last_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
166 if (last_bit == -1)
167 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168
Dave Chinner372cc85e2012-06-22 18:50:12 +1000169 /*
170 * initial count for a dirty buffer is 2 vectors - the format structure
171 * and the first dirty region.
172 */
173 nvecs = 2;
174
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 while (last_bit != -1) {
176 /*
177 * This takes the bit number to start looking from and
178 * returns the next set bit from there. It returns -1
179 * if there are no more bits set or the start bit is
180 * beyond the end of the bitmap.
181 */
Dave Chinner372cc85e2012-06-22 18:50:12 +1000182 next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
183 last_bit + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 /*
185 * If we run out of bits, leave the loop,
186 * else if we find a new set of bits bump the number of vecs,
187 * else keep scanning the current set of bits.
188 */
189 if (next_bit == -1) {
Dave Chinner372cc85e2012-06-22 18:50:12 +1000190 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 } else if (next_bit != last_bit + 1) {
192 last_bit = next_bit;
193 nvecs++;
Dave Chinnerc1155412010-05-07 11:05:19 +1000194 } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) !=
195 (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) +
196 XFS_BLF_CHUNK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 last_bit = next_bit;
198 nvecs++;
199 } else {
200 last_bit++;
201 }
202 }
203
Dave Chinner372cc85e2012-06-22 18:50:12 +1000204 return nvecs;
205}
206
207/*
208 * This returns the number of log iovecs needed to log the given buf log item.
209 *
210 * It calculates this as 1 iovec for the buf log format structure and 1 for each
211 * stretch of non-contiguous chunks to be logged. Contiguous chunks are logged
212 * in a single iovec.
213 *
214 * Discontiguous buffers need a format structure per region that that is being
215 * logged. This makes the changes in the buffer appear to log recovery as though
216 * they came from separate buffers, just like would occur if multiple buffers
217 * were used instead of a single discontiguous buffer. This enables
218 * discontiguous buffers to be in-memory constructs, completely transparent to
219 * what ends up on disk.
220 *
221 * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log
222 * format structures.
223 */
224STATIC uint
225xfs_buf_item_size(
226 struct xfs_log_item *lip)
227{
228 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
229 uint nvecs;
230 int i;
231
232 ASSERT(atomic_read(&bip->bli_refcount) > 0);
233 if (bip->bli_flags & XFS_BLI_STALE) {
234 /*
235 * The buffer is stale, so all we need to log
236 * is the buf log format structure with the
237 * cancel flag in it.
238 */
239 trace_xfs_buf_item_size_stale(bip);
240 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
241 return bip->bli_format_count;
242 }
243
244 ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
245
246 /*
247 * the vector count is based on the number of buffer vectors we have
248 * dirty bits in. This will only be greater than one when we have a
249 * compound buffer with more than one segment dirty. Hence for compound
250 * buffers we need to track which segment the dirty bits correspond to,
251 * and when we move from one segment to the next increment the vector
252 * count for the extra buf log format structure that will need to be
253 * written.
254 */
255 nvecs = 0;
256 for (i = 0; i < bip->bli_format_count; i++) {
257 nvecs += xfs_buf_item_size_segment(bip, &bip->bli_formats[i]);
258 }
259
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000260 trace_xfs_buf_item_size(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 return nvecs;
262}
263
Dave Chinner372cc85e2012-06-22 18:50:12 +1000264static struct xfs_log_iovec *
265xfs_buf_item_format_segment(
266 struct xfs_buf_log_item *bip,
267 struct xfs_log_iovec *vecp,
268 uint offset,
269 struct xfs_buf_log_format *blfp)
270{
271 struct xfs_buf *bp = bip->bli_buf;
272 uint base_size;
273 uint nvecs;
274 int first_bit;
275 int last_bit;
276 int next_bit;
277 uint nbits;
278 uint buffer_offset;
279
280 /* copy the flags across from the base format item */
281 blfp->blf_flags = bip->bli_format.blf_flags;
282
283 /*
284 * Base size is the actual size of the ondisk structure - it reflects
285 * the actual size of the dirty bitmap rather than the size of the in
286 * memory structure.
287 */
288 base_size = offsetof(struct xfs_buf_log_format, blf_data_map) +
289 (blfp->blf_map_size * sizeof(blfp->blf_data_map[0]));
290 vecp->i_addr = blfp;
291 vecp->i_len = base_size;
292 vecp->i_type = XLOG_REG_TYPE_BFORMAT;
293 vecp++;
294 nvecs = 1;
295
296 if (bip->bli_flags & XFS_BLI_STALE) {
297 /*
298 * The buffer is stale, so all we need to log
299 * is the buf log format structure with the
300 * cancel flag in it.
301 */
302 trace_xfs_buf_item_format_stale(bip);
303 ASSERT(blfp->blf_flags & XFS_BLF_CANCEL);
304 blfp->blf_size = nvecs;
305 return vecp;
306 }
307
308 /*
309 * Fill in an iovec for each set of contiguous chunks.
310 */
311 first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
312 ASSERT(first_bit != -1);
313 last_bit = first_bit;
314 nbits = 1;
315 for (;;) {
316 /*
317 * This takes the bit number to start looking from and
318 * returns the next set bit from there. It returns -1
319 * if there are no more bits set or the start bit is
320 * beyond the end of the bitmap.
321 */
322 next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
323 (uint)last_bit + 1);
324 /*
325 * If we run out of bits fill in the last iovec and get
326 * out of the loop.
327 * Else if we start a new set of bits then fill in the
328 * iovec for the series we were looking at and start
329 * counting the bits in the new one.
330 * Else we're still in the same set of bits so just
331 * keep counting and scanning.
332 */
333 if (next_bit == -1) {
334 buffer_offset = offset + first_bit * XFS_BLF_CHUNK;
335 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
336 vecp->i_len = nbits * XFS_BLF_CHUNK;
337 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
338 nvecs++;
339 break;
340 } else if (next_bit != last_bit + 1) {
341 buffer_offset = offset + first_bit * XFS_BLF_CHUNK;
342 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
343 vecp->i_len = nbits * XFS_BLF_CHUNK;
344 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
345 nvecs++;
346 vecp++;
347 first_bit = next_bit;
348 last_bit = next_bit;
349 nbits = 1;
350 } else if (xfs_buf_offset(bp, offset +
351 (next_bit << XFS_BLF_SHIFT)) !=
352 (xfs_buf_offset(bp, offset +
353 (last_bit << XFS_BLF_SHIFT)) +
354 XFS_BLF_CHUNK)) {
355 buffer_offset = offset + first_bit * XFS_BLF_CHUNK;
356 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
357 vecp->i_len = nbits * XFS_BLF_CHUNK;
358 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
359/*
360 * You would think we need to bump the nvecs here too, but we do not
361 * this number is used by recovery, and it gets confused by the boundary
362 * split here
363 * nvecs++;
364 */
365 vecp++;
366 first_bit = next_bit;
367 last_bit = next_bit;
368 nbits = 1;
369 } else {
370 last_bit++;
371 nbits++;
372 }
373 }
374 bip->bli_format.blf_size = nvecs;
375 return vecp;
376}
377
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378/*
379 * This is called to fill in the vector of log iovecs for the
380 * given log buf item. It fills the first entry with a buf log
381 * format structure, and the rest point to contiguous chunks
382 * within the buffer.
383 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000384STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385xfs_buf_item_format(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000386 struct xfs_log_item *lip,
387 struct xfs_log_iovec *vecp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000389 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
Dave Chinner372cc85e2012-06-22 18:50:12 +1000390 struct xfs_buf *bp = bip->bli_buf;
391 uint offset = 0;
392 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393
394 ASSERT(atomic_read(&bip->bli_refcount) > 0);
395 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
396 (bip->bli_flags & XFS_BLI_STALE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
398 /*
Dave Chinnerccf7c232010-05-20 23:19:42 +1000399 * If it is an inode buffer, transfer the in-memory state to the
400 * format flags and clear the in-memory state. We do not transfer
401 * this state if the inode buffer allocation has not yet been committed
402 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
403 * correct replay of the inode allocation.
404 */
405 if (bip->bli_flags & XFS_BLI_INODE_BUF) {
406 if (!((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000407 xfs_log_item_in_current_chkpt(lip)))
Dave Chinnerccf7c232010-05-20 23:19:42 +1000408 bip->bli_format.blf_flags |= XFS_BLF_INODE_BUF;
409 bip->bli_flags &= ~XFS_BLI_INODE_BUF;
410 }
411
Dave Chinner372cc85e2012-06-22 18:50:12 +1000412 for (i = 0; i < bip->bli_format_count; i++) {
413 vecp = xfs_buf_item_format_segment(bip, vecp, offset,
414 &bip->bli_formats[i]);
415 offset += bp->b_maps[i].bm_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 }
417
418 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 * Check to make sure everything is consistent.
420 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000421 trace_xfs_buf_item_format(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 xfs_buf_item_log_check(bip);
423}
424
425/*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000426 * This is called to pin the buffer associated with the buf log item in memory
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000427 * so it cannot be written out.
Dave Chinner64fc35d2010-05-07 11:04:34 +1000428 *
429 * We also always take a reference to the buffer log item here so that the bli
430 * is held while the item is pinned in memory. This means that we can
431 * unconditionally drop the reference count a transaction holds when the
432 * transaction is completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000434STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435xfs_buf_item_pin(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000436 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000438 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 ASSERT(atomic_read(&bip->bli_refcount) > 0);
441 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
442 (bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000443
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000444 trace_xfs_buf_item_pin(bip);
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000445
446 atomic_inc(&bip->bli_refcount);
447 atomic_inc(&bip->bli_buf->b_pin_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448}
449
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450/*
451 * This is called to unpin the buffer associated with the buf log
452 * item which was previously pinned with a call to xfs_buf_item_pin().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 *
454 * Also drop the reference to the buf item for the current transaction.
455 * If the XFS_BLI_STALE flag is set and we are the last reference,
456 * then free up the buf log item and unlock the buffer.
Christoph Hellwig9412e312010-06-23 18:11:15 +1000457 *
458 * If the remove flag is set we are called from uncommit in the
459 * forced-shutdown path. If that is true and the reference count on
460 * the log item is going to drop to zero we need to free the item's
461 * descriptor in the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000463STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464xfs_buf_item_unpin(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000465 struct xfs_log_item *lip,
Christoph Hellwig9412e312010-06-23 18:11:15 +1000466 int remove)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000468 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000469 xfs_buf_t *bp = bip->bli_buf;
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000470 struct xfs_ail *ailp = lip->li_ailp;
Dave Chinner8e123852010-03-08 11:26:03 +1100471 int stale = bip->bli_flags & XFS_BLI_STALE;
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000472 int freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200474 ASSERT(bp->b_fspriv == bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 ASSERT(atomic_read(&bip->bli_refcount) > 0);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000476
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000477 trace_xfs_buf_item_unpin(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
479 freed = atomic_dec_and_test(&bip->bli_refcount);
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000480
481 if (atomic_dec_and_test(&bp->b_pin_count))
482 wake_up_all(&bp->b_waiters);
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000483
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 if (freed && stale) {
485 ASSERT(bip->bli_flags & XFS_BLI_STALE);
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200486 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 ASSERT(XFS_BUF_ISSTALE(bp));
Dave Chinnerc1155412010-05-07 11:05:19 +1000488 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000489
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000490 trace_xfs_buf_item_unpin_stale(bip);
491
Christoph Hellwig9412e312010-06-23 18:11:15 +1000492 if (remove) {
493 /*
Dave Chinnere34a3142011-01-27 12:13:35 +1100494 * If we are in a transaction context, we have to
495 * remove the log item from the transaction as we are
496 * about to release our reference to the buffer. If we
497 * don't, the unlock that occurs later in
498 * xfs_trans_uncommit() will try to reference the
Christoph Hellwig9412e312010-06-23 18:11:15 +1000499 * buffer which we no longer have a hold on.
500 */
Dave Chinnere34a3142011-01-27 12:13:35 +1100501 if (lip->li_desc)
502 xfs_trans_del_item(lip);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000503
504 /*
505 * Since the transaction no longer refers to the buffer,
506 * the buffer should no longer refer to the transaction.
507 */
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200508 bp->b_transp = NULL;
Christoph Hellwig9412e312010-06-23 18:11:15 +1000509 }
510
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 /*
512 * If we get called here because of an IO error, we may
David Chinner783a2f62008-10-30 17:39:58 +1100513 * or may not have the item on the AIL. xfs_trans_ail_delete()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 * will take care of that situation.
David Chinner783a2f62008-10-30 17:39:58 +1100515 * xfs_trans_ail_delete() drops the AIL lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 */
517 if (bip->bli_flags & XFS_BLI_STALE_INODE) {
Dave Chinnerc90821a2010-12-03 17:00:52 +1100518 xfs_buf_do_callbacks(bp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200519 bp->b_fspriv = NULL;
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200520 bp->b_iodone = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 } else {
David Chinner783a2f62008-10-30 17:39:58 +1100522 spin_lock(&ailp->xa_lock);
Dave Chinner04913fd2012-04-23 15:58:41 +1000523 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_LOG_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 xfs_buf_item_relse(bp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200525 ASSERT(bp->b_fspriv == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 }
527 xfs_buf_relse(bp);
Christoph Hellwig960c60a2012-04-23 15:58:38 +1000528 } else if (freed && remove) {
529 xfs_buf_lock(bp);
530 xfs_buf_ioerror(bp, EIO);
531 XFS_BUF_UNDONE(bp);
532 xfs_buf_stale(bp);
533 xfs_buf_ioend(bp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 }
535}
536
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000537STATIC uint
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000538xfs_buf_item_push(
539 struct xfs_log_item *lip,
540 struct list_head *buffer_list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000542 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
543 struct xfs_buf *bp = bip->bli_buf;
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000544 uint rval = XFS_ITEM_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
Chandra Seetharaman811e64c2011-07-22 23:40:27 +0000546 if (xfs_buf_ispinned(bp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 return XFS_ITEM_PINNED;
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200548 if (!xfs_buf_trylock(bp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 return XFS_ITEM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000552
553 trace_xfs_buf_item_push(bip);
554
555 if (!xfs_buf_delwri_queue(bp, buffer_list))
556 rval = XFS_ITEM_FLUSHING;
557 xfs_buf_unlock(bp);
558 return rval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559}
560
561/*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000562 * Release the buffer associated with the buf log item. If there is no dirty
563 * logged data associated with the buffer recorded in the buf log item, then
564 * free the buf log item and remove the reference to it in the buffer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 *
Dave Chinner64fc35d2010-05-07 11:04:34 +1000566 * This call ignores the recursion count. It is only called when the buffer
567 * should REALLY be unlocked, regardless of the recursion count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 *
Dave Chinner64fc35d2010-05-07 11:04:34 +1000569 * We unconditionally drop the transaction's reference to the log item. If the
570 * item was logged, then another reference was taken when it was pinned, so we
571 * can safely drop the transaction reference now. This also allows us to avoid
572 * potential races with the unpin code freeing the bli by not referencing the
573 * bli after we've dropped the reference count.
574 *
575 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
576 * if necessary but do not unlock the buffer. This is for support of
577 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
578 * free the item.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000580STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581xfs_buf_item_unlock(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000582 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000584 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
585 struct xfs_buf *bp = bip->bli_buf;
586 int aborted;
587 uint hold;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
Dave Chinner64fc35d2010-05-07 11:04:34 +1000589 /* Clear the buffer's association with this transaction. */
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200590 bp->b_transp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
592 /*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000593 * If this is a transaction abort, don't return early. Instead, allow
594 * the brelse to happen. Normally it would be done for stale
595 * (cancelled) buffers at unpin time, but we'll never go through the
596 * pin/unpin cycle if we abort inside commit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 */
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000598 aborted = (lip->li_flags & XFS_LI_ABORTED) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599
600 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 * Before possibly freeing the buf item, determine if we should
602 * release the buffer at the end of this routine.
603 */
604 hold = bip->bli_flags & XFS_BLI_HOLD;
Dave Chinner64fc35d2010-05-07 11:04:34 +1000605
606 /* Clear the per transaction state. */
607 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD);
608
609 /*
610 * If the buf item is marked stale, then don't do anything. We'll
611 * unlock the buffer and free the buf item when the buffer is unpinned
612 * for the last time.
613 */
614 if (bip->bli_flags & XFS_BLI_STALE) {
615 trace_xfs_buf_item_unlock_stale(bip);
Dave Chinnerc1155412010-05-07 11:05:19 +1000616 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Dave Chinner64fc35d2010-05-07 11:04:34 +1000617 if (!aborted) {
618 atomic_dec(&bip->bli_refcount);
619 return;
620 }
621 }
622
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000623 trace_xfs_buf_item_unlock(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
625 /*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000626 * If the buf item isn't tracking any data, free it, otherwise drop the
627 * reference we hold to it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 */
Eric Sandeen24ad33f2007-06-28 16:43:30 +1000629 if (xfs_bitmap_empty(bip->bli_format.blf_data_map,
Dave Chinner64fc35d2010-05-07 11:04:34 +1000630 bip->bli_format.blf_map_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 xfs_buf_item_relse(bp);
Dave Chinner64fc35d2010-05-07 11:04:34 +1000632 else
633 atomic_dec(&bip->bli_refcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634
Dave Chinner64fc35d2010-05-07 11:04:34 +1000635 if (!hold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637}
638
639/*
640 * This is called to find out where the oldest active copy of the
641 * buf log item in the on disk log resides now that the last log
642 * write of it completed at the given lsn.
643 * We always re-log all the dirty data in a buffer, so usually the
644 * latest copy in the on disk log is the only one that matters. For
645 * those cases we simply return the given lsn.
646 *
647 * The one exception to this is for buffers full of newly allocated
648 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
649 * flag set, indicating that only the di_next_unlinked fields from the
650 * inodes in the buffers will be replayed during recovery. If the
651 * original newly allocated inode images have not yet been flushed
652 * when the buffer is so relogged, then we need to make sure that we
653 * keep the old images in the 'active' portion of the log. We do this
654 * by returning the original lsn of that transaction here rather than
655 * the current one.
656 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000657STATIC xfs_lsn_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658xfs_buf_item_committed(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000659 struct xfs_log_item *lip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 xfs_lsn_t lsn)
661{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000662 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
663
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000664 trace_xfs_buf_item_committed(bip);
665
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000666 if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0)
667 return lip->li_lsn;
668 return lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669}
670
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000671STATIC void
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000672xfs_buf_item_committing(
673 struct xfs_log_item *lip,
674 xfs_lsn_t commit_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675{
676}
677
678/*
679 * This is the ops vector shared by all buf log items.
680 */
Christoph Hellwig272e42b2011-10-28 09:54:24 +0000681static const struct xfs_item_ops xfs_buf_item_ops = {
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000682 .iop_size = xfs_buf_item_size,
683 .iop_format = xfs_buf_item_format,
684 .iop_pin = xfs_buf_item_pin,
685 .iop_unpin = xfs_buf_item_unpin,
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000686 .iop_unlock = xfs_buf_item_unlock,
687 .iop_committed = xfs_buf_item_committed,
688 .iop_push = xfs_buf_item_push,
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000689 .iop_committing = xfs_buf_item_committing
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690};
691
Dave Chinner372cc85e2012-06-22 18:50:12 +1000692STATIC int
693xfs_buf_item_get_format(
694 struct xfs_buf_log_item *bip,
695 int count)
696{
697 ASSERT(bip->bli_formats == NULL);
698 bip->bli_format_count = count;
699
700 if (count == 1) {
701 bip->bli_formats = &bip->bli_format;
702 return 0;
703 }
704
705 bip->bli_formats = kmem_zalloc(count * sizeof(struct xfs_buf_log_format),
706 KM_SLEEP);
707 if (!bip->bli_formats)
708 return ENOMEM;
709 return 0;
710}
711
712STATIC void
713xfs_buf_item_free_format(
714 struct xfs_buf_log_item *bip)
715{
716 if (bip->bli_formats != &bip->bli_format) {
717 kmem_free(bip->bli_formats);
718 bip->bli_formats = NULL;
719 }
720}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
722/*
723 * Allocate a new buf log item to go with the given buffer.
724 * Set the buffer's b_fsprivate field to point to the new
725 * buf log item. If there are other item's attached to the
726 * buffer (see xfs_buf_attach_iodone() below), then put the
727 * buf log item at the front.
728 */
729void
730xfs_buf_item_init(
731 xfs_buf_t *bp,
732 xfs_mount_t *mp)
733{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200734 xfs_log_item_t *lip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 xfs_buf_log_item_t *bip;
736 int chunks;
737 int map_size;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000738 int error;
739 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740
741 /*
742 * Check to see if there is already a buf log item for
743 * this buffer. If there is, it is guaranteed to be
744 * the first. If we do already have one, there is
745 * nothing to do here so return.
746 */
Dave Chinnerebad8612010-09-22 10:47:20 +1000747 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200748 if (lip != NULL && lip->li_type == XFS_LI_BUF)
749 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
Dave Chinner372cc85e2012-06-22 18:50:12 +1000751 bip = kmem_zone_zalloc(xfs_buf_item_zone, KM_SLEEP);
Dave Chinner43f5efc2010-03-23 10:10:00 +1100752 xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 bip->bli_buf = bp;
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000754 xfs_buf_hold(bp);
Dave Chinner372cc85e2012-06-22 18:50:12 +1000755
756 /*
757 * chunks is the number of XFS_BLF_CHUNK size pieces the buffer
758 * can be divided into. Make sure not to truncate any pieces.
759 * map_size is the size of the bitmap needed to describe the
760 * chunks of the buffer.
761 *
762 * Discontiguous buffer support follows the layout of the underlying
763 * buffer. This makes the implementation as simple as possible.
764 */
765 error = xfs_buf_item_get_format(bip, bp->b_map_count);
766 ASSERT(error == 0);
767
768 for (i = 0; i < bip->bli_format_count; i++) {
769 chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len),
770 XFS_BLF_CHUNK);
771 map_size = DIV_ROUND_UP(chunks, NBWORD);
772
773 bip->bli_formats[i].blf_type = XFS_LI_BUF;
774 bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn;
775 bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len;
776 bip->bli_formats[i].blf_map_size = map_size;
777 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778
779#ifdef XFS_TRANS_DEBUG
780 /*
781 * Allocate the arrays for tracking what needs to be logged
782 * and what our callers request to be logged. bli_orig
783 * holds a copy of the original, clean buffer for comparison
784 * against, and bli_logged keeps a 1 bit flag per byte in
785 * the buffer to indicate which bytes the callers have asked
786 * to have logged.
787 */
Dave Chinneraa0e8832012-04-23 15:58:52 +1000788 bip->bli_orig = kmem_alloc(BBTOB(bp->b_length), KM_SLEEP);
789 memcpy(bip->bli_orig, bp->b_addr, BBTOB(bp->b_length));
790 bip->bli_logged = kmem_zalloc(BBTOB(bp->b_length) / NBBY, KM_SLEEP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791#endif
792
793 /*
794 * Put the buf item into the list of items attached to the
795 * buffer at the front.
796 */
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200797 if (bp->b_fspriv)
798 bip->bli_item.li_bio_list = bp->b_fspriv;
799 bp->b_fspriv = bip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800}
801
802
803/*
804 * Mark bytes first through last inclusive as dirty in the buf
805 * item's bitmap.
806 */
807void
Dave Chinner372cc85e2012-06-22 18:50:12 +1000808xfs_buf_item_log_segment(
809 struct xfs_buf_log_item *bip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 uint first,
Dave Chinner372cc85e2012-06-22 18:50:12 +1000811 uint last,
812 uint *map)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813{
814 uint first_bit;
815 uint last_bit;
816 uint bits_to_set;
817 uint bits_set;
818 uint word_num;
819 uint *wordp;
820 uint bit;
821 uint end_bit;
822 uint mask;
823
824 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 * Convert byte offsets to bit numbers.
826 */
Dave Chinnerc1155412010-05-07 11:05:19 +1000827 first_bit = first >> XFS_BLF_SHIFT;
828 last_bit = last >> XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829
830 /*
831 * Calculate the total number of bits to be set.
832 */
833 bits_to_set = last_bit - first_bit + 1;
834
835 /*
836 * Get a pointer to the first word in the bitmap
837 * to set a bit in.
838 */
839 word_num = first_bit >> BIT_TO_WORD_SHIFT;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000840 wordp = &map[word_num];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841
842 /*
843 * Calculate the starting bit in the first word.
844 */
845 bit = first_bit & (uint)(NBWORD - 1);
846
847 /*
848 * First set any bits in the first word of our range.
849 * If it starts at bit 0 of the word, it will be
850 * set below rather than here. That is what the variable
851 * bit tells us. The variable bits_set tracks the number
852 * of bits that have been set so far. End_bit is the number
853 * of the last bit to be set in this word plus one.
854 */
855 if (bit) {
856 end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
857 mask = ((1 << (end_bit - bit)) - 1) << bit;
858 *wordp |= mask;
859 wordp++;
860 bits_set = end_bit - bit;
861 } else {
862 bits_set = 0;
863 }
864
865 /*
866 * Now set bits a whole word at a time that are between
867 * first_bit and last_bit.
868 */
869 while ((bits_to_set - bits_set) >= NBWORD) {
870 *wordp |= 0xffffffff;
871 bits_set += NBWORD;
872 wordp++;
873 }
874
875 /*
876 * Finally, set any bits left to be set in one last partial word.
877 */
878 end_bit = bits_to_set - bits_set;
879 if (end_bit) {
880 mask = (1 << end_bit) - 1;
881 *wordp |= mask;
882 }
883
884 xfs_buf_item_log_debug(bip, first, last);
885}
886
Dave Chinner372cc85e2012-06-22 18:50:12 +1000887/*
888 * Mark bytes first through last inclusive as dirty in the buf
889 * item's bitmap.
890 */
891void
892xfs_buf_item_log(
893 xfs_buf_log_item_t *bip,
894 uint first,
895 uint last)
896{
897 int i;
898 uint start;
899 uint end;
900 struct xfs_buf *bp = bip->bli_buf;
901
902 /*
903 * Mark the item as having some dirty data for
904 * quick reference in xfs_buf_item_dirty.
905 */
906 bip->bli_flags |= XFS_BLI_DIRTY;
907
908 /*
909 * walk each buffer segment and mark them dirty appropriately.
910 */
911 start = 0;
912 for (i = 0; i < bip->bli_format_count; i++) {
913 if (start > last)
914 break;
915 end = start + BBTOB(bp->b_maps[i].bm_len);
916 if (first > end) {
917 start += BBTOB(bp->b_maps[i].bm_len);
918 continue;
919 }
920 if (first < start)
921 first = start;
922 if (end > last)
923 end = last;
924
925 xfs_buf_item_log_segment(bip, first, end,
926 &bip->bli_formats[i].blf_data_map[0]);
927
928 start += bp->b_maps[i].bm_len;
929 }
930}
931
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932
933/*
934 * Return 1 if the buffer has some data that has been logged (at any
935 * point, not just the current transaction) and 0 if not.
936 */
937uint
938xfs_buf_item_dirty(
939 xfs_buf_log_item_t *bip)
940{
941 return (bip->bli_flags & XFS_BLI_DIRTY);
942}
943
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000944STATIC void
945xfs_buf_item_free(
946 xfs_buf_log_item_t *bip)
947{
948#ifdef XFS_TRANS_DEBUG
949 kmem_free(bip->bli_orig);
950 kmem_free(bip->bli_logged);
951#endif /* XFS_TRANS_DEBUG */
952
Dave Chinner372cc85e2012-06-22 18:50:12 +1000953 xfs_buf_item_free_format(bip);
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000954 kmem_zone_free(xfs_buf_item_zone, bip);
955}
956
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957/*
958 * This is called when the buf log item is no longer needed. It should
959 * free the buf log item associated with the given buffer and clear
960 * the buffer's pointer to the buf log item. If there are no more
961 * items in the list, clear the b_iodone field of the buffer (see
962 * xfs_buf_attach_iodone() below).
963 */
964void
965xfs_buf_item_relse(
966 xfs_buf_t *bp)
967{
968 xfs_buf_log_item_t *bip;
969
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000970 trace_xfs_buf_item_relse(bp, _RET_IP_);
971
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200972 bip = bp->b_fspriv;
973 bp->b_fspriv = bip->bli_item.li_bio_list;
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200974 if (bp->b_fspriv == NULL)
975 bp->b_iodone = NULL;
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200976
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000977 xfs_buf_rele(bp);
978 xfs_buf_item_free(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979}
980
981
982/*
983 * Add the given log item with its callback to the list of callbacks
984 * to be called when the buffer's I/O completes. If it is not set
985 * already, set the buffer's b_iodone() routine to be
986 * xfs_buf_iodone_callbacks() and link the log item into the list of
987 * items rooted at b_fsprivate. Items are always added as the second
988 * entry in the list if there is a first, because the buf item code
989 * assumes that the buf log item is first.
990 */
991void
992xfs_buf_attach_iodone(
993 xfs_buf_t *bp,
994 void (*cb)(xfs_buf_t *, xfs_log_item_t *),
995 xfs_log_item_t *lip)
996{
997 xfs_log_item_t *head_lip;
998
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200999 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
1001 lip->li_cb = cb;
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001002 head_lip = bp->b_fspriv;
1003 if (head_lip) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 lip->li_bio_list = head_lip->li_bio_list;
1005 head_lip->li_bio_list = lip;
1006 } else {
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001007 bp->b_fspriv = lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 }
1009
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02001010 ASSERT(bp->b_iodone == NULL ||
1011 bp->b_iodone == xfs_buf_iodone_callbacks);
1012 bp->b_iodone = xfs_buf_iodone_callbacks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013}
1014
Dave Chinnerc90821a2010-12-03 17:00:52 +11001015/*
1016 * We can have many callbacks on a buffer. Running the callbacks individually
1017 * can cause a lot of contention on the AIL lock, so we allow for a single
1018 * callback to be able to scan the remaining lip->li_bio_list for other items
1019 * of the same type and callback to be processed in the first call.
1020 *
1021 * As a result, the loop walking the callback list below will also modify the
1022 * list. it removes the first item from the list and then runs the callback.
1023 * The loop then restarts from the new head of the list. This allows the
1024 * callback to scan and modify the list attached to the buffer and we don't
1025 * have to care about maintaining a next item pointer.
1026 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027STATIC void
1028xfs_buf_do_callbacks(
Dave Chinnerc90821a2010-12-03 17:00:52 +11001029 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030{
Dave Chinnerc90821a2010-12-03 17:00:52 +11001031 struct xfs_log_item *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001033 while ((lip = bp->b_fspriv) != NULL) {
1034 bp->b_fspriv = lip->li_bio_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 ASSERT(lip->li_cb != NULL);
1036 /*
1037 * Clear the next pointer so we don't have any
1038 * confusion if the item is added to another buf.
1039 * Don't touch the log item after calling its
1040 * callback, because it could have freed itself.
1041 */
1042 lip->li_bio_list = NULL;
1043 lip->li_cb(bp, lip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 }
1045}
1046
1047/*
1048 * This is the iodone() function for buffers which have had callbacks
1049 * attached to them by xfs_buf_attach_iodone(). It should remove each
1050 * log item from the buffer's list and call the callback of each in turn.
1051 * When done, the buffer's fsprivate field is set to NULL and the buffer
1052 * is unlocked with a call to iodone().
1053 */
1054void
1055xfs_buf_iodone_callbacks(
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001056 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057{
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001058 struct xfs_log_item *lip = bp->b_fspriv;
1059 struct xfs_mount *mp = lip->li_mountp;
1060 static ulong lasttime;
1061 static xfs_buftarg_t *lasttarg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00001063 if (likely(!xfs_buf_geterror(bp)))
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001064 goto do_callbacks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001066 /*
1067 * If we've already decided to shutdown the filesystem because of
1068 * I/O errors, there's no point in giving this a retry.
1069 */
1070 if (XFS_FORCED_SHUTDOWN(mp)) {
Christoph Hellwigc867cb62011-10-10 16:52:46 +00001071 xfs_buf_stale(bp);
Christoph Hellwigc867cb62011-10-10 16:52:46 +00001072 XFS_BUF_DONE(bp);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001073 trace_xfs_buf_item_iodone(bp, _RET_IP_);
1074 goto do_callbacks;
1075 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076
Chandra Seetharaman49074c02011-07-22 23:40:40 +00001077 if (bp->b_target != lasttarg ||
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001078 time_after(jiffies, (lasttime + 5*HZ))) {
1079 lasttime = jiffies;
Christoph Hellwigb38505b2011-10-10 16:52:50 +00001080 xfs_buf_ioerror_alert(bp, __func__);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001081 }
Chandra Seetharaman49074c02011-07-22 23:40:40 +00001082 lasttarg = bp->b_target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001084 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001085 * If the write was asynchronous then no one will be looking for the
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001086 * error. Clear the error state and write the buffer out again.
1087 *
Christoph Hellwig43ff2122012-04-23 15:58:39 +10001088 * XXX: This helps against transient write errors, but we need to find
1089 * a way to shut the filesystem down if the writes keep failing.
1090 *
1091 * In practice we'll shut the filesystem down soon as non-transient
1092 * erorrs tend to affect the whole device and a failing log write
1093 * will make us give up. But we really ought to do better here.
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001094 */
1095 if (XFS_BUF_ISASYNC(bp)) {
Christoph Hellwig43ff2122012-04-23 15:58:39 +10001096 ASSERT(bp->b_iodone != NULL);
1097
1098 trace_xfs_buf_item_iodone_async(bp, _RET_IP_);
1099
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00001100 xfs_buf_ioerror(bp, 0); /* errno of 0 unsets the flag */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001102 if (!XFS_BUF_ISSTALE(bp)) {
Christoph Hellwig43ff2122012-04-23 15:58:39 +10001103 bp->b_flags |= XBF_WRITE | XBF_ASYNC | XBF_DONE;
Christoph Hellwiga2dcf5d2012-07-13 02:24:10 -04001104 xfs_buf_iorequest(bp);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10001105 } else {
1106 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 }
Christoph Hellwig43ff2122012-04-23 15:58:39 +10001108
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 return;
1110 }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001111
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001112 /*
1113 * If the write of the buffer was synchronous, we want to make
1114 * sure to return the error to the caller of xfs_bwrite().
1115 */
Christoph Hellwigc867cb62011-10-10 16:52:46 +00001116 xfs_buf_stale(bp);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001117 XFS_BUF_DONE(bp);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001118
1119 trace_xfs_buf_error_relse(bp, _RET_IP_);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001120
1121do_callbacks:
Dave Chinnerc90821a2010-12-03 17:00:52 +11001122 xfs_buf_do_callbacks(bp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001123 bp->b_fspriv = NULL;
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02001124 bp->b_iodone = NULL;
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00001125 xfs_buf_ioend(bp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126}
1127
1128/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 * This is the iodone() function for buffers which have been
1130 * logged. It is called when they are eventually flushed out.
1131 * It should remove the buf item from the AIL, and free the buf item.
1132 * It is called by xfs_buf_iodone_callbacks() above which will take
1133 * care of cleaning up the buffer itself.
1134 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135void
1136xfs_buf_iodone(
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001137 struct xfs_buf *bp,
1138 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139{
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001140 struct xfs_ail *ailp = lip->li_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001142 ASSERT(BUF_ITEM(lip)->bli_buf == bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +10001144 xfs_buf_rele(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
1146 /*
1147 * If we are forcibly shutting down, this may well be
1148 * off the AIL already. That's because we simulate the
1149 * log-committed callbacks to unpin these buffers. Or we may never
1150 * have put this item on AIL because of the transaction was
David Chinner783a2f62008-10-30 17:39:58 +11001151 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 *
1153 * Either way, AIL is useless if we're forcing a shutdown.
1154 */
David Chinnerfc1829f2008-10-30 17:39:46 +11001155 spin_lock(&ailp->xa_lock);
Dave Chinner04913fd2012-04-23 15:58:41 +10001156 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001157 xfs_buf_item_free(BUF_ITEM(lip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158}