blob: f1d85cfc0a54d1cb54ebe95937872127f2e63142 [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
Dave Chinnerc90821a2010-12-03 17:00:52 +110040STATIC void xfs_buf_do_callbacks(struct xfs_buf *bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Dave Chinner166d1362013-08-12 20:50:04 +100042static inline int
43xfs_buf_log_format_size(
44 struct xfs_buf_log_format *blfp)
45{
46 return offsetof(struct xfs_buf_log_format, blf_data_map) +
47 (blfp->blf_map_size * sizeof(blfp->blf_data_map[0]));
48}
49
Linus Torvalds1da177e2005-04-16 15:20:36 -070050/*
51 * This returns the number of log iovecs needed to log the
52 * given buf log item.
53 *
54 * It calculates this as 1 iovec for the buf log format structure
55 * and 1 for each stretch of non-contiguous chunks to be logged.
56 * Contiguous chunks are logged in a single iovec.
57 *
58 * If the XFS_BLI_STALE flag has been set, then log nothing.
59 */
Dave Chinner166d1362013-08-12 20:50:04 +100060STATIC void
Dave Chinner372cc85e2012-06-22 18:50:12 +100061xfs_buf_item_size_segment(
62 struct xfs_buf_log_item *bip,
Dave Chinner166d1362013-08-12 20:50:04 +100063 struct xfs_buf_log_format *blfp,
64 int *nvecs,
65 int *nbytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -070066{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +100067 struct xfs_buf *bp = bip->bli_buf;
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +100068 int next_bit;
69 int last_bit;
Linus Torvalds1da177e2005-04-16 15:20:36 -070070
Dave Chinner372cc85e2012-06-22 18:50:12 +100071 last_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
72 if (last_bit == -1)
Dave Chinner166d1362013-08-12 20:50:04 +100073 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
Dave Chinner372cc85e2012-06-22 18:50:12 +100075 /*
76 * initial count for a dirty buffer is 2 vectors - the format structure
77 * and the first dirty region.
78 */
Dave Chinner166d1362013-08-12 20:50:04 +100079 *nvecs += 2;
80 *nbytes += xfs_buf_log_format_size(blfp) + XFS_BLF_CHUNK;
Dave Chinner372cc85e2012-06-22 18:50:12 +100081
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 while (last_bit != -1) {
83 /*
84 * This takes the bit number to start looking from and
85 * returns the next set bit from there. It returns -1
86 * if there are no more bits set or the start bit is
87 * beyond the end of the bitmap.
88 */
Dave Chinner372cc85e2012-06-22 18:50:12 +100089 next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
90 last_bit + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 /*
92 * If we run out of bits, leave the loop,
93 * else if we find a new set of bits bump the number of vecs,
94 * else keep scanning the current set of bits.
95 */
96 if (next_bit == -1) {
Dave Chinner372cc85e2012-06-22 18:50:12 +100097 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 } else if (next_bit != last_bit + 1) {
99 last_bit = next_bit;
Dave Chinner166d1362013-08-12 20:50:04 +1000100 (*nvecs)++;
Dave Chinnerc1155412010-05-07 11:05:19 +1000101 } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) !=
102 (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) +
103 XFS_BLF_CHUNK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 last_bit = next_bit;
Dave Chinner166d1362013-08-12 20:50:04 +1000105 (*nvecs)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 } else {
107 last_bit++;
108 }
Dave Chinner166d1362013-08-12 20:50:04 +1000109 *nbytes += XFS_BLF_CHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 }
Dave Chinner372cc85e2012-06-22 18:50:12 +1000111}
112
113/*
114 * This returns the number of log iovecs needed to log the given buf log item.
115 *
116 * It calculates this as 1 iovec for the buf log format structure and 1 for each
117 * stretch of non-contiguous chunks to be logged. Contiguous chunks are logged
118 * in a single iovec.
119 *
120 * Discontiguous buffers need a format structure per region that that is being
121 * logged. This makes the changes in the buffer appear to log recovery as though
122 * they came from separate buffers, just like would occur if multiple buffers
123 * were used instead of a single discontiguous buffer. This enables
124 * discontiguous buffers to be in-memory constructs, completely transparent to
125 * what ends up on disk.
126 *
127 * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log
128 * format structures.
129 */
Dave Chinner166d1362013-08-12 20:50:04 +1000130STATIC void
Dave Chinner372cc85e2012-06-22 18:50:12 +1000131xfs_buf_item_size(
Dave Chinner166d1362013-08-12 20:50:04 +1000132 struct xfs_log_item *lip,
133 int *nvecs,
134 int *nbytes)
Dave Chinner372cc85e2012-06-22 18:50:12 +1000135{
136 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
Dave Chinner372cc85e2012-06-22 18:50:12 +1000137 int i;
138
139 ASSERT(atomic_read(&bip->bli_refcount) > 0);
140 if (bip->bli_flags & XFS_BLI_STALE) {
141 /*
142 * The buffer is stale, so all we need to log
143 * is the buf log format structure with the
144 * cancel flag in it.
145 */
146 trace_xfs_buf_item_size_stale(bip);
Mark Tinguelyb9438172012-12-04 17:18:03 -0600147 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
Dave Chinner166d1362013-08-12 20:50:04 +1000148 *nvecs += bip->bli_format_count;
149 for (i = 0; i < bip->bli_format_count; i++) {
150 *nbytes += xfs_buf_log_format_size(&bip->bli_formats[i]);
151 }
152 return;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000153 }
154
155 ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
156
Dave Chinner5f6bed72013-06-27 16:04:52 +1000157 if (bip->bli_flags & XFS_BLI_ORDERED) {
158 /*
159 * The buffer has been logged just to order it.
160 * It is not being included in the transaction
161 * commit, so no vectors are used at all.
162 */
163 trace_xfs_buf_item_size_ordered(bip);
Dave Chinner166d1362013-08-12 20:50:04 +1000164 *nvecs = XFS_LOG_VEC_ORDERED;
165 return;
Dave Chinner5f6bed72013-06-27 16:04:52 +1000166 }
167
Dave Chinner372cc85e2012-06-22 18:50:12 +1000168 /*
169 * the vector count is based on the number of buffer vectors we have
170 * dirty bits in. This will only be greater than one when we have a
171 * compound buffer with more than one segment dirty. Hence for compound
172 * buffers we need to track which segment the dirty bits correspond to,
173 * and when we move from one segment to the next increment the vector
174 * count for the extra buf log format structure that will need to be
175 * written.
176 */
Dave Chinner372cc85e2012-06-22 18:50:12 +1000177 for (i = 0; i < bip->bli_format_count; i++) {
Dave Chinner166d1362013-08-12 20:50:04 +1000178 xfs_buf_item_size_segment(bip, &bip->bli_formats[i],
179 nvecs, nbytes);
Dave Chinner372cc85e2012-06-22 18:50:12 +1000180 }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000181 trace_xfs_buf_item_size(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182}
183
Dave Chinner372cc85e2012-06-22 18:50:12 +1000184static struct xfs_log_iovec *
185xfs_buf_item_format_segment(
186 struct xfs_buf_log_item *bip,
187 struct xfs_log_iovec *vecp,
188 uint offset,
189 struct xfs_buf_log_format *blfp)
190{
191 struct xfs_buf *bp = bip->bli_buf;
192 uint base_size;
193 uint nvecs;
194 int first_bit;
195 int last_bit;
196 int next_bit;
197 uint nbits;
198 uint buffer_offset;
199
200 /* copy the flags across from the base format item */
Mark Tinguelyb9438172012-12-04 17:18:03 -0600201 blfp->blf_flags = bip->__bli_format.blf_flags;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000202
203 /*
204 * Base size is the actual size of the ondisk structure - it reflects
205 * the actual size of the dirty bitmap rather than the size of the in
206 * memory structure.
207 */
Dave Chinner166d1362013-08-12 20:50:04 +1000208 base_size = xfs_buf_log_format_size(blfp);
Mark Tinguely820a5542012-12-04 17:18:04 -0600209
210 nvecs = 0;
211 first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
212 if (!(bip->bli_flags & XFS_BLI_STALE) && first_bit == -1) {
213 /*
214 * If the map is not be dirty in the transaction, mark
215 * the size as zero and do not advance the vector pointer.
216 */
217 goto out;
218 }
219
Dave Chinner372cc85e2012-06-22 18:50:12 +1000220 vecp->i_addr = blfp;
221 vecp->i_len = base_size;
222 vecp->i_type = XLOG_REG_TYPE_BFORMAT;
223 vecp++;
224 nvecs = 1;
225
226 if (bip->bli_flags & XFS_BLI_STALE) {
227 /*
228 * The buffer is stale, so all we need to log
229 * is the buf log format structure with the
230 * cancel flag in it.
231 */
232 trace_xfs_buf_item_format_stale(bip);
233 ASSERT(blfp->blf_flags & XFS_BLF_CANCEL);
Mark Tinguely820a5542012-12-04 17:18:04 -0600234 goto out;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000235 }
236
Dave Chinner5f6bed72013-06-27 16:04:52 +1000237
Dave Chinner372cc85e2012-06-22 18:50:12 +1000238 /*
239 * Fill in an iovec for each set of contiguous chunks.
240 */
Mark Tinguely820a5542012-12-04 17:18:04 -0600241
Dave Chinner372cc85e2012-06-22 18:50:12 +1000242 last_bit = first_bit;
243 nbits = 1;
244 for (;;) {
245 /*
246 * This takes the bit number to start looking from and
247 * returns the next set bit from there. It returns -1
248 * if there are no more bits set or the start bit is
249 * beyond the end of the bitmap.
250 */
251 next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
252 (uint)last_bit + 1);
253 /*
254 * If we run out of bits fill in the last iovec and get
255 * out of the loop.
256 * Else if we start a new set of bits then fill in the
257 * iovec for the series we were looking at and start
258 * counting the bits in the new one.
259 * Else we're still in the same set of bits so just
260 * keep counting and scanning.
261 */
262 if (next_bit == -1) {
263 buffer_offset = offset + first_bit * XFS_BLF_CHUNK;
264 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
265 vecp->i_len = nbits * XFS_BLF_CHUNK;
266 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
267 nvecs++;
268 break;
269 } else if (next_bit != last_bit + 1) {
270 buffer_offset = offset + first_bit * XFS_BLF_CHUNK;
271 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
272 vecp->i_len = nbits * XFS_BLF_CHUNK;
273 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
274 nvecs++;
275 vecp++;
276 first_bit = next_bit;
277 last_bit = next_bit;
278 nbits = 1;
279 } else if (xfs_buf_offset(bp, offset +
280 (next_bit << XFS_BLF_SHIFT)) !=
281 (xfs_buf_offset(bp, offset +
282 (last_bit << XFS_BLF_SHIFT)) +
283 XFS_BLF_CHUNK)) {
284 buffer_offset = offset + first_bit * XFS_BLF_CHUNK;
285 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
286 vecp->i_len = nbits * XFS_BLF_CHUNK;
287 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
Dave Chinner709da6a2013-05-27 16:38:23 +1000288 nvecs++;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000289 vecp++;
290 first_bit = next_bit;
291 last_bit = next_bit;
292 nbits = 1;
293 } else {
294 last_bit++;
295 nbits++;
296 }
297 }
Mark Tinguely820a5542012-12-04 17:18:04 -0600298out:
299 blfp->blf_size = nvecs;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000300 return vecp;
301}
302
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303/*
304 * This is called to fill in the vector of log iovecs for the
305 * given log buf item. It fills the first entry with a buf log
306 * format structure, and the rest point to contiguous chunks
307 * within the buffer.
308 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000309STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310xfs_buf_item_format(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000311 struct xfs_log_item *lip,
312 struct xfs_log_iovec *vecp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000314 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
Dave Chinner372cc85e2012-06-22 18:50:12 +1000315 struct xfs_buf *bp = bip->bli_buf;
316 uint offset = 0;
317 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
319 ASSERT(atomic_read(&bip->bli_refcount) > 0);
320 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
321 (bip->bli_flags & XFS_BLI_STALE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
323 /*
Dave Chinnerccf7c232010-05-20 23:19:42 +1000324 * If it is an inode buffer, transfer the in-memory state to the
Dave Chinnerddf6ad02013-06-27 16:04:56 +1000325 * format flags and clear the in-memory state.
326 *
327 * For buffer based inode allocation, we do not transfer
Dave Chinnerccf7c232010-05-20 23:19:42 +1000328 * this state if the inode buffer allocation has not yet been committed
329 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
330 * correct replay of the inode allocation.
Dave Chinnerddf6ad02013-06-27 16:04:56 +1000331 *
332 * For icreate item based inode allocation, the buffers aren't written
333 * to the journal during allocation, and hence we should always tag the
334 * buffer as an inode buffer so that the correct unlinked list replay
335 * occurs during recovery.
Dave Chinnerccf7c232010-05-20 23:19:42 +1000336 */
337 if (bip->bli_flags & XFS_BLI_INODE_BUF) {
Dave Chinnerddf6ad02013-06-27 16:04:56 +1000338 if (xfs_sb_version_hascrc(&lip->li_mountp->m_sb) ||
339 !((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000340 xfs_log_item_in_current_chkpt(lip)))
Mark Tinguelyb9438172012-12-04 17:18:03 -0600341 bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF;
Dave Chinnerccf7c232010-05-20 23:19:42 +1000342 bip->bli_flags &= ~XFS_BLI_INODE_BUF;
343 }
344
Dave Chinner5f6bed72013-06-27 16:04:52 +1000345 if ((bip->bli_flags & (XFS_BLI_ORDERED|XFS_BLI_STALE)) ==
346 XFS_BLI_ORDERED) {
347 /*
348 * The buffer has been logged just to order it. It is not being
349 * included in the transaction commit, so don't format it.
350 */
351 trace_xfs_buf_item_format_ordered(bip);
352 return;
353 }
354
Dave Chinner372cc85e2012-06-22 18:50:12 +1000355 for (i = 0; i < bip->bli_format_count; i++) {
356 vecp = xfs_buf_item_format_segment(bip, vecp, offset,
357 &bip->bli_formats[i]);
358 offset += bp->b_maps[i].bm_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 }
360
361 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 * Check to make sure everything is consistent.
363 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000364 trace_xfs_buf_item_format(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365}
366
367/*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000368 * This is called to pin the buffer associated with the buf log item in memory
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000369 * so it cannot be written out.
Dave Chinner64fc35d2010-05-07 11:04:34 +1000370 *
371 * We also always take a reference to the buffer log item here so that the bli
372 * is held while the item is pinned in memory. This means that we can
373 * unconditionally drop the reference count a transaction holds when the
374 * transaction is completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000376STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377xfs_buf_item_pin(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000378 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000380 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 ASSERT(atomic_read(&bip->bli_refcount) > 0);
383 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
Dave Chinner5f6bed72013-06-27 16:04:52 +1000384 (bip->bli_flags & XFS_BLI_ORDERED) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 (bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000386
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000387 trace_xfs_buf_item_pin(bip);
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000388
389 atomic_inc(&bip->bli_refcount);
390 atomic_inc(&bip->bli_buf->b_pin_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391}
392
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393/*
394 * This is called to unpin the buffer associated with the buf log
395 * item which was previously pinned with a call to xfs_buf_item_pin().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 *
397 * Also drop the reference to the buf item for the current transaction.
398 * If the XFS_BLI_STALE flag is set and we are the last reference,
399 * then free up the buf log item and unlock the buffer.
Christoph Hellwig9412e312010-06-23 18:11:15 +1000400 *
401 * If the remove flag is set we are called from uncommit in the
402 * forced-shutdown path. If that is true and the reference count on
403 * the log item is going to drop to zero we need to free the item's
404 * descriptor in the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000406STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407xfs_buf_item_unpin(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000408 struct xfs_log_item *lip,
Christoph Hellwig9412e312010-06-23 18:11:15 +1000409 int remove)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000411 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000412 xfs_buf_t *bp = bip->bli_buf;
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000413 struct xfs_ail *ailp = lip->li_ailp;
Dave Chinner8e123852010-03-08 11:26:03 +1100414 int stale = bip->bli_flags & XFS_BLI_STALE;
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000415 int freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200417 ASSERT(bp->b_fspriv == bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 ASSERT(atomic_read(&bip->bli_refcount) > 0);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000419
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000420 trace_xfs_buf_item_unpin(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
422 freed = atomic_dec_and_test(&bip->bli_refcount);
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000423
424 if (atomic_dec_and_test(&bp->b_pin_count))
425 wake_up_all(&bp->b_waiters);
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000426
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 if (freed && stale) {
428 ASSERT(bip->bli_flags & XFS_BLI_STALE);
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200429 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 ASSERT(XFS_BUF_ISSTALE(bp));
Mark Tinguelyb9438172012-12-04 17:18:03 -0600431 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000432
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000433 trace_xfs_buf_item_unpin_stale(bip);
434
Christoph Hellwig9412e312010-06-23 18:11:15 +1000435 if (remove) {
436 /*
Dave Chinnere34a3142011-01-27 12:13:35 +1100437 * If we are in a transaction context, we have to
438 * remove the log item from the transaction as we are
439 * about to release our reference to the buffer. If we
440 * don't, the unlock that occurs later in
441 * xfs_trans_uncommit() will try to reference the
Christoph Hellwig9412e312010-06-23 18:11:15 +1000442 * buffer which we no longer have a hold on.
443 */
Dave Chinnere34a3142011-01-27 12:13:35 +1100444 if (lip->li_desc)
445 xfs_trans_del_item(lip);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000446
447 /*
448 * Since the transaction no longer refers to the buffer,
449 * the buffer should no longer refer to the transaction.
450 */
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200451 bp->b_transp = NULL;
Christoph Hellwig9412e312010-06-23 18:11:15 +1000452 }
453
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 /*
455 * If we get called here because of an IO error, we may
David Chinner783a2f62008-10-30 17:39:58 +1100456 * or may not have the item on the AIL. xfs_trans_ail_delete()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 * will take care of that situation.
David Chinner783a2f62008-10-30 17:39:58 +1100458 * xfs_trans_ail_delete() drops the AIL lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 */
460 if (bip->bli_flags & XFS_BLI_STALE_INODE) {
Dave Chinnerc90821a2010-12-03 17:00:52 +1100461 xfs_buf_do_callbacks(bp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200462 bp->b_fspriv = NULL;
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200463 bp->b_iodone = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 } else {
David Chinner783a2f62008-10-30 17:39:58 +1100465 spin_lock(&ailp->xa_lock);
Dave Chinner04913fd2012-04-23 15:58:41 +1000466 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_LOG_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 xfs_buf_item_relse(bp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200468 ASSERT(bp->b_fspriv == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 }
470 xfs_buf_relse(bp);
Christoph Hellwig960c60a2012-04-23 15:58:38 +1000471 } else if (freed && remove) {
Dave Chinner137fff02012-11-02 14:23:12 +1100472 /*
473 * There are currently two references to the buffer - the active
474 * LRU reference and the buf log item. What we are about to do
475 * here - simulate a failed IO completion - requires 3
476 * references.
477 *
478 * The LRU reference is removed by the xfs_buf_stale() call. The
479 * buf item reference is removed by the xfs_buf_iodone()
480 * callback that is run by xfs_buf_do_callbacks() during ioend
481 * processing (via the bp->b_iodone callback), and then finally
482 * the ioend processing will drop the IO reference if the buffer
483 * is marked XBF_ASYNC.
484 *
485 * Hence we need to take an additional reference here so that IO
486 * completion processing doesn't free the buffer prematurely.
487 */
Christoph Hellwig960c60a2012-04-23 15:58:38 +1000488 xfs_buf_lock(bp);
Dave Chinner137fff02012-11-02 14:23:12 +1100489 xfs_buf_hold(bp);
490 bp->b_flags |= XBF_ASYNC;
Christoph Hellwig960c60a2012-04-23 15:58:38 +1000491 xfs_buf_ioerror(bp, EIO);
492 XFS_BUF_UNDONE(bp);
493 xfs_buf_stale(bp);
494 xfs_buf_ioend(bp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 }
496}
497
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000498STATIC uint
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000499xfs_buf_item_push(
500 struct xfs_log_item *lip,
501 struct list_head *buffer_list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000503 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
504 struct xfs_buf *bp = bip->bli_buf;
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000505 uint rval = XFS_ITEM_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
Chandra Seetharaman811e64c2011-07-22 23:40:27 +0000507 if (xfs_buf_ispinned(bp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 return XFS_ITEM_PINNED;
Brian Foster5337fe92013-02-11 10:08:21 -0500509 if (!xfs_buf_trylock(bp)) {
510 /*
511 * If we have just raced with a buffer being pinned and it has
512 * been marked stale, we could end up stalling until someone else
513 * issues a log force to unpin the stale buffer. Check for the
514 * race condition here so xfsaild recognizes the buffer is pinned
515 * and queues a log force to move it along.
516 */
517 if (xfs_buf_ispinned(bp))
518 return XFS_ITEM_PINNED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 return XFS_ITEM_LOCKED;
Brian Foster5337fe92013-02-11 10:08:21 -0500520 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000523
524 trace_xfs_buf_item_push(bip);
525
526 if (!xfs_buf_delwri_queue(bp, buffer_list))
527 rval = XFS_ITEM_FLUSHING;
528 xfs_buf_unlock(bp);
529 return rval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530}
531
532/*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000533 * Release the buffer associated with the buf log item. If there is no dirty
534 * logged data associated with the buffer recorded in the buf log item, then
535 * free the buf log item and remove the reference to it in the buffer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 *
Dave Chinner64fc35d2010-05-07 11:04:34 +1000537 * This call ignores the recursion count. It is only called when the buffer
538 * should REALLY be unlocked, regardless of the recursion count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 *
Dave Chinner64fc35d2010-05-07 11:04:34 +1000540 * We unconditionally drop the transaction's reference to the log item. If the
541 * item was logged, then another reference was taken when it was pinned, so we
542 * can safely drop the transaction reference now. This also allows us to avoid
543 * potential races with the unpin code freeing the bli by not referencing the
544 * bli after we've dropped the reference count.
545 *
546 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
547 * if necessary but do not unlock the buffer. This is for support of
548 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
549 * free the item.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000551STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552xfs_buf_item_unlock(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000553 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000555 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
556 struct xfs_buf *bp = bip->bli_buf;
Dave Chinner5f6bed72013-06-27 16:04:52 +1000557 bool clean;
558 bool aborted;
559 int flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
Dave Chinner64fc35d2010-05-07 11:04:34 +1000561 /* Clear the buffer's association with this transaction. */
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200562 bp->b_transp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
564 /*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000565 * If this is a transaction abort, don't return early. Instead, allow
566 * the brelse to happen. Normally it would be done for stale
567 * (cancelled) buffers at unpin time, but we'll never go through the
568 * pin/unpin cycle if we abort inside commit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 */
Dave Chinner5f6bed72013-06-27 16:04:52 +1000570 aborted = (lip->li_flags & XFS_LI_ABORTED) ? true : false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 /*
Dave Chinner5f6bed72013-06-27 16:04:52 +1000572 * Before possibly freeing the buf item, copy the per-transaction state
573 * so we can reference it safely later after clearing it from the
574 * buffer log item.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 */
Dave Chinner5f6bed72013-06-27 16:04:52 +1000576 flags = bip->bli_flags;
577 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD | XFS_BLI_ORDERED);
Dave Chinner64fc35d2010-05-07 11:04:34 +1000578
579 /*
580 * If the buf item is marked stale, then don't do anything. We'll
581 * unlock the buffer and free the buf item when the buffer is unpinned
582 * for the last time.
583 */
Dave Chinner5f6bed72013-06-27 16:04:52 +1000584 if (flags & XFS_BLI_STALE) {
Dave Chinner64fc35d2010-05-07 11:04:34 +1000585 trace_xfs_buf_item_unlock_stale(bip);
Mark Tinguelyb9438172012-12-04 17:18:03 -0600586 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
Dave Chinner64fc35d2010-05-07 11:04:34 +1000587 if (!aborted) {
588 atomic_dec(&bip->bli_refcount);
589 return;
590 }
591 }
592
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000593 trace_xfs_buf_item_unlock(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
595 /*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000596 * If the buf item isn't tracking any data, free it, otherwise drop the
Dave Chinner3b190342013-01-21 23:53:55 +1100597 * reference we hold to it. If we are aborting the transaction, this may
598 * be the only reference to the buf item, so we free it anyway
599 * regardless of whether it is dirty or not. A dirty abort implies a
600 * shutdown, anyway.
Dave Chinner5f6bed72013-06-27 16:04:52 +1000601 *
602 * Ordered buffers are dirty but may have no recorded changes, so ensure
603 * we only release clean items here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 */
Dave Chinner5f6bed72013-06-27 16:04:52 +1000605 clean = (flags & XFS_BLI_DIRTY) ? false : true;
606 if (clean) {
607 int i;
608 for (i = 0; i < bip->bli_format_count; i++) {
609 if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map,
610 bip->bli_formats[i].blf_map_size)) {
611 clean = false;
612 break;
613 }
Mark Tinguelyc883d0c2012-12-04 17:18:05 -0600614 }
615 }
Dave Chinner46f9d2e2013-09-03 21:47:37 +1000616
617 /*
618 * Clean buffers, by definition, cannot be in the AIL. However, aborted
619 * buffers may be dirty and hence in the AIL. Therefore if we are
620 * aborting a buffer and we've just taken the last refernce away, we
621 * have to check if it is in the AIL before freeing it. We need to free
622 * it in this case, because an aborted transaction has already shut the
623 * filesystem down and this is the last chance we will have to do so.
624 */
625 if (atomic_dec_and_test(&bip->bli_refcount)) {
626 if (clean)
627 xfs_buf_item_relse(bp);
628 else if (aborted) {
629 ASSERT(XFS_FORCED_SHUTDOWN(lip->li_mountp));
630 if (lip->li_flags & XFS_LI_IN_AIL) {
Dave Chinner48852352013-09-24 16:01:13 +1000631 spin_lock(&lip->li_ailp->xa_lock);
Dave Chinner46f9d2e2013-09-03 21:47:37 +1000632 xfs_trans_ail_delete(lip->li_ailp, lip,
633 SHUTDOWN_LOG_IO_ERROR);
634 }
Dave Chinner3b190342013-01-21 23:53:55 +1100635 xfs_buf_item_relse(bp);
636 }
Dave Chinner46f9d2e2013-09-03 21:47:37 +1000637 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638
Dave Chinner5f6bed72013-06-27 16:04:52 +1000639 if (!(flags & XFS_BLI_HOLD))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641}
642
643/*
644 * This is called to find out where the oldest active copy of the
645 * buf log item in the on disk log resides now that the last log
646 * write of it completed at the given lsn.
647 * We always re-log all the dirty data in a buffer, so usually the
648 * latest copy in the on disk log is the only one that matters. For
649 * those cases we simply return the given lsn.
650 *
651 * The one exception to this is for buffers full of newly allocated
652 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
653 * flag set, indicating that only the di_next_unlinked fields from the
654 * inodes in the buffers will be replayed during recovery. If the
655 * original newly allocated inode images have not yet been flushed
656 * when the buffer is so relogged, then we need to make sure that we
657 * keep the old images in the 'active' portion of the log. We do this
658 * by returning the original lsn of that transaction here rather than
659 * the current one.
660 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000661STATIC xfs_lsn_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662xfs_buf_item_committed(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000663 struct xfs_log_item *lip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 xfs_lsn_t lsn)
665{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000666 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
667
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000668 trace_xfs_buf_item_committed(bip);
669
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000670 if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0)
671 return lip->li_lsn;
672 return lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673}
674
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000675STATIC void
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000676xfs_buf_item_committing(
677 struct xfs_log_item *lip,
678 xfs_lsn_t commit_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
680}
681
682/*
683 * This is the ops vector shared by all buf log items.
684 */
Christoph Hellwig272e42b2011-10-28 09:54:24 +0000685static const struct xfs_item_ops xfs_buf_item_ops = {
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000686 .iop_size = xfs_buf_item_size,
687 .iop_format = xfs_buf_item_format,
688 .iop_pin = xfs_buf_item_pin,
689 .iop_unpin = xfs_buf_item_unpin,
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000690 .iop_unlock = xfs_buf_item_unlock,
691 .iop_committed = xfs_buf_item_committed,
692 .iop_push = xfs_buf_item_push,
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000693 .iop_committing = xfs_buf_item_committing
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694};
695
Dave Chinner372cc85e2012-06-22 18:50:12 +1000696STATIC int
697xfs_buf_item_get_format(
698 struct xfs_buf_log_item *bip,
699 int count)
700{
701 ASSERT(bip->bli_formats == NULL);
702 bip->bli_format_count = count;
703
704 if (count == 1) {
Mark Tinguelyb9438172012-12-04 17:18:03 -0600705 bip->bli_formats = &bip->__bli_format;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000706 return 0;
707 }
708
709 bip->bli_formats = kmem_zalloc(count * sizeof(struct xfs_buf_log_format),
710 KM_SLEEP);
711 if (!bip->bli_formats)
712 return ENOMEM;
713 return 0;
714}
715
716STATIC void
717xfs_buf_item_free_format(
718 struct xfs_buf_log_item *bip)
719{
Mark Tinguelyb9438172012-12-04 17:18:03 -0600720 if (bip->bli_formats != &bip->__bli_format) {
Dave Chinner372cc85e2012-06-22 18:50:12 +1000721 kmem_free(bip->bli_formats);
722 bip->bli_formats = NULL;
723 }
724}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
726/*
727 * Allocate a new buf log item to go with the given buffer.
728 * Set the buffer's b_fsprivate field to point to the new
729 * buf log item. If there are other item's attached to the
730 * buffer (see xfs_buf_attach_iodone() below), then put the
731 * buf log item at the front.
732 */
733void
734xfs_buf_item_init(
735 xfs_buf_t *bp,
736 xfs_mount_t *mp)
737{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200738 xfs_log_item_t *lip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 xfs_buf_log_item_t *bip;
740 int chunks;
741 int map_size;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000742 int error;
743 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744
745 /*
746 * Check to see if there is already a buf log item for
747 * this buffer. If there is, it is guaranteed to be
748 * the first. If we do already have one, there is
749 * nothing to do here so return.
750 */
Dave Chinnerebad8612010-09-22 10:47:20 +1000751 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200752 if (lip != NULL && lip->li_type == XFS_LI_BUF)
753 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754
Dave Chinner372cc85e2012-06-22 18:50:12 +1000755 bip = kmem_zone_zalloc(xfs_buf_item_zone, KM_SLEEP);
Dave Chinner43f5efc2010-03-23 10:10:00 +1100756 xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 bip->bli_buf = bp;
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000758 xfs_buf_hold(bp);
Dave Chinner372cc85e2012-06-22 18:50:12 +1000759
760 /*
761 * chunks is the number of XFS_BLF_CHUNK size pieces the buffer
762 * can be divided into. Make sure not to truncate any pieces.
763 * map_size is the size of the bitmap needed to describe the
764 * chunks of the buffer.
765 *
766 * Discontiguous buffer support follows the layout of the underlying
767 * buffer. This makes the implementation as simple as possible.
768 */
769 error = xfs_buf_item_get_format(bip, bp->b_map_count);
770 ASSERT(error == 0);
771
772 for (i = 0; i < bip->bli_format_count; i++) {
773 chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len),
774 XFS_BLF_CHUNK);
775 map_size = DIV_ROUND_UP(chunks, NBWORD);
776
777 bip->bli_formats[i].blf_type = XFS_LI_BUF;
778 bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn;
779 bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len;
780 bip->bli_formats[i].blf_map_size = map_size;
781 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
783#ifdef XFS_TRANS_DEBUG
784 /*
785 * Allocate the arrays for tracking what needs to be logged
786 * and what our callers request to be logged. bli_orig
787 * holds a copy of the original, clean buffer for comparison
788 * against, and bli_logged keeps a 1 bit flag per byte in
789 * the buffer to indicate which bytes the callers have asked
790 * to have logged.
791 */
Dave Chinneraa0e8832012-04-23 15:58:52 +1000792 bip->bli_orig = kmem_alloc(BBTOB(bp->b_length), KM_SLEEP);
793 memcpy(bip->bli_orig, bp->b_addr, BBTOB(bp->b_length));
794 bip->bli_logged = kmem_zalloc(BBTOB(bp->b_length) / NBBY, KM_SLEEP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795#endif
796
797 /*
798 * Put the buf item into the list of items attached to the
799 * buffer at the front.
800 */
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200801 if (bp->b_fspriv)
802 bip->bli_item.li_bio_list = bp->b_fspriv;
803 bp->b_fspriv = bip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804}
805
806
807/*
808 * Mark bytes first through last inclusive as dirty in the buf
809 * item's bitmap.
810 */
811void
Dave Chinner372cc85e2012-06-22 18:50:12 +1000812xfs_buf_item_log_segment(
813 struct xfs_buf_log_item *bip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 uint first,
Dave Chinner372cc85e2012-06-22 18:50:12 +1000815 uint last,
816 uint *map)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817{
818 uint first_bit;
819 uint last_bit;
820 uint bits_to_set;
821 uint bits_set;
822 uint word_num;
823 uint *wordp;
824 uint bit;
825 uint end_bit;
826 uint mask;
827
828 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 * Convert byte offsets to bit numbers.
830 */
Dave Chinnerc1155412010-05-07 11:05:19 +1000831 first_bit = first >> XFS_BLF_SHIFT;
832 last_bit = last >> XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833
834 /*
835 * Calculate the total number of bits to be set.
836 */
837 bits_to_set = last_bit - first_bit + 1;
838
839 /*
840 * Get a pointer to the first word in the bitmap
841 * to set a bit in.
842 */
843 word_num = first_bit >> BIT_TO_WORD_SHIFT;
Dave Chinner372cc85e2012-06-22 18:50:12 +1000844 wordp = &map[word_num];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845
846 /*
847 * Calculate the starting bit in the first word.
848 */
849 bit = first_bit & (uint)(NBWORD - 1);
850
851 /*
852 * First set any bits in the first word of our range.
853 * If it starts at bit 0 of the word, it will be
854 * set below rather than here. That is what the variable
855 * bit tells us. The variable bits_set tracks the number
856 * of bits that have been set so far. End_bit is the number
857 * of the last bit to be set in this word plus one.
858 */
859 if (bit) {
860 end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
861 mask = ((1 << (end_bit - bit)) - 1) << bit;
862 *wordp |= mask;
863 wordp++;
864 bits_set = end_bit - bit;
865 } else {
866 bits_set = 0;
867 }
868
869 /*
870 * Now set bits a whole word at a time that are between
871 * first_bit and last_bit.
872 */
873 while ((bits_to_set - bits_set) >= NBWORD) {
874 *wordp |= 0xffffffff;
875 bits_set += NBWORD;
876 wordp++;
877 }
878
879 /*
880 * Finally, set any bits left to be set in one last partial word.
881 */
882 end_bit = bits_to_set - bits_set;
883 if (end_bit) {
884 mask = (1 << end_bit) - 1;
885 *wordp |= mask;
886 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887}
888
Dave Chinner372cc85e2012-06-22 18:50:12 +1000889/*
890 * Mark bytes first through last inclusive as dirty in the buf
891 * item's bitmap.
892 */
893void
894xfs_buf_item_log(
895 xfs_buf_log_item_t *bip,
896 uint first,
897 uint last)
898{
899 int i;
900 uint start;
901 uint end;
902 struct xfs_buf *bp = bip->bli_buf;
903
904 /*
Dave Chinner372cc85e2012-06-22 18:50:12 +1000905 * walk each buffer segment and mark them dirty appropriately.
906 */
907 start = 0;
908 for (i = 0; i < bip->bli_format_count; i++) {
909 if (start > last)
910 break;
911 end = start + BBTOB(bp->b_maps[i].bm_len);
912 if (first > end) {
913 start += BBTOB(bp->b_maps[i].bm_len);
914 continue;
915 }
916 if (first < start)
917 first = start;
918 if (end > last)
919 end = last;
920
921 xfs_buf_item_log_segment(bip, first, end,
922 &bip->bli_formats[i].blf_data_map[0]);
923
924 start += bp->b_maps[i].bm_len;
925 }
926}
927
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928
929/*
Dave Chinner5f6bed72013-06-27 16:04:52 +1000930 * Return 1 if the buffer has been logged or ordered in a transaction (at any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 * point, not just the current transaction) and 0 if not.
932 */
933uint
934xfs_buf_item_dirty(
935 xfs_buf_log_item_t *bip)
936{
937 return (bip->bli_flags & XFS_BLI_DIRTY);
938}
939
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000940STATIC void
941xfs_buf_item_free(
942 xfs_buf_log_item_t *bip)
943{
944#ifdef XFS_TRANS_DEBUG
945 kmem_free(bip->bli_orig);
946 kmem_free(bip->bli_logged);
947#endif /* XFS_TRANS_DEBUG */
948
Dave Chinner372cc85e2012-06-22 18:50:12 +1000949 xfs_buf_item_free_format(bip);
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000950 kmem_zone_free(xfs_buf_item_zone, bip);
951}
952
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953/*
954 * This is called when the buf log item is no longer needed. It should
955 * free the buf log item associated with the given buffer and clear
956 * the buffer's pointer to the buf log item. If there are no more
957 * items in the list, clear the b_iodone field of the buffer (see
958 * xfs_buf_attach_iodone() below).
959 */
960void
961xfs_buf_item_relse(
962 xfs_buf_t *bp)
963{
Dave Chinner5f6bed72013-06-27 16:04:52 +1000964 xfs_buf_log_item_t *bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000966 trace_xfs_buf_item_relse(bp, _RET_IP_);
Dave Chinner5f6bed72013-06-27 16:04:52 +1000967 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000968
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200969 bp->b_fspriv = bip->bli_item.li_bio_list;
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200970 if (bp->b_fspriv == NULL)
971 bp->b_iodone = NULL;
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200972
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000973 xfs_buf_rele(bp);
974 xfs_buf_item_free(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975}
976
977
978/*
979 * Add the given log item with its callback to the list of callbacks
980 * to be called when the buffer's I/O completes. If it is not set
981 * already, set the buffer's b_iodone() routine to be
982 * xfs_buf_iodone_callbacks() and link the log item into the list of
983 * items rooted at b_fsprivate. Items are always added as the second
984 * entry in the list if there is a first, because the buf item code
985 * assumes that the buf log item is first.
986 */
987void
988xfs_buf_attach_iodone(
989 xfs_buf_t *bp,
990 void (*cb)(xfs_buf_t *, xfs_log_item_t *),
991 xfs_log_item_t *lip)
992{
993 xfs_log_item_t *head_lip;
994
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200995 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996
997 lip->li_cb = cb;
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200998 head_lip = bp->b_fspriv;
999 if (head_lip) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 lip->li_bio_list = head_lip->li_bio_list;
1001 head_lip->li_bio_list = lip;
1002 } else {
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001003 bp->b_fspriv = lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 }
1005
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02001006 ASSERT(bp->b_iodone == NULL ||
1007 bp->b_iodone == xfs_buf_iodone_callbacks);
1008 bp->b_iodone = xfs_buf_iodone_callbacks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009}
1010
Dave Chinnerc90821a2010-12-03 17:00:52 +11001011/*
1012 * We can have many callbacks on a buffer. Running the callbacks individually
1013 * can cause a lot of contention on the AIL lock, so we allow for a single
1014 * callback to be able to scan the remaining lip->li_bio_list for other items
1015 * of the same type and callback to be processed in the first call.
1016 *
1017 * As a result, the loop walking the callback list below will also modify the
1018 * list. it removes the first item from the list and then runs the callback.
1019 * The loop then restarts from the new head of the list. This allows the
1020 * callback to scan and modify the list attached to the buffer and we don't
1021 * have to care about maintaining a next item pointer.
1022 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023STATIC void
1024xfs_buf_do_callbacks(
Dave Chinnerc90821a2010-12-03 17:00:52 +11001025 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026{
Dave Chinnerc90821a2010-12-03 17:00:52 +11001027 struct xfs_log_item *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001029 while ((lip = bp->b_fspriv) != NULL) {
1030 bp->b_fspriv = lip->li_bio_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 ASSERT(lip->li_cb != NULL);
1032 /*
1033 * Clear the next pointer so we don't have any
1034 * confusion if the item is added to another buf.
1035 * Don't touch the log item after calling its
1036 * callback, because it could have freed itself.
1037 */
1038 lip->li_bio_list = NULL;
1039 lip->li_cb(bp, lip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 }
1041}
1042
1043/*
1044 * This is the iodone() function for buffers which have had callbacks
1045 * attached to them by xfs_buf_attach_iodone(). It should remove each
1046 * log item from the buffer's list and call the callback of each in turn.
1047 * When done, the buffer's fsprivate field is set to NULL and the buffer
1048 * is unlocked with a call to iodone().
1049 */
1050void
1051xfs_buf_iodone_callbacks(
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001052 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053{
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001054 struct xfs_log_item *lip = bp->b_fspriv;
1055 struct xfs_mount *mp = lip->li_mountp;
1056 static ulong lasttime;
1057 static xfs_buftarg_t *lasttarg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00001059 if (likely(!xfs_buf_geterror(bp)))
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001060 goto do_callbacks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001062 /*
1063 * If we've already decided to shutdown the filesystem because of
1064 * I/O errors, there's no point in giving this a retry.
1065 */
1066 if (XFS_FORCED_SHUTDOWN(mp)) {
Christoph Hellwigc867cb62011-10-10 16:52:46 +00001067 xfs_buf_stale(bp);
Christoph Hellwigc867cb62011-10-10 16:52:46 +00001068 XFS_BUF_DONE(bp);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001069 trace_xfs_buf_item_iodone(bp, _RET_IP_);
1070 goto do_callbacks;
1071 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072
Chandra Seetharaman49074c02011-07-22 23:40:40 +00001073 if (bp->b_target != lasttarg ||
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001074 time_after(jiffies, (lasttime + 5*HZ))) {
1075 lasttime = jiffies;
Christoph Hellwigb38505b2011-10-10 16:52:50 +00001076 xfs_buf_ioerror_alert(bp, __func__);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001077 }
Chandra Seetharaman49074c02011-07-22 23:40:40 +00001078 lasttarg = bp->b_target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001080 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001081 * If the write was asynchronous then no one will be looking for the
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001082 * error. Clear the error state and write the buffer out again.
1083 *
Christoph Hellwig43ff2122012-04-23 15:58:39 +10001084 * XXX: This helps against transient write errors, but we need to find
1085 * a way to shut the filesystem down if the writes keep failing.
1086 *
1087 * In practice we'll shut the filesystem down soon as non-transient
1088 * erorrs tend to affect the whole device and a failing log write
1089 * will make us give up. But we really ought to do better here.
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001090 */
1091 if (XFS_BUF_ISASYNC(bp)) {
Christoph Hellwig43ff2122012-04-23 15:58:39 +10001092 ASSERT(bp->b_iodone != NULL);
1093
1094 trace_xfs_buf_item_iodone_async(bp, _RET_IP_);
1095
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00001096 xfs_buf_ioerror(bp, 0); /* errno of 0 unsets the flag */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001098 if (!XFS_BUF_ISSTALE(bp)) {
Christoph Hellwig43ff2122012-04-23 15:58:39 +10001099 bp->b_flags |= XBF_WRITE | XBF_ASYNC | XBF_DONE;
Christoph Hellwiga2dcf5d2012-07-13 02:24:10 -04001100 xfs_buf_iorequest(bp);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10001101 } else {
1102 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 }
Christoph Hellwig43ff2122012-04-23 15:58:39 +10001104
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 return;
1106 }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001107
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001108 /*
1109 * If the write of the buffer was synchronous, we want to make
1110 * sure to return the error to the caller of xfs_bwrite().
1111 */
Christoph Hellwigc867cb62011-10-10 16:52:46 +00001112 xfs_buf_stale(bp);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001113 XFS_BUF_DONE(bp);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001114
1115 trace_xfs_buf_error_relse(bp, _RET_IP_);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001116
1117do_callbacks:
Dave Chinnerc90821a2010-12-03 17:00:52 +11001118 xfs_buf_do_callbacks(bp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001119 bp->b_fspriv = NULL;
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02001120 bp->b_iodone = NULL;
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00001121 xfs_buf_ioend(bp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122}
1123
1124/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 * This is the iodone() function for buffers which have been
1126 * logged. It is called when they are eventually flushed out.
1127 * It should remove the buf item from the AIL, and free the buf item.
1128 * It is called by xfs_buf_iodone_callbacks() above which will take
1129 * care of cleaning up the buffer itself.
1130 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131void
1132xfs_buf_iodone(
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001133 struct xfs_buf *bp,
1134 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135{
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001136 struct xfs_ail *ailp = lip->li_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001138 ASSERT(BUF_ITEM(lip)->bli_buf == bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +10001140 xfs_buf_rele(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
1142 /*
1143 * If we are forcibly shutting down, this may well be
1144 * off the AIL already. That's because we simulate the
1145 * log-committed callbacks to unpin these buffers. Or we may never
1146 * have put this item on AIL because of the transaction was
David Chinner783a2f62008-10-30 17:39:58 +11001147 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 *
1149 * Either way, AIL is useless if we're forcing a shutdown.
1150 */
David Chinnerfc1829f2008-10-30 17:39:46 +11001151 spin_lock(&ailp->xa_lock);
Dave Chinner04913fd2012-04-23 15:58:41 +10001152 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001153 xfs_buf_item_free(BUF_ITEM(lip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154}