]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blame - fs/gfs2/meta_io.c
gfs2: Extended attribute readahead
[mirror_ubuntu-hirsute-kernel.git] / fs / gfs2 / meta_io.c
CommitLineData
b3b94faa
DT
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
091806ed 3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
b3b94faa
DT
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
e9fc2aa0 7 * of the GNU General Public License version 2.
b3b94faa
DT
8 */
9
10#include <linux/sched.h>
11#include <linux/slab.h>
12#include <linux/spinlock.h>
13#include <linux/completion.h>
14#include <linux/buffer_head.h>
15#include <linux/mm.h>
16#include <linux/pagemap.h>
17#include <linux/writeback.h>
18#include <linux/swap.h>
19#include <linux/delay.h>
2e565bb6 20#include <linux/bio.h>
5c676f6d 21#include <linux/gfs2_ondisk.h>
b3b94faa
DT
22
23#include "gfs2.h"
5c676f6d 24#include "incore.h"
b3b94faa
DT
25#include "glock.h"
26#include "glops.h"
27#include "inode.h"
28#include "log.h"
29#include "lops.h"
30#include "meta_io.h"
31#include "rgrp.h"
32#include "trans.h"
5c676f6d 33#include "util.h"
627c10b7 34#include "trace_gfs2.h"
b3b94faa 35
4a0f9a32 36static int gfs2_aspace_writepage(struct page *page, struct writeback_control *wbc)
b3b94faa 37{
4a0f9a32
SW
38 struct buffer_head *bh, *head;
39 int nr_underway = 0;
65299a3b 40 int write_op = REQ_META | REQ_PRIO |
721a9602 41 (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE);
b3b94faa 42
4a0f9a32
SW
43 BUG_ON(!PageLocked(page));
44 BUG_ON(!page_has_buffers(page));
45
46 head = page_buffers(page);
47 bh = head;
48
49 do {
50 if (!buffer_mapped(bh))
51 continue;
52 /*
53 * If it's a fully non-blocking write attempt and we cannot
54 * lock the buffer then redirty the page. Note that this can
e76e0ec9 55 * potentially cause a busy-wait loop from flusher thread and kswapd
4a0f9a32
SW
56 * activity, but those code paths have their own higher-level
57 * throttling.
58 */
1b430bee 59 if (wbc->sync_mode != WB_SYNC_NONE) {
4a0f9a32
SW
60 lock_buffer(bh);
61 } else if (!trylock_buffer(bh)) {
62 redirty_page_for_writepage(wbc, page);
63 continue;
64 }
65 if (test_clear_buffer_dirty(bh)) {
66 mark_buffer_async_write(bh);
67 } else {
68 unlock_buffer(bh);
69 }
70 } while ((bh = bh->b_this_page) != head);
71
72 /*
73 * The page and its buffers are protected by PageWriteback(), so we can
74 * drop the bh refcounts early.
75 */
76 BUG_ON(PageWriteback(page));
77 set_page_writeback(page);
78
79 do {
80 struct buffer_head *next = bh->b_this_page;
81 if (buffer_async_write(bh)) {
82 submit_bh(write_op, bh);
83 nr_underway++;
84 }
85 bh = next;
86 } while (bh != head);
87 unlock_page(page);
88
4a0f9a32
SW
89 if (nr_underway == 0)
90 end_page_writeback(page);
91
eaefbf96 92 return 0;
b3b94faa
DT
93}
94
009d8518 95const struct address_space_operations gfs2_meta_aops = {
b3b94faa 96 .writepage = gfs2_aspace_writepage,
4340fe62 97 .releasepage = gfs2_releasepage,
b3b94faa
DT
98};
99
1b2ad412
SW
100const struct address_space_operations gfs2_rgrp_aops = {
101 .writepage = gfs2_aspace_writepage,
102 .releasepage = gfs2_releasepage,
103};
104
b3b94faa 105/**
6802e340 106 * gfs2_getbuf - Get a buffer with a given address space
cb4c0313 107 * @gl: the glock
b3b94faa
DT
108 * @blkno: the block number (filesystem scope)
109 * @create: 1 if the buffer should be created
110 *
111 * Returns: the buffer
112 */
113
6802e340 114struct buffer_head *gfs2_getbuf(struct gfs2_glock *gl, u64 blkno, int create)
b3b94faa 115{
009d8518 116 struct address_space *mapping = gfs2_glock2aspace(gl);
15562c43 117 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
b3b94faa
DT
118 struct page *page;
119 struct buffer_head *bh;
120 unsigned int shift;
121 unsigned long index;
122 unsigned int bufnum;
123
70d4ee94
SW
124 if (mapping == NULL)
125 mapping = &sdp->sd_aspace;
126
b3b94faa
DT
127 shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift;
128 index = blkno >> shift; /* convert block to page */
129 bufnum = blkno - (index << shift); /* block buf index within page */
130
131 if (create) {
132 for (;;) {
cb4c0313 133 page = grab_cache_page(mapping, index);
b3b94faa
DT
134 if (page)
135 break;
136 yield();
137 }
138 } else {
2457aec6
MG
139 page = find_get_page_flags(mapping, index,
140 FGP_LOCK|FGP_ACCESSED);
b3b94faa
DT
141 if (!page)
142 return NULL;
143 }
144
145 if (!page_has_buffers(page))
146 create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0);
147
148 /* Locate header for our buffer within our page */
149 for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page)
150 /* Do nothing */;
151 get_bh(bh);
152
153 if (!buffer_mapped(bh))
154 map_bh(bh, sdp->sd_vfs, blkno);
155
156 unlock_page(page);
b3b94faa
DT
157 page_cache_release(page);
158
159 return bh;
160}
161
162static void meta_prep_new(struct buffer_head *bh)
163{
164 struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data;
165
166 lock_buffer(bh);
167 clear_buffer_dirty(bh);
168 set_buffer_uptodate(bh);
169 unlock_buffer(bh);
170
171 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
172}
173
174/**
175 * gfs2_meta_new - Get a block
176 * @gl: The glock associated with this block
177 * @blkno: The block number
178 *
179 * Returns: The buffer
180 */
181
cd915493 182struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno)
b3b94faa
DT
183{
184 struct buffer_head *bh;
6802e340 185 bh = gfs2_getbuf(gl, blkno, CREATE);
b3b94faa
DT
186 meta_prep_new(bh);
187 return bh;
188}
189
c8d57703
AG
190static void gfs2_meta_readahead(struct gfs2_glock *gl, u64 blkno)
191{
192 struct buffer_head *bh;
193
194 bh = gfs2_getbuf(gl, blkno, 1);
195 lock_buffer(bh);
196 if (buffer_uptodate(bh)) {
197 unlock_buffer(bh);
198 brelse(bh);
199 return;
200 }
201 bh->b_end_io = end_buffer_read_sync;
202 submit_bh(READA | REQ_META | REQ_PRIO, bh);
203}
204
b3b94faa
DT
205/**
206 * gfs2_meta_read - Read a block from disk
207 * @gl: The glock covering the block
208 * @blkno: The block number
7276b3b0 209 * @flags: flags
b3b94faa
DT
210 * @bhp: the place where the buffer is returned (NULL on failure)
211 *
212 * Returns: errno
213 */
214
cd915493 215int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags,
c8d57703 216 int rahead, struct buffer_head **bhp)
b3b94faa 217{
15562c43 218 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
c969f58c
SW
219 struct buffer_head *bh;
220
44b8db13
MY
221 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
222 *bhp = NULL;
c969f58c 223 return -EIO;
44b8db13 224 }
c969f58c
SW
225
226 *bhp = bh = gfs2_getbuf(gl, blkno, CREATE);
227
228 lock_buffer(bh);
229 if (buffer_uptodate(bh)) {
230 unlock_buffer(bh);
c8d57703
AG
231 if (rahead)
232 gfs2_meta_readahead(gl, blkno + 1);
c969f58c
SW
233 return 0;
234 }
235 bh->b_end_io = end_buffer_read_sync;
236 get_bh(bh);
65299a3b 237 submit_bh(READ_SYNC | REQ_META | REQ_PRIO, bh);
c8d57703
AG
238 if (rahead)
239 gfs2_meta_readahead(gl, blkno + 1);
c969f58c
SW
240 if (!(flags & DIO_WAIT))
241 return 0;
242
243 wait_on_buffer(bh);
244 if (unlikely(!buffer_uptodate(bh))) {
245 struct gfs2_trans *tr = current->journal_info;
246 if (tr && tr->tr_touched)
247 gfs2_io_error_bh(sdp, bh);
248 brelse(bh);
44b8db13 249 *bhp = NULL;
c969f58c 250 return -EIO;
7276b3b0 251 }
b3b94faa 252
7276b3b0 253 return 0;
b3b94faa
DT
254}
255
256/**
7276b3b0 257 * gfs2_meta_wait - Reread a block from disk
b3b94faa 258 * @sdp: the filesystem
7276b3b0 259 * @bh: The block to wait for
b3b94faa
DT
260 *
261 * Returns: errno
262 */
263
7276b3b0 264int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh)
b3b94faa
DT
265{
266 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
267 return -EIO;
268
7276b3b0 269 wait_on_buffer(bh);
b3b94faa 270
7276b3b0
SW
271 if (!buffer_uptodate(bh)) {
272 struct gfs2_trans *tr = current->journal_info;
273 if (tr && tr->tr_touched)
274 gfs2_io_error_bh(sdp, bh);
275 return -EIO;
b3b94faa 276 }
7276b3b0
SW
277 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
278 return -EIO;
b3b94faa
DT
279
280 return 0;
281}
282
16615be1
SW
283void gfs2_remove_from_journal(struct buffer_head *bh, struct gfs2_trans *tr, int meta)
284{
009d8518
SW
285 struct address_space *mapping = bh->b_page->mapping;
286 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
16615be1 287 struct gfs2_bufdata *bd = bh->b_private;
502be2a3 288 int was_pinned = 0;
009d8518 289
16615be1 290 if (test_clear_buffer_pinned(bh)) {
627c10b7 291 trace_gfs2_pin(bd, 0);
5e687eac 292 atomic_dec(&sdp->sd_log_pinned);
c0752aa7 293 list_del_init(&bd->bd_list);
022ef4fe 294 if (meta)
16615be1 295 tr->tr_num_buf_rm++;
022ef4fe 296 else
16615be1 297 tr->tr_num_databuf_rm++;
16615be1 298 tr->tr_touched = 1;
502be2a3 299 was_pinned = 1;
16615be1
SW
300 brelse(bh);
301 }
302 if (bd) {
c618e87a 303 spin_lock(&sdp->sd_ail_lock);
16ca9412 304 if (bd->bd_tr) {
16615be1 305 gfs2_trans_add_revoke(sdp, bd);
502be2a3
BP
306 } else if (was_pinned) {
307 bh->b_private = NULL;
308 kmem_cache_free(gfs2_bufdata_cachep, bd);
16615be1 309 }
c618e87a 310 spin_unlock(&sdp->sd_ail_lock);
16615be1
SW
311 }
312 clear_buffer_dirty(bh);
313 clear_buffer_uptodate(bh);
314}
315
b3b94faa
DT
316/**
317 * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore
318 * @ip: the inode who owns the buffers
319 * @bstart: the first buffer in the run
320 * @blen: the number of buffers in the run
321 *
322 */
323
cd915493 324void gfs2_meta_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen)
b3b94faa 325{
feaa7bba 326 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
b3b94faa
DT
327 struct buffer_head *bh;
328
329 while (blen) {
6802e340 330 bh = gfs2_getbuf(ip->i_gl, bstart, NO_CREATE);
b3b94faa 331 if (bh) {
1ad38c43
SW
332 lock_buffer(bh);
333 gfs2_log_lock(sdp);
16615be1 334 gfs2_remove_from_journal(bh, current->journal_info, 1);
1ad38c43 335 gfs2_log_unlock(sdp);
b3b94faa 336 unlock_buffer(bh);
b3b94faa
DT
337 brelse(bh);
338 }
339
340 bstart++;
341 blen--;
342 }
343}
344
b3b94faa
DT
345/**
346 * gfs2_meta_indirect_buffer - Get a metadata buffer
347 * @ip: The GFS2 inode
348 * @height: The level of this buf in the metadata (indir addr) tree (if any)
349 * @num: The block number (device relative) of the buffer
b3b94faa
DT
350 * @bhp: the buffer is returned here
351 *
b3b94faa
DT
352 * Returns: errno
353 */
354
cd915493 355int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, u64 num,
f2f9c812 356 struct buffer_head **bhp)
b3b94faa 357{
7276b3b0
SW
358 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
359 struct gfs2_glock *gl = ip->i_gl;
f91a0d3e
SW
360 struct buffer_head *bh;
361 int ret = 0;
f2f9c812 362 u32 mtype = height ? GFS2_METATYPE_IN : GFS2_METATYPE_DI;
c8d57703
AG
363 int rahead = 0;
364
365 if (num == ip->i_no_addr)
366 rahead = ip->i_rahead;
b3b94faa 367
c8d57703 368 ret = gfs2_meta_read(gl, num, DIO_WAIT, rahead, &bh);
f2f9c812
BP
369 if (ret == 0 && gfs2_metatype_check(sdp, bh, mtype)) {
370 brelse(bh);
371 ret = -EIO;
b3b94faa 372 }
b3b94faa 373 *bhp = bh;
f91a0d3e 374 return ret;
b3b94faa
DT
375}
376
377/**
378 * gfs2_meta_ra - start readahead on an extent of a file
379 * @gl: the glock the blocks belong to
380 * @dblock: the starting disk block
381 * @extlen: the number of blocks in the extent
382 *
7276b3b0 383 * returns: the first buffer in the extent
b3b94faa
DT
384 */
385
7276b3b0 386struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen)
b3b94faa 387{
15562c43 388 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
b3b94faa 389 struct buffer_head *first_bh, *bh;
cd915493 390 u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >>
568f4c96 391 sdp->sd_sb.sb_bsize_shift;
b3b94faa 392
7276b3b0
SW
393 BUG_ON(!extlen);
394
395 if (max_ra < 1)
396 max_ra = 1;
b3b94faa
DT
397 if (extlen > max_ra)
398 extlen = max_ra;
399
6802e340 400 first_bh = gfs2_getbuf(gl, dblock, CREATE);
b3b94faa
DT
401
402 if (buffer_uptodate(first_bh))
403 goto out;
7276b3b0 404 if (!buffer_locked(first_bh))
20ed0535 405 ll_rw_block(READ_SYNC | REQ_META, 1, &first_bh);
b3b94faa
DT
406
407 dblock++;
408 extlen--;
409
410 while (extlen) {
6802e340 411 bh = gfs2_getbuf(gl, dblock, CREATE);
b3b94faa 412
7276b3b0 413 if (!buffer_uptodate(bh) && !buffer_locked(bh))
20ed0535 414 ll_rw_block(READA | REQ_META, 1, &bh);
7276b3b0 415 brelse(bh);
b3b94faa
DT
416 dblock++;
417 extlen--;
7276b3b0
SW
418 if (!buffer_locked(first_bh) && buffer_uptodate(first_bh))
419 goto out;
b3b94faa
DT
420 }
421
7276b3b0 422 wait_on_buffer(first_bh);
a91ea69f 423out:
7276b3b0 424 return first_bh;
b3b94faa
DT
425}
426