1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * io.c
5  *
6  * Buffer cache handling
7  *
8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9  *
10  * This program is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU General Public
12  * License as published by the Free Software Foundation; either
13  * version 2 of the License, or (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public
21  * License along with this program; if not, write to the
22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23  * Boston, MA 021110-1307, USA.
24  */
25 
26 #include <linux/fs.h>
27 #include <linux/types.h>
28 #include <linux/highmem.h>
29 #include <linux/bio.h>
30 
31 #include <cluster/masklog.h>
32 
33 #include "ocfs2.h"
34 
35 #include "alloc.h"
36 #include "inode.h"
37 #include "journal.h"
38 #include "uptodate.h"
39 #include "buffer_head_io.h"
40 #include "ocfs2_trace.h"
41 
42 /*
43  * Bits on bh->b_state used by ocfs2.
44  *
45  * These MUST be after the JBD2 bits.  Hence, we use BH_JBDPrivateStart.
46  */
47 enum ocfs2_state_bits {
48 	BH_NeedsValidate = BH_JBDPrivateStart,
49 };
50 
51 /* Expand the magic b_state functions */
52 BUFFER_FNS(NeedsValidate, needs_validate);
53 
ocfs2_write_block(struct ocfs2_super * osb,struct buffer_head * bh,struct ocfs2_caching_info * ci)54 int ocfs2_write_block(struct ocfs2_super *osb, struct buffer_head *bh,
55 		      struct ocfs2_caching_info *ci)
56 {
57 	int ret = 0;
58 
59 	trace_ocfs2_write_block((unsigned long long)bh->b_blocknr, ci);
60 
61 	BUG_ON(bh->b_blocknr < OCFS2_SUPER_BLOCK_BLKNO);
62 	BUG_ON(buffer_jbd(bh));
63 
64 	/* No need to check for a soft readonly file system here. non
65 	 * journalled writes are only ever done on system files which
66 	 * can get modified during recovery even if read-only. */
67 	if (ocfs2_is_hard_readonly(osb)) {
68 		ret = -EROFS;
69 		mlog_errno(ret);
70 		goto out;
71 	}
72 
73 	ocfs2_metadata_cache_io_lock(ci);
74 
75 	lock_buffer(bh);
76 	set_buffer_uptodate(bh);
77 
78 	/* remove from dirty list before I/O. */
79 	clear_buffer_dirty(bh);
80 
81 	get_bh(bh); /* for end_buffer_write_sync() */
82 	bh->b_end_io = end_buffer_write_sync;
83 	submit_bh(REQ_OP_WRITE, 0, bh);
84 
85 	wait_on_buffer(bh);
86 
87 	if (buffer_uptodate(bh)) {
88 		ocfs2_set_buffer_uptodate(ci, bh);
89 	} else {
90 		/* We don't need to remove the clustered uptodate
91 		 * information for this bh as it's not marked locally
92 		 * uptodate. */
93 		ret = -EIO;
94 		mlog_errno(ret);
95 	}
96 
97 	ocfs2_metadata_cache_io_unlock(ci);
98 out:
99 	return ret;
100 }
101 
102 /* Caller must provide a bhs[] with all NULL or non-NULL entries, so it
103  * will be easier to handle read failure.
104  */
ocfs2_read_blocks_sync(struct ocfs2_super * osb,u64 block,unsigned int nr,struct buffer_head * bhs[])105 int ocfs2_read_blocks_sync(struct ocfs2_super *osb, u64 block,
106 			   unsigned int nr, struct buffer_head *bhs[])
107 {
108 	int status = 0;
109 	unsigned int i;
110 	struct buffer_head *bh;
111 	int new_bh = 0;
112 
113 	trace_ocfs2_read_blocks_sync((unsigned long long)block, nr);
114 
115 	if (!nr)
116 		goto bail;
117 
118 	/* Don't put buffer head and re-assign it to NULL if it is allocated
119 	 * outside since the caller can't be aware of this alternation!
120 	 */
121 	new_bh = (bhs[0] == NULL);
122 
123 	for (i = 0 ; i < nr ; i++) {
124 		if (bhs[i] == NULL) {
125 			bhs[i] = sb_getblk(osb->sb, block++);
126 			if (bhs[i] == NULL) {
127 				status = -ENOMEM;
128 				mlog_errno(status);
129 				break;
130 			}
131 		}
132 		bh = bhs[i];
133 
134 		if (buffer_jbd(bh)) {
135 			trace_ocfs2_read_blocks_sync_jbd(
136 					(unsigned long long)bh->b_blocknr);
137 			continue;
138 		}
139 
140 		if (buffer_dirty(bh)) {
141 			/* This should probably be a BUG, or
142 			 * at least return an error. */
143 			mlog(ML_ERROR,
144 			     "trying to sync read a dirty "
145 			     "buffer! (blocknr = %llu), skipping\n",
146 			     (unsigned long long)bh->b_blocknr);
147 			continue;
148 		}
149 
150 		lock_buffer(bh);
151 		if (buffer_jbd(bh)) {
152 #ifdef CATCH_BH_JBD_RACES
153 			mlog(ML_ERROR,
154 			     "block %llu had the JBD bit set "
155 			     "while I was in lock_buffer!",
156 			     (unsigned long long)bh->b_blocknr);
157 			BUG();
158 #else
159 			unlock_buffer(bh);
160 			continue;
161 #endif
162 		}
163 
164 		get_bh(bh); /* for end_buffer_read_sync() */
165 		bh->b_end_io = end_buffer_read_sync;
166 		submit_bh(REQ_OP_READ, 0, bh);
167 	}
168 
169 read_failure:
170 	for (i = nr; i > 0; i--) {
171 		bh = bhs[i - 1];
172 
173 		if (unlikely(status)) {
174 			if (new_bh && bh) {
175 				/* If middle bh fails, let previous bh
176 				 * finish its read and then put it to
177 				 * aovoid bh leak
178 				 */
179 				if (!buffer_jbd(bh))
180 					wait_on_buffer(bh);
181 				put_bh(bh);
182 				bhs[i - 1] = NULL;
183 			} else if (bh && buffer_uptodate(bh)) {
184 				clear_buffer_uptodate(bh);
185 			}
186 			continue;
187 		}
188 
189 		/* No need to wait on the buffer if it's managed by JBD. */
190 		if (!buffer_jbd(bh))
191 			wait_on_buffer(bh);
192 
193 		if (!buffer_uptodate(bh)) {
194 			/* Status won't be cleared from here on out,
195 			 * so we can safely record this and loop back
196 			 * to cleanup the other buffers. */
197 			status = -EIO;
198 			goto read_failure;
199 		}
200 	}
201 
202 bail:
203 	return status;
204 }
205 
206 /* Caller must provide a bhs[] with all NULL or non-NULL entries, so it
207  * will be easier to handle read failure.
208  */
ocfs2_read_blocks(struct ocfs2_caching_info * ci,u64 block,int nr,struct buffer_head * bhs[],int flags,int (* validate)(struct super_block * sb,struct buffer_head * bh))209 int ocfs2_read_blocks(struct ocfs2_caching_info *ci, u64 block, int nr,
210 		      struct buffer_head *bhs[], int flags,
211 		      int (*validate)(struct super_block *sb,
212 				      struct buffer_head *bh))
213 {
214 	int status = 0;
215 	int i, ignore_cache = 0;
216 	struct buffer_head *bh;
217 	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
218 	int new_bh = 0;
219 
220 	trace_ocfs2_read_blocks_begin(ci, (unsigned long long)block, nr, flags);
221 
222 	BUG_ON(!ci);
223 	BUG_ON((flags & OCFS2_BH_READAHEAD) &&
224 	       (flags & OCFS2_BH_IGNORE_CACHE));
225 
226 	if (bhs == NULL) {
227 		status = -EINVAL;
228 		mlog_errno(status);
229 		goto bail;
230 	}
231 
232 	if (nr < 0) {
233 		mlog(ML_ERROR, "asked to read %d blocks!\n", nr);
234 		status = -EINVAL;
235 		mlog_errno(status);
236 		goto bail;
237 	}
238 
239 	if (nr == 0) {
240 		status = 0;
241 		goto bail;
242 	}
243 
244 	/* Don't put buffer head and re-assign it to NULL if it is allocated
245 	 * outside since the caller can't be aware of this alternation!
246 	 */
247 	new_bh = (bhs[0] == NULL);
248 
249 	ocfs2_metadata_cache_io_lock(ci);
250 	for (i = 0 ; i < nr ; i++) {
251 		if (bhs[i] == NULL) {
252 			bhs[i] = sb_getblk(sb, block++);
253 			if (bhs[i] == NULL) {
254 				ocfs2_metadata_cache_io_unlock(ci);
255 				status = -ENOMEM;
256 				mlog_errno(status);
257 				/* Don't forget to put previous bh! */
258 				break;
259 			}
260 		}
261 		bh = bhs[i];
262 		ignore_cache = (flags & OCFS2_BH_IGNORE_CACHE);
263 
264 		/* There are three read-ahead cases here which we need to
265 		 * be concerned with. All three assume a buffer has
266 		 * previously been submitted with OCFS2_BH_READAHEAD
267 		 * and it hasn't yet completed I/O.
268 		 *
269 		 * 1) The current request is sync to disk. This rarely
270 		 *    happens these days, and never when performance
271 		 *    matters - the code can just wait on the buffer
272 		 *    lock and re-submit.
273 		 *
274 		 * 2) The current request is cached, but not
275 		 *    readahead. ocfs2_buffer_uptodate() will return
276 		 *    false anyway, so we'll wind up waiting on the
277 		 *    buffer lock to do I/O. We re-check the request
278 		 *    with after getting the lock to avoid a re-submit.
279 		 *
280 		 * 3) The current request is readahead (and so must
281 		 *    also be a caching one). We short circuit if the
282 		 *    buffer is locked (under I/O) and if it's in the
283 		 *    uptodate cache. The re-check from #2 catches the
284 		 *    case that the previous read-ahead completes just
285 		 *    before our is-it-in-flight check.
286 		 */
287 
288 		if (!ignore_cache && !ocfs2_buffer_uptodate(ci, bh)) {
289 			trace_ocfs2_read_blocks_from_disk(
290 			     (unsigned long long)bh->b_blocknr,
291 			     (unsigned long long)ocfs2_metadata_cache_owner(ci));
292 			/* We're using ignore_cache here to say
293 			 * "go to disk" */
294 			ignore_cache = 1;
295 		}
296 
297 		trace_ocfs2_read_blocks_bh((unsigned long long)bh->b_blocknr,
298 			ignore_cache, buffer_jbd(bh), buffer_dirty(bh));
299 
300 		if (buffer_jbd(bh)) {
301 			continue;
302 		}
303 
304 		if (ignore_cache) {
305 			if (buffer_dirty(bh)) {
306 				/* This should probably be a BUG, or
307 				 * at least return an error. */
308 				continue;
309 			}
310 
311 			/* A read-ahead request was made - if the
312 			 * buffer is already under read-ahead from a
313 			 * previously submitted request than we are
314 			 * done here. */
315 			if ((flags & OCFS2_BH_READAHEAD)
316 			    && ocfs2_buffer_read_ahead(ci, bh))
317 				continue;
318 
319 			lock_buffer(bh);
320 			if (buffer_jbd(bh)) {
321 #ifdef CATCH_BH_JBD_RACES
322 				mlog(ML_ERROR, "block %llu had the JBD bit set "
323 					       "while I was in lock_buffer!",
324 				     (unsigned long long)bh->b_blocknr);
325 				BUG();
326 #else
327 				unlock_buffer(bh);
328 				continue;
329 #endif
330 			}
331 
332 			/* Re-check ocfs2_buffer_uptodate() as a
333 			 * previously read-ahead buffer may have
334 			 * completed I/O while we were waiting for the
335 			 * buffer lock. */
336 			if (!(flags & OCFS2_BH_IGNORE_CACHE)
337 			    && !(flags & OCFS2_BH_READAHEAD)
338 			    && ocfs2_buffer_uptodate(ci, bh)) {
339 				unlock_buffer(bh);
340 				continue;
341 			}
342 
343 			get_bh(bh); /* for end_buffer_read_sync() */
344 			if (validate)
345 				set_buffer_needs_validate(bh);
346 			bh->b_end_io = end_buffer_read_sync;
347 			submit_bh(REQ_OP_READ, 0, bh);
348 			continue;
349 		}
350 	}
351 
352 read_failure:
353 	for (i = (nr - 1); i >= 0; i--) {
354 		bh = bhs[i];
355 
356 		if (!(flags & OCFS2_BH_READAHEAD)) {
357 			if (unlikely(status)) {
358 				/* Clear the buffers on error including those
359 				 * ever succeeded in reading
360 				 */
361 				if (new_bh && bh) {
362 					/* If middle bh fails, let previous bh
363 					 * finish its read and then put it to
364 					 * aovoid bh leak
365 					 */
366 					if (!buffer_jbd(bh))
367 						wait_on_buffer(bh);
368 					put_bh(bh);
369 					bhs[i] = NULL;
370 				} else if (bh && buffer_uptodate(bh)) {
371 					clear_buffer_uptodate(bh);
372 				}
373 				continue;
374 			}
375 			/* We know this can't have changed as we hold the
376 			 * owner sem. Avoid doing any work on the bh if the
377 			 * journal has it. */
378 			if (!buffer_jbd(bh))
379 				wait_on_buffer(bh);
380 
381 			if (!buffer_uptodate(bh)) {
382 				/* Status won't be cleared from here on out,
383 				 * so we can safely record this and loop back
384 				 * to cleanup the other buffers. Don't need to
385 				 * remove the clustered uptodate information
386 				 * for this bh as it's not marked locally
387 				 * uptodate. */
388 				status = -EIO;
389 				clear_buffer_needs_validate(bh);
390 				goto read_failure;
391 			}
392 
393 			if (buffer_needs_validate(bh)) {
394 				/* We never set NeedsValidate if the
395 				 * buffer was held by the journal, so
396 				 * that better not have changed */
397 				BUG_ON(buffer_jbd(bh));
398 				clear_buffer_needs_validate(bh);
399 				status = validate(sb, bh);
400 				if (status)
401 					goto read_failure;
402 			}
403 		}
404 
405 		/* Always set the buffer in the cache, even if it was
406 		 * a forced read, or read-ahead which hasn't yet
407 		 * completed. */
408 		ocfs2_set_buffer_uptodate(ci, bh);
409 	}
410 	ocfs2_metadata_cache_io_unlock(ci);
411 
412 	trace_ocfs2_read_blocks_end((unsigned long long)block, nr,
413 				    flags, ignore_cache);
414 
415 bail:
416 
417 	return status;
418 }
419 
420 /* Check whether the blkno is the super block or one of the backups. */
ocfs2_check_super_or_backup(struct super_block * sb,sector_t blkno)421 static void ocfs2_check_super_or_backup(struct super_block *sb,
422 					sector_t blkno)
423 {
424 	int i;
425 	u64 backup_blkno;
426 
427 	if (blkno == OCFS2_SUPER_BLOCK_BLKNO)
428 		return;
429 
430 	for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
431 		backup_blkno = ocfs2_backup_super_blkno(sb, i);
432 		if (backup_blkno == blkno)
433 			return;
434 	}
435 
436 	BUG();
437 }
438 
439 /*
440  * Write super block and backups doesn't need to collaborate with journal,
441  * so we don't need to lock ip_io_mutex and ci doesn't need to bea passed
442  * into this function.
443  */
ocfs2_write_super_or_backup(struct ocfs2_super * osb,struct buffer_head * bh)444 int ocfs2_write_super_or_backup(struct ocfs2_super *osb,
445 				struct buffer_head *bh)
446 {
447 	int ret = 0;
448 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
449 
450 	BUG_ON(buffer_jbd(bh));
451 	ocfs2_check_super_or_backup(osb->sb, bh->b_blocknr);
452 
453 	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) {
454 		ret = -EROFS;
455 		mlog_errno(ret);
456 		goto out;
457 	}
458 
459 	lock_buffer(bh);
460 	set_buffer_uptodate(bh);
461 
462 	/* remove from dirty list before I/O. */
463 	clear_buffer_dirty(bh);
464 
465 	get_bh(bh); /* for end_buffer_write_sync() */
466 	bh->b_end_io = end_buffer_write_sync;
467 	ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &di->i_check);
468 	submit_bh(REQ_OP_WRITE, 0, bh);
469 
470 	wait_on_buffer(bh);
471 
472 	if (!buffer_uptodate(bh)) {
473 		ret = -EIO;
474 		mlog_errno(ret);
475 	}
476 
477 out:
478 	return ret;
479 }
480