xref: /openbmc/linux/fs/gfs2/rgrp.c (revision 82ced6fd)
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
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
7  * of the GNU General Public License version 2.
8  */
9 
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/fs.h>
15 #include <linux/gfs2_ondisk.h>
16 #include <linux/prefetch.h>
17 #include <linux/blkdev.h>
18 
19 #include "gfs2.h"
20 #include "incore.h"
21 #include "glock.h"
22 #include "glops.h"
23 #include "lops.h"
24 #include "meta_io.h"
25 #include "quota.h"
26 #include "rgrp.h"
27 #include "super.h"
28 #include "trans.h"
29 #include "util.h"
30 #include "log.h"
31 #include "inode.h"
32 #include "ops_address.h"
33 
34 #define BFITNOENT ((u32)~0)
35 #define NO_BLOCK ((u64)~0)
36 
37 #if BITS_PER_LONG == 32
38 #define LBITMASK   (0x55555555UL)
39 #define LBITSKIP55 (0x55555555UL)
40 #define LBITSKIP00 (0x00000000UL)
41 #else
42 #define LBITMASK   (0x5555555555555555UL)
43 #define LBITSKIP55 (0x5555555555555555UL)
44 #define LBITSKIP00 (0x0000000000000000UL)
45 #endif
46 
47 /*
48  * These routines are used by the resource group routines (rgrp.c)
49  * to keep track of block allocation.  Each block is represented by two
50  * bits.  So, each byte represents GFS2_NBBY (i.e. 4) blocks.
51  *
52  * 0 = Free
53  * 1 = Used (not metadata)
54  * 2 = Unlinked (still in use) inode
55  * 3 = Used (metadata)
56  */
57 
58 static const char valid_change[16] = {
59 	        /* current */
60 	/* n */ 0, 1, 1, 1,
61 	/* e */ 1, 0, 0, 0,
62 	/* w */ 0, 0, 0, 1,
63 	        1, 0, 0, 0
64 };
65 
66 static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal,
67                         unsigned char old_state, unsigned char new_state,
68 			unsigned int *n);
69 
70 /**
71  * gfs2_setbit - Set a bit in the bitmaps
72  * @buffer: the buffer that holds the bitmaps
73  * @buflen: the length (in bytes) of the buffer
74  * @block: the block to set
75  * @new_state: the new state of the block
76  *
77  */
78 
79 static inline void gfs2_setbit(struct gfs2_rgrpd *rgd, unsigned char *buf1,
80 			       unsigned char *buf2, unsigned int offset,
81 			       unsigned int buflen, u32 block,
82 			       unsigned char new_state)
83 {
84 	unsigned char *byte1, *byte2, *end, cur_state;
85 	const unsigned int bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE;
86 
87 	byte1 = buf1 + offset + (block / GFS2_NBBY);
88 	end = buf1 + offset + buflen;
89 
90 	BUG_ON(byte1 >= end);
91 
92 	cur_state = (*byte1 >> bit) & GFS2_BIT_MASK;
93 
94 	if (unlikely(!valid_change[new_state * 4 + cur_state])) {
95 		gfs2_consist_rgrpd(rgd);
96 		return;
97 	}
98 	*byte1 ^= (cur_state ^ new_state) << bit;
99 
100 	if (buf2) {
101 		byte2 = buf2 + offset + (block / GFS2_NBBY);
102 		cur_state = (*byte2 >> bit) & GFS2_BIT_MASK;
103 		*byte2 ^= (cur_state ^ new_state) << bit;
104 	}
105 }
106 
107 /**
108  * gfs2_testbit - test a bit in the bitmaps
109  * @buffer: the buffer that holds the bitmaps
110  * @buflen: the length (in bytes) of the buffer
111  * @block: the block to read
112  *
113  */
114 
115 static inline unsigned char gfs2_testbit(struct gfs2_rgrpd *rgd,
116 					 const unsigned char *buffer,
117 					 unsigned int buflen, u32 block)
118 {
119 	const unsigned char *byte, *end;
120 	unsigned char cur_state;
121 	unsigned int bit;
122 
123 	byte = buffer + (block / GFS2_NBBY);
124 	bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE;
125 	end = buffer + buflen;
126 
127 	gfs2_assert(rgd->rd_sbd, byte < end);
128 
129 	cur_state = (*byte >> bit) & GFS2_BIT_MASK;
130 
131 	return cur_state;
132 }
133 
134 /**
135  * gfs2_bit_search
136  * @ptr: Pointer to bitmap data
137  * @mask: Mask to use (normally 0x55555.... but adjusted for search start)
138  * @state: The state we are searching for
139  *
140  * We xor the bitmap data with a patter which is the bitwise opposite
141  * of what we are looking for, this gives rise to a pattern of ones
142  * wherever there is a match. Since we have two bits per entry, we
143  * take this pattern, shift it down by one place and then and it with
144  * the original. All the even bit positions (0,2,4, etc) then represent
145  * successful matches, so we mask with 0x55555..... to remove the unwanted
146  * odd bit positions.
147  *
148  * This allows searching of a whole u64 at once (32 blocks) with a
149  * single test (on 64 bit arches).
150  */
151 
152 static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state)
153 {
154 	u64 tmp;
155 	static const u64 search[] = {
156 		[0] = 0xffffffffffffffffULL,
157 		[1] = 0xaaaaaaaaaaaaaaaaULL,
158 		[2] = 0x5555555555555555ULL,
159 		[3] = 0x0000000000000000ULL,
160 	};
161 	tmp = le64_to_cpu(*ptr) ^ search[state];
162 	tmp &= (tmp >> 1);
163 	tmp &= mask;
164 	return tmp;
165 }
166 
167 /**
168  * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing
169  *       a block in a given allocation state.
170  * @buffer: the buffer that holds the bitmaps
171  * @len: the length (in bytes) of the buffer
172  * @goal: start search at this block's bit-pair (within @buffer)
173  * @state: GFS2_BLKST_XXX the state of the block we're looking for.
174  *
175  * Scope of @goal and returned block number is only within this bitmap buffer,
176  * not entire rgrp or filesystem.  @buffer will be offset from the actual
177  * beginning of a bitmap block buffer, skipping any header structures, but
178  * headers are always a multiple of 64 bits long so that the buffer is
179  * always aligned to a 64 bit boundary.
180  *
181  * The size of the buffer is in bytes, but is it assumed that it is
182  * always ok to to read a complete multiple of 64 bits at the end
183  * of the block in case the end is no aligned to a natural boundary.
184  *
185  * Return: the block number (bitmap buffer scope) that was found
186  */
187 
188 static u32 gfs2_bitfit(const u8 *buf, const unsigned int len,
189 		       u32 goal, u8 state)
190 {
191 	u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1);
192 	const __le64 *ptr = ((__le64 *)buf) + (goal >> 5);
193 	const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64)));
194 	u64 tmp;
195 	u64 mask = 0x5555555555555555ULL;
196 	u32 bit;
197 
198 	BUG_ON(state > 3);
199 
200 	/* Mask off bits we don't care about at the start of the search */
201 	mask <<= spoint;
202 	tmp = gfs2_bit_search(ptr, mask, state);
203 	ptr++;
204 	while(tmp == 0 && ptr < end) {
205 		tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state);
206 		ptr++;
207 	}
208 	/* Mask off any bits which are more than len bytes from the start */
209 	if (ptr == end && (len & (sizeof(u64) - 1)))
210 		tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1))));
211 	/* Didn't find anything, so return */
212 	if (tmp == 0)
213 		return BFITNOENT;
214 	ptr--;
215 	bit = __ffs64(tmp);
216 	bit /= 2;	/* two bits per entry in the bitmap */
217 	return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit;
218 }
219 
220 /**
221  * gfs2_bitcount - count the number of bits in a certain state
222  * @buffer: the buffer that holds the bitmaps
223  * @buflen: the length (in bytes) of the buffer
224  * @state: the state of the block we're looking for
225  *
226  * Returns: The number of bits
227  */
228 
229 static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer,
230 			 unsigned int buflen, u8 state)
231 {
232 	const u8 *byte = buffer;
233 	const u8 *end = buffer + buflen;
234 	const u8 state1 = state << 2;
235 	const u8 state2 = state << 4;
236 	const u8 state3 = state << 6;
237 	u32 count = 0;
238 
239 	for (; byte < end; byte++) {
240 		if (((*byte) & 0x03) == state)
241 			count++;
242 		if (((*byte) & 0x0C) == state1)
243 			count++;
244 		if (((*byte) & 0x30) == state2)
245 			count++;
246 		if (((*byte) & 0xC0) == state3)
247 			count++;
248 	}
249 
250 	return count;
251 }
252 
253 /**
254  * gfs2_rgrp_verify - Verify that a resource group is consistent
255  * @sdp: the filesystem
256  * @rgd: the rgrp
257  *
258  */
259 
260 void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd)
261 {
262 	struct gfs2_sbd *sdp = rgd->rd_sbd;
263 	struct gfs2_bitmap *bi = NULL;
264 	u32 length = rgd->rd_length;
265 	u32 count[4], tmp;
266 	int buf, x;
267 
268 	memset(count, 0, 4 * sizeof(u32));
269 
270 	/* Count # blocks in each of 4 possible allocation states */
271 	for (buf = 0; buf < length; buf++) {
272 		bi = rgd->rd_bits + buf;
273 		for (x = 0; x < 4; x++)
274 			count[x] += gfs2_bitcount(rgd,
275 						  bi->bi_bh->b_data +
276 						  bi->bi_offset,
277 						  bi->bi_len, x);
278 	}
279 
280 	if (count[0] != rgd->rd_free) {
281 		if (gfs2_consist_rgrpd(rgd))
282 			fs_err(sdp, "free data mismatch:  %u != %u\n",
283 			       count[0], rgd->rd_free);
284 		return;
285 	}
286 
287 	tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes;
288 	if (count[1] + count[2] != tmp) {
289 		if (gfs2_consist_rgrpd(rgd))
290 			fs_err(sdp, "used data mismatch:  %u != %u\n",
291 			       count[1], tmp);
292 		return;
293 	}
294 
295 	if (count[3] != rgd->rd_dinodes) {
296 		if (gfs2_consist_rgrpd(rgd))
297 			fs_err(sdp, "used metadata mismatch:  %u != %u\n",
298 			       count[3], rgd->rd_dinodes);
299 		return;
300 	}
301 
302 	if (count[2] > count[3]) {
303 		if (gfs2_consist_rgrpd(rgd))
304 			fs_err(sdp, "unlinked inodes > inodes:  %u\n",
305 			       count[2]);
306 		return;
307 	}
308 
309 }
310 
311 static inline int rgrp_contains_block(struct gfs2_rgrpd *rgd, u64 block)
312 {
313 	u64 first = rgd->rd_data0;
314 	u64 last = first + rgd->rd_data;
315 	return first <= block && block < last;
316 }
317 
318 /**
319  * gfs2_blk2rgrpd - Find resource group for a given data/meta block number
320  * @sdp: The GFS2 superblock
321  * @n: The data block number
322  *
323  * Returns: The resource group, or NULL if not found
324  */
325 
326 struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk)
327 {
328 	struct gfs2_rgrpd *rgd;
329 
330 	spin_lock(&sdp->sd_rindex_spin);
331 
332 	list_for_each_entry(rgd, &sdp->sd_rindex_mru_list, rd_list_mru) {
333 		if (rgrp_contains_block(rgd, blk)) {
334 			list_move(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list);
335 			spin_unlock(&sdp->sd_rindex_spin);
336 			return rgd;
337 		}
338 	}
339 
340 	spin_unlock(&sdp->sd_rindex_spin);
341 
342 	return NULL;
343 }
344 
345 /**
346  * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem
347  * @sdp: The GFS2 superblock
348  *
349  * Returns: The first rgrp in the filesystem
350  */
351 
352 struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp)
353 {
354 	gfs2_assert(sdp, !list_empty(&sdp->sd_rindex_list));
355 	return list_entry(sdp->sd_rindex_list.next, struct gfs2_rgrpd, rd_list);
356 }
357 
358 /**
359  * gfs2_rgrpd_get_next - get the next RG
360  * @rgd: A RG
361  *
362  * Returns: The next rgrp
363  */
364 
365 struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd)
366 {
367 	if (rgd->rd_list.next == &rgd->rd_sbd->sd_rindex_list)
368 		return NULL;
369 	return list_entry(rgd->rd_list.next, struct gfs2_rgrpd, rd_list);
370 }
371 
372 static void clear_rgrpdi(struct gfs2_sbd *sdp)
373 {
374 	struct list_head *head;
375 	struct gfs2_rgrpd *rgd;
376 	struct gfs2_glock *gl;
377 
378 	spin_lock(&sdp->sd_rindex_spin);
379 	sdp->sd_rindex_forward = NULL;
380 	spin_unlock(&sdp->sd_rindex_spin);
381 
382 	head = &sdp->sd_rindex_list;
383 	while (!list_empty(head)) {
384 		rgd = list_entry(head->next, struct gfs2_rgrpd, rd_list);
385 		gl = rgd->rd_gl;
386 
387 		list_del(&rgd->rd_list);
388 		list_del(&rgd->rd_list_mru);
389 
390 		if (gl) {
391 			gl->gl_object = NULL;
392 			gfs2_glock_put(gl);
393 		}
394 
395 		kfree(rgd->rd_bits);
396 		kmem_cache_free(gfs2_rgrpd_cachep, rgd);
397 	}
398 }
399 
400 void gfs2_clear_rgrpd(struct gfs2_sbd *sdp)
401 {
402 	mutex_lock(&sdp->sd_rindex_mutex);
403 	clear_rgrpdi(sdp);
404 	mutex_unlock(&sdp->sd_rindex_mutex);
405 }
406 
407 static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd)
408 {
409 	printk(KERN_INFO "  ri_addr = %llu\n", (unsigned long long)rgd->rd_addr);
410 	printk(KERN_INFO "  ri_length = %u\n", rgd->rd_length);
411 	printk(KERN_INFO "  ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0);
412 	printk(KERN_INFO "  ri_data = %u\n", rgd->rd_data);
413 	printk(KERN_INFO "  ri_bitbytes = %u\n", rgd->rd_bitbytes);
414 }
415 
416 /**
417  * gfs2_compute_bitstructs - Compute the bitmap sizes
418  * @rgd: The resource group descriptor
419  *
420  * Calculates bitmap descriptors, one for each block that contains bitmap data
421  *
422  * Returns: errno
423  */
424 
425 static int compute_bitstructs(struct gfs2_rgrpd *rgd)
426 {
427 	struct gfs2_sbd *sdp = rgd->rd_sbd;
428 	struct gfs2_bitmap *bi;
429 	u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */
430 	u32 bytes_left, bytes;
431 	int x;
432 
433 	if (!length)
434 		return -EINVAL;
435 
436 	rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS);
437 	if (!rgd->rd_bits)
438 		return -ENOMEM;
439 
440 	bytes_left = rgd->rd_bitbytes;
441 
442 	for (x = 0; x < length; x++) {
443 		bi = rgd->rd_bits + x;
444 
445 		/* small rgrp; bitmap stored completely in header block */
446 		if (length == 1) {
447 			bytes = bytes_left;
448 			bi->bi_offset = sizeof(struct gfs2_rgrp);
449 			bi->bi_start = 0;
450 			bi->bi_len = bytes;
451 		/* header block */
452 		} else if (x == 0) {
453 			bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp);
454 			bi->bi_offset = sizeof(struct gfs2_rgrp);
455 			bi->bi_start = 0;
456 			bi->bi_len = bytes;
457 		/* last block */
458 		} else if (x + 1 == length) {
459 			bytes = bytes_left;
460 			bi->bi_offset = sizeof(struct gfs2_meta_header);
461 			bi->bi_start = rgd->rd_bitbytes - bytes_left;
462 			bi->bi_len = bytes;
463 		/* other blocks */
464 		} else {
465 			bytes = sdp->sd_sb.sb_bsize -
466 				sizeof(struct gfs2_meta_header);
467 			bi->bi_offset = sizeof(struct gfs2_meta_header);
468 			bi->bi_start = rgd->rd_bitbytes - bytes_left;
469 			bi->bi_len = bytes;
470 		}
471 
472 		bytes_left -= bytes;
473 	}
474 
475 	if (bytes_left) {
476 		gfs2_consist_rgrpd(rgd);
477 		return -EIO;
478 	}
479 	bi = rgd->rd_bits + (length - 1);
480 	if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) {
481 		if (gfs2_consist_rgrpd(rgd)) {
482 			gfs2_rindex_print(rgd);
483 			fs_err(sdp, "start=%u len=%u offset=%u\n",
484 			       bi->bi_start, bi->bi_len, bi->bi_offset);
485 		}
486 		return -EIO;
487 	}
488 
489 	return 0;
490 }
491 
492 /**
493  * gfs2_ri_total - Total up the file system space, according to the rindex.
494  *
495  */
496 u64 gfs2_ri_total(struct gfs2_sbd *sdp)
497 {
498 	u64 total_data = 0;
499 	struct inode *inode = sdp->sd_rindex;
500 	struct gfs2_inode *ip = GFS2_I(inode);
501 	char buf[sizeof(struct gfs2_rindex)];
502 	struct file_ra_state ra_state;
503 	int error, rgrps;
504 
505 	mutex_lock(&sdp->sd_rindex_mutex);
506 	file_ra_state_init(&ra_state, inode->i_mapping);
507 	for (rgrps = 0;; rgrps++) {
508 		loff_t pos = rgrps * sizeof(struct gfs2_rindex);
509 
510 		if (pos + sizeof(struct gfs2_rindex) >= ip->i_disksize)
511 			break;
512 		error = gfs2_internal_read(ip, &ra_state, buf, &pos,
513 					   sizeof(struct gfs2_rindex));
514 		if (error != sizeof(struct gfs2_rindex))
515 			break;
516 		total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data);
517 	}
518 	mutex_unlock(&sdp->sd_rindex_mutex);
519 	return total_data;
520 }
521 
522 static void gfs2_rindex_in(struct gfs2_rgrpd *rgd, const void *buf)
523 {
524 	const struct gfs2_rindex *str = buf;
525 
526 	rgd->rd_addr = be64_to_cpu(str->ri_addr);
527 	rgd->rd_length = be32_to_cpu(str->ri_length);
528 	rgd->rd_data0 = be64_to_cpu(str->ri_data0);
529 	rgd->rd_data = be32_to_cpu(str->ri_data);
530 	rgd->rd_bitbytes = be32_to_cpu(str->ri_bitbytes);
531 }
532 
533 /**
534  * read_rindex_entry - Pull in a new resource index entry from the disk
535  * @gl: The glock covering the rindex inode
536  *
537  * Returns: 0 on success, error code otherwise
538  */
539 
540 static int read_rindex_entry(struct gfs2_inode *ip,
541 			     struct file_ra_state *ra_state)
542 {
543 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
544 	loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex);
545 	char buf[sizeof(struct gfs2_rindex)];
546 	int error;
547 	struct gfs2_rgrpd *rgd;
548 
549 	error = gfs2_internal_read(ip, ra_state, buf, &pos,
550 				   sizeof(struct gfs2_rindex));
551 	if (!error)
552 		return 0;
553 	if (error != sizeof(struct gfs2_rindex)) {
554 		if (error > 0)
555 			error = -EIO;
556 		return error;
557 	}
558 
559 	rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS);
560 	error = -ENOMEM;
561 	if (!rgd)
562 		return error;
563 
564 	mutex_init(&rgd->rd_mutex);
565 	lops_init_le(&rgd->rd_le, &gfs2_rg_lops);
566 	rgd->rd_sbd = sdp;
567 
568 	list_add_tail(&rgd->rd_list, &sdp->sd_rindex_list);
569 	list_add_tail(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list);
570 
571 	gfs2_rindex_in(rgd, buf);
572 	error = compute_bitstructs(rgd);
573 	if (error)
574 		return error;
575 
576 	error = gfs2_glock_get(sdp, rgd->rd_addr,
577 			       &gfs2_rgrp_glops, CREATE, &rgd->rd_gl);
578 	if (error)
579 		return error;
580 
581 	rgd->rd_gl->gl_object = rgd;
582 	rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
583 	rgd->rd_flags |= GFS2_RDF_CHECK;
584 	return error;
585 }
586 
587 /**
588  * gfs2_ri_update - Pull in a new resource index from the disk
589  * @ip: pointer to the rindex inode
590  *
591  * Returns: 0 on successful update, error code otherwise
592  */
593 
594 static int gfs2_ri_update(struct gfs2_inode *ip)
595 {
596 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
597 	struct inode *inode = &ip->i_inode;
598 	struct file_ra_state ra_state;
599 	u64 rgrp_count = ip->i_disksize;
600 	int error;
601 
602 	if (do_div(rgrp_count, sizeof(struct gfs2_rindex))) {
603 		gfs2_consist_inode(ip);
604 		return -EIO;
605 	}
606 
607 	clear_rgrpdi(sdp);
608 
609 	file_ra_state_init(&ra_state, inode->i_mapping);
610 	for (sdp->sd_rgrps = 0; sdp->sd_rgrps < rgrp_count; sdp->sd_rgrps++) {
611 		error = read_rindex_entry(ip, &ra_state);
612 		if (error) {
613 			clear_rgrpdi(sdp);
614 			return error;
615 		}
616 	}
617 
618 	sdp->sd_rindex_uptodate = 1;
619 	return 0;
620 }
621 
622 /**
623  * gfs2_ri_update_special - Pull in a new resource index from the disk
624  *
625  * This is a special version that's safe to call from gfs2_inplace_reserve_i.
626  * In this case we know that we don't have any resource groups in memory yet.
627  *
628  * @ip: pointer to the rindex inode
629  *
630  * Returns: 0 on successful update, error code otherwise
631  */
632 static int gfs2_ri_update_special(struct gfs2_inode *ip)
633 {
634 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
635 	struct inode *inode = &ip->i_inode;
636 	struct file_ra_state ra_state;
637 	int error;
638 
639 	file_ra_state_init(&ra_state, inode->i_mapping);
640 	for (sdp->sd_rgrps = 0;; sdp->sd_rgrps++) {
641 		/* Ignore partials */
642 		if ((sdp->sd_rgrps + 1) * sizeof(struct gfs2_rindex) >
643 		    ip->i_disksize)
644 			break;
645 		error = read_rindex_entry(ip, &ra_state);
646 		if (error) {
647 			clear_rgrpdi(sdp);
648 			return error;
649 		}
650 	}
651 
652 	sdp->sd_rindex_uptodate = 1;
653 	return 0;
654 }
655 
656 /**
657  * gfs2_rindex_hold - Grab a lock on the rindex
658  * @sdp: The GFS2 superblock
659  * @ri_gh: the glock holder
660  *
661  * We grab a lock on the rindex inode to make sure that it doesn't
662  * change whilst we are performing an operation. We keep this lock
663  * for quite long periods of time compared to other locks. This
664  * doesn't matter, since it is shared and it is very, very rarely
665  * accessed in the exclusive mode (i.e. only when expanding the filesystem).
666  *
667  * This makes sure that we're using the latest copy of the resource index
668  * special file, which might have been updated if someone expanded the
669  * filesystem (via gfs2_grow utility), which adds new resource groups.
670  *
671  * Returns: 0 on success, error code otherwise
672  */
673 
674 int gfs2_rindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ri_gh)
675 {
676 	struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex);
677 	struct gfs2_glock *gl = ip->i_gl;
678 	int error;
679 
680 	error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, ri_gh);
681 	if (error)
682 		return error;
683 
684 	/* Read new copy from disk if we don't have the latest */
685 	if (!sdp->sd_rindex_uptodate) {
686 		mutex_lock(&sdp->sd_rindex_mutex);
687 		if (!sdp->sd_rindex_uptodate) {
688 			error = gfs2_ri_update(ip);
689 			if (error)
690 				gfs2_glock_dq_uninit(ri_gh);
691 		}
692 		mutex_unlock(&sdp->sd_rindex_mutex);
693 	}
694 
695 	return error;
696 }
697 
698 static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf)
699 {
700 	const struct gfs2_rgrp *str = buf;
701 	u32 rg_flags;
702 
703 	rg_flags = be32_to_cpu(str->rg_flags);
704 	if (rg_flags & GFS2_RGF_NOALLOC)
705 		rgd->rd_flags |= GFS2_RDF_NOALLOC;
706 	else
707 		rgd->rd_flags &= ~GFS2_RDF_NOALLOC;
708 	rgd->rd_free = be32_to_cpu(str->rg_free);
709 	rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes);
710 	rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration);
711 }
712 
713 static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf)
714 {
715 	struct gfs2_rgrp *str = buf;
716 	u32 rg_flags = 0;
717 
718 	if (rgd->rd_flags & GFS2_RDF_NOALLOC)
719 		rg_flags |= GFS2_RGF_NOALLOC;
720 	str->rg_flags = cpu_to_be32(rg_flags);
721 	str->rg_free = cpu_to_be32(rgd->rd_free);
722 	str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes);
723 	str->__pad = cpu_to_be32(0);
724 	str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration);
725 	memset(&str->rg_reserved, 0, sizeof(str->rg_reserved));
726 }
727 
728 /**
729  * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps
730  * @rgd: the struct gfs2_rgrpd describing the RG to read in
731  *
732  * Read in all of a Resource Group's header and bitmap blocks.
733  * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps.
734  *
735  * Returns: errno
736  */
737 
738 int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd)
739 {
740 	struct gfs2_sbd *sdp = rgd->rd_sbd;
741 	struct gfs2_glock *gl = rgd->rd_gl;
742 	unsigned int length = rgd->rd_length;
743 	struct gfs2_bitmap *bi;
744 	unsigned int x, y;
745 	int error;
746 
747 	mutex_lock(&rgd->rd_mutex);
748 
749 	spin_lock(&sdp->sd_rindex_spin);
750 	if (rgd->rd_bh_count) {
751 		rgd->rd_bh_count++;
752 		spin_unlock(&sdp->sd_rindex_spin);
753 		mutex_unlock(&rgd->rd_mutex);
754 		return 0;
755 	}
756 	spin_unlock(&sdp->sd_rindex_spin);
757 
758 	for (x = 0; x < length; x++) {
759 		bi = rgd->rd_bits + x;
760 		error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, &bi->bi_bh);
761 		if (error)
762 			goto fail;
763 	}
764 
765 	for (y = length; y--;) {
766 		bi = rgd->rd_bits + y;
767 		error = gfs2_meta_wait(sdp, bi->bi_bh);
768 		if (error)
769 			goto fail;
770 		if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB :
771 					      GFS2_METATYPE_RG)) {
772 			error = -EIO;
773 			goto fail;
774 		}
775 	}
776 
777 	if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) {
778 		gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data);
779 		rgd->rd_flags |= GFS2_RDF_UPTODATE;
780 	}
781 
782 	spin_lock(&sdp->sd_rindex_spin);
783 	rgd->rd_free_clone = rgd->rd_free;
784 	rgd->rd_bh_count++;
785 	spin_unlock(&sdp->sd_rindex_spin);
786 
787 	mutex_unlock(&rgd->rd_mutex);
788 
789 	return 0;
790 
791 fail:
792 	while (x--) {
793 		bi = rgd->rd_bits + x;
794 		brelse(bi->bi_bh);
795 		bi->bi_bh = NULL;
796 		gfs2_assert_warn(sdp, !bi->bi_clone);
797 	}
798 	mutex_unlock(&rgd->rd_mutex);
799 
800 	return error;
801 }
802 
803 void gfs2_rgrp_bh_hold(struct gfs2_rgrpd *rgd)
804 {
805 	struct gfs2_sbd *sdp = rgd->rd_sbd;
806 
807 	spin_lock(&sdp->sd_rindex_spin);
808 	gfs2_assert_warn(rgd->rd_sbd, rgd->rd_bh_count);
809 	rgd->rd_bh_count++;
810 	spin_unlock(&sdp->sd_rindex_spin);
811 }
812 
813 /**
814  * gfs2_rgrp_bh_put - Release RG bitmaps read in with gfs2_rgrp_bh_get()
815  * @rgd: the struct gfs2_rgrpd describing the RG to read in
816  *
817  */
818 
819 void gfs2_rgrp_bh_put(struct gfs2_rgrpd *rgd)
820 {
821 	struct gfs2_sbd *sdp = rgd->rd_sbd;
822 	int x, length = rgd->rd_length;
823 
824 	spin_lock(&sdp->sd_rindex_spin);
825 	gfs2_assert_warn(rgd->rd_sbd, rgd->rd_bh_count);
826 	if (--rgd->rd_bh_count) {
827 		spin_unlock(&sdp->sd_rindex_spin);
828 		return;
829 	}
830 
831 	for (x = 0; x < length; x++) {
832 		struct gfs2_bitmap *bi = rgd->rd_bits + x;
833 		kfree(bi->bi_clone);
834 		bi->bi_clone = NULL;
835 		brelse(bi->bi_bh);
836 		bi->bi_bh = NULL;
837 	}
838 
839 	spin_unlock(&sdp->sd_rindex_spin);
840 }
841 
842 static void gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset,
843 				    const struct gfs2_bitmap *bi)
844 {
845 	struct super_block *sb = sdp->sd_vfs;
846 	struct block_device *bdev = sb->s_bdev;
847 	const unsigned int sects_per_blk = sdp->sd_sb.sb_bsize /
848 					   bdev_hardsect_size(sb->s_bdev);
849 	u64 blk;
850 	sector_t start = 0;
851 	sector_t nr_sects = 0;
852 	int rv;
853 	unsigned int x;
854 
855 	for (x = 0; x < bi->bi_len; x++) {
856 		const u8 *orig = bi->bi_bh->b_data + bi->bi_offset + x;
857 		const u8 *clone = bi->bi_clone + bi->bi_offset + x;
858 		u8 diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1));
859 		diff &= 0x55;
860 		if (diff == 0)
861 			continue;
862 		blk = offset + ((bi->bi_start + x) * GFS2_NBBY);
863 		blk *= sects_per_blk; /* convert to sectors */
864 		while(diff) {
865 			if (diff & 1) {
866 				if (nr_sects == 0)
867 					goto start_new_extent;
868 				if ((start + nr_sects) != blk) {
869 					rv = blkdev_issue_discard(bdev, start,
870 							    nr_sects, GFP_NOFS);
871 					if (rv)
872 						goto fail;
873 					nr_sects = 0;
874 start_new_extent:
875 					start = blk;
876 				}
877 				nr_sects += sects_per_blk;
878 			}
879 			diff >>= 2;
880 			blk += sects_per_blk;
881 		}
882 	}
883 	if (nr_sects) {
884 		rv = blkdev_issue_discard(bdev, start, nr_sects, GFP_NOFS);
885 		if (rv)
886 			goto fail;
887 	}
888 	return;
889 fail:
890 	fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv);
891 	sdp->sd_args.ar_discard = 0;
892 }
893 
894 void gfs2_rgrp_repolish_clones(struct gfs2_rgrpd *rgd)
895 {
896 	struct gfs2_sbd *sdp = rgd->rd_sbd;
897 	unsigned int length = rgd->rd_length;
898 	unsigned int x;
899 
900 	for (x = 0; x < length; x++) {
901 		struct gfs2_bitmap *bi = rgd->rd_bits + x;
902 		if (!bi->bi_clone)
903 			continue;
904 		if (sdp->sd_args.ar_discard)
905 			gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bi);
906 		memcpy(bi->bi_clone + bi->bi_offset,
907 		       bi->bi_bh->b_data + bi->bi_offset, bi->bi_len);
908 	}
909 
910 	spin_lock(&sdp->sd_rindex_spin);
911 	rgd->rd_free_clone = rgd->rd_free;
912 	spin_unlock(&sdp->sd_rindex_spin);
913 }
914 
915 /**
916  * gfs2_alloc_get - get the struct gfs2_alloc structure for an inode
917  * @ip: the incore GFS2 inode structure
918  *
919  * Returns: the struct gfs2_alloc
920  */
921 
922 struct gfs2_alloc *gfs2_alloc_get(struct gfs2_inode *ip)
923 {
924 	BUG_ON(ip->i_alloc != NULL);
925 	ip->i_alloc = kzalloc(sizeof(struct gfs2_alloc), GFP_KERNEL);
926 	return ip->i_alloc;
927 }
928 
929 /**
930  * try_rgrp_fit - See if a given reservation will fit in a given RG
931  * @rgd: the RG data
932  * @al: the struct gfs2_alloc structure describing the reservation
933  *
934  * If there's room for the requested blocks to be allocated from the RG:
935  *   Sets the $al_rgd field in @al.
936  *
937  * Returns: 1 on success (it fits), 0 on failure (it doesn't fit)
938  */
939 
940 static int try_rgrp_fit(struct gfs2_rgrpd *rgd, struct gfs2_alloc *al)
941 {
942 	struct gfs2_sbd *sdp = rgd->rd_sbd;
943 	int ret = 0;
944 
945 	if (rgd->rd_flags & GFS2_RDF_NOALLOC)
946 		return 0;
947 
948 	spin_lock(&sdp->sd_rindex_spin);
949 	if (rgd->rd_free_clone >= al->al_requested) {
950 		al->al_rgd = rgd;
951 		ret = 1;
952 	}
953 	spin_unlock(&sdp->sd_rindex_spin);
954 
955 	return ret;
956 }
957 
958 /**
959  * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes
960  * @rgd: The rgrp
961  *
962  * Returns: The inode, if one has been found
963  */
964 
965 static struct inode *try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked)
966 {
967 	struct inode *inode;
968 	u32 goal = 0, block;
969 	u64 no_addr;
970 	struct gfs2_sbd *sdp = rgd->rd_sbd;
971 	unsigned int n;
972 
973 	for(;;) {
974 		if (goal >= rgd->rd_data)
975 			break;
976 		down_write(&sdp->sd_log_flush_lock);
977 		n = 1;
978 		block = rgblk_search(rgd, goal, GFS2_BLKST_UNLINKED,
979 				     GFS2_BLKST_UNLINKED, &n);
980 		up_write(&sdp->sd_log_flush_lock);
981 		if (block == BFITNOENT)
982 			break;
983 		/* rgblk_search can return a block < goal, so we need to
984 		   keep it marching forward. */
985 		no_addr = block + rgd->rd_data0;
986 		goal++;
987 		if (*last_unlinked != NO_BLOCK && no_addr <= *last_unlinked)
988 			continue;
989 		*last_unlinked = no_addr;
990 		inode = gfs2_inode_lookup(rgd->rd_sbd->sd_vfs, DT_UNKNOWN,
991 					  no_addr, -1, 1);
992 		if (!IS_ERR(inode))
993 			return inode;
994 	}
995 
996 	rgd->rd_flags &= ~GFS2_RDF_CHECK;
997 	return NULL;
998 }
999 
1000 /**
1001  * recent_rgrp_next - get next RG from "recent" list
1002  * @cur_rgd: current rgrp
1003  *
1004  * Returns: The next rgrp in the recent list
1005  */
1006 
1007 static struct gfs2_rgrpd *recent_rgrp_next(struct gfs2_rgrpd *cur_rgd)
1008 {
1009 	struct gfs2_sbd *sdp = cur_rgd->rd_sbd;
1010 	struct list_head *head;
1011 	struct gfs2_rgrpd *rgd;
1012 
1013 	spin_lock(&sdp->sd_rindex_spin);
1014 	head = &sdp->sd_rindex_mru_list;
1015 	if (unlikely(cur_rgd->rd_list_mru.next == head)) {
1016 		spin_unlock(&sdp->sd_rindex_spin);
1017 		return NULL;
1018 	}
1019 	rgd = list_entry(cur_rgd->rd_list_mru.next, struct gfs2_rgrpd, rd_list_mru);
1020 	spin_unlock(&sdp->sd_rindex_spin);
1021 	return rgd;
1022 }
1023 
1024 /**
1025  * forward_rgrp_get - get an rgrp to try next from full list
1026  * @sdp: The GFS2 superblock
1027  *
1028  * Returns: The rgrp to try next
1029  */
1030 
1031 static struct gfs2_rgrpd *forward_rgrp_get(struct gfs2_sbd *sdp)
1032 {
1033 	struct gfs2_rgrpd *rgd;
1034 	unsigned int journals = gfs2_jindex_size(sdp);
1035 	unsigned int rg = 0, x;
1036 
1037 	spin_lock(&sdp->sd_rindex_spin);
1038 
1039 	rgd = sdp->sd_rindex_forward;
1040 	if (!rgd) {
1041 		if (sdp->sd_rgrps >= journals)
1042 			rg = sdp->sd_rgrps * sdp->sd_jdesc->jd_jid / journals;
1043 
1044 		for (x = 0, rgd = gfs2_rgrpd_get_first(sdp); x < rg;
1045 		     x++, rgd = gfs2_rgrpd_get_next(rgd))
1046 			/* Do Nothing */;
1047 
1048 		sdp->sd_rindex_forward = rgd;
1049 	}
1050 
1051 	spin_unlock(&sdp->sd_rindex_spin);
1052 
1053 	return rgd;
1054 }
1055 
1056 /**
1057  * forward_rgrp_set - set the forward rgrp pointer
1058  * @sdp: the filesystem
1059  * @rgd: The new forward rgrp
1060  *
1061  */
1062 
1063 static void forward_rgrp_set(struct gfs2_sbd *sdp, struct gfs2_rgrpd *rgd)
1064 {
1065 	spin_lock(&sdp->sd_rindex_spin);
1066 	sdp->sd_rindex_forward = rgd;
1067 	spin_unlock(&sdp->sd_rindex_spin);
1068 }
1069 
1070 /**
1071  * get_local_rgrp - Choose and lock a rgrp for allocation
1072  * @ip: the inode to reserve space for
1073  * @rgp: the chosen and locked rgrp
1074  *
1075  * Try to acquire rgrp in way which avoids contending with others.
1076  *
1077  * Returns: errno
1078  */
1079 
1080 static struct inode *get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked)
1081 {
1082 	struct inode *inode = NULL;
1083 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1084 	struct gfs2_rgrpd *rgd, *begin = NULL;
1085 	struct gfs2_alloc *al = ip->i_alloc;
1086 	int flags = LM_FLAG_TRY;
1087 	int skipped = 0;
1088 	int loops = 0;
1089 	int error, rg_locked;
1090 
1091 	rgd = gfs2_blk2rgrpd(sdp, ip->i_goal);
1092 
1093 	while (rgd) {
1094 		rg_locked = 0;
1095 
1096 		if (gfs2_glock_is_locked_by_me(rgd->rd_gl)) {
1097 			rg_locked = 1;
1098 			error = 0;
1099 		} else {
1100 			error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1101 						   LM_FLAG_TRY, &al->al_rgd_gh);
1102 		}
1103 		switch (error) {
1104 		case 0:
1105 			if (try_rgrp_fit(rgd, al))
1106 				goto out;
1107 			if (rgd->rd_flags & GFS2_RDF_CHECK)
1108 				inode = try_rgrp_unlink(rgd, last_unlinked);
1109 			if (!rg_locked)
1110 				gfs2_glock_dq_uninit(&al->al_rgd_gh);
1111 			if (inode)
1112 				return inode;
1113 			/* fall through */
1114 		case GLR_TRYFAILED:
1115 			rgd = recent_rgrp_next(rgd);
1116 			break;
1117 
1118 		default:
1119 			return ERR_PTR(error);
1120 		}
1121 	}
1122 
1123 	/* Go through full list of rgrps */
1124 
1125 	begin = rgd = forward_rgrp_get(sdp);
1126 
1127 	for (;;) {
1128 		rg_locked = 0;
1129 
1130 		if (gfs2_glock_is_locked_by_me(rgd->rd_gl)) {
1131 			rg_locked = 1;
1132 			error = 0;
1133 		} else {
1134 			error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, flags,
1135 						   &al->al_rgd_gh);
1136 		}
1137 		switch (error) {
1138 		case 0:
1139 			if (try_rgrp_fit(rgd, al))
1140 				goto out;
1141 			if (rgd->rd_flags & GFS2_RDF_CHECK)
1142 				inode = try_rgrp_unlink(rgd, last_unlinked);
1143 			if (!rg_locked)
1144 				gfs2_glock_dq_uninit(&al->al_rgd_gh);
1145 			if (inode)
1146 				return inode;
1147 			break;
1148 
1149 		case GLR_TRYFAILED:
1150 			skipped++;
1151 			break;
1152 
1153 		default:
1154 			return ERR_PTR(error);
1155 		}
1156 
1157 		rgd = gfs2_rgrpd_get_next(rgd);
1158 		if (!rgd)
1159 			rgd = gfs2_rgrpd_get_first(sdp);
1160 
1161 		if (rgd == begin) {
1162 			if (++loops >= 3)
1163 				return ERR_PTR(-ENOSPC);
1164 			if (!skipped)
1165 				loops++;
1166 			flags = 0;
1167 			if (loops == 2)
1168 				gfs2_log_flush(sdp, NULL);
1169 		}
1170 	}
1171 
1172 out:
1173 	if (begin) {
1174 		spin_lock(&sdp->sd_rindex_spin);
1175 		list_move(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list);
1176 		spin_unlock(&sdp->sd_rindex_spin);
1177 		rgd = gfs2_rgrpd_get_next(rgd);
1178 		if (!rgd)
1179 			rgd = gfs2_rgrpd_get_first(sdp);
1180 		forward_rgrp_set(sdp, rgd);
1181 	}
1182 
1183 	return NULL;
1184 }
1185 
1186 /**
1187  * gfs2_inplace_reserve_i - Reserve space in the filesystem
1188  * @ip: the inode to reserve space for
1189  *
1190  * Returns: errno
1191  */
1192 
1193 int gfs2_inplace_reserve_i(struct gfs2_inode *ip, char *file, unsigned int line)
1194 {
1195 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1196 	struct gfs2_alloc *al = ip->i_alloc;
1197 	struct inode *inode;
1198 	int error = 0;
1199 	u64 last_unlinked = NO_BLOCK;
1200 
1201 	if (gfs2_assert_warn(sdp, al->al_requested))
1202 		return -EINVAL;
1203 
1204 try_again:
1205 	/* We need to hold the rindex unless the inode we're using is
1206 	   the rindex itself, in which case it's already held. */
1207 	if (ip != GFS2_I(sdp->sd_rindex))
1208 		error = gfs2_rindex_hold(sdp, &al->al_ri_gh);
1209 	else if (!sdp->sd_rgrps) /* We may not have the rindex read in, so: */
1210 		error = gfs2_ri_update_special(ip);
1211 
1212 	if (error)
1213 		return error;
1214 
1215 	inode = get_local_rgrp(ip, &last_unlinked);
1216 	if (inode) {
1217 		if (ip != GFS2_I(sdp->sd_rindex))
1218 			gfs2_glock_dq_uninit(&al->al_ri_gh);
1219 		if (IS_ERR(inode))
1220 			return PTR_ERR(inode);
1221 		iput(inode);
1222 		gfs2_log_flush(sdp, NULL);
1223 		goto try_again;
1224 	}
1225 
1226 	al->al_file = file;
1227 	al->al_line = line;
1228 
1229 	return 0;
1230 }
1231 
1232 /**
1233  * gfs2_inplace_release - release an inplace reservation
1234  * @ip: the inode the reservation was taken out on
1235  *
1236  * Release a reservation made by gfs2_inplace_reserve().
1237  */
1238 
1239 void gfs2_inplace_release(struct gfs2_inode *ip)
1240 {
1241 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1242 	struct gfs2_alloc *al = ip->i_alloc;
1243 
1244 	if (gfs2_assert_warn(sdp, al->al_alloced <= al->al_requested) == -1)
1245 		fs_warn(sdp, "al_alloced = %u, al_requested = %u "
1246 			     "al_file = %s, al_line = %u\n",
1247 		             al->al_alloced, al->al_requested, al->al_file,
1248 			     al->al_line);
1249 
1250 	al->al_rgd = NULL;
1251 	if (al->al_rgd_gh.gh_gl)
1252 		gfs2_glock_dq_uninit(&al->al_rgd_gh);
1253 	if (ip != GFS2_I(sdp->sd_rindex))
1254 		gfs2_glock_dq_uninit(&al->al_ri_gh);
1255 }
1256 
1257 /**
1258  * gfs2_get_block_type - Check a block in a RG is of given type
1259  * @rgd: the resource group holding the block
1260  * @block: the block number
1261  *
1262  * Returns: The block type (GFS2_BLKST_*)
1263  */
1264 
1265 unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block)
1266 {
1267 	struct gfs2_bitmap *bi = NULL;
1268 	u32 length, rgrp_block, buf_block;
1269 	unsigned int buf;
1270 	unsigned char type;
1271 
1272 	length = rgd->rd_length;
1273 	rgrp_block = block - rgd->rd_data0;
1274 
1275 	for (buf = 0; buf < length; buf++) {
1276 		bi = rgd->rd_bits + buf;
1277 		if (rgrp_block < (bi->bi_start + bi->bi_len) * GFS2_NBBY)
1278 			break;
1279 	}
1280 
1281 	gfs2_assert(rgd->rd_sbd, buf < length);
1282 	buf_block = rgrp_block - bi->bi_start * GFS2_NBBY;
1283 
1284 	type = gfs2_testbit(rgd, bi->bi_bh->b_data + bi->bi_offset,
1285 			   bi->bi_len, buf_block);
1286 
1287 	return type;
1288 }
1289 
1290 /**
1291  * rgblk_search - find a block in @old_state, change allocation
1292  *           state to @new_state
1293  * @rgd: the resource group descriptor
1294  * @goal: the goal block within the RG (start here to search for avail block)
1295  * @old_state: GFS2_BLKST_XXX the before-allocation state to find
1296  * @new_state: GFS2_BLKST_XXX the after-allocation block state
1297  * @n: The extent length
1298  *
1299  * Walk rgrp's bitmap to find bits that represent a block in @old_state.
1300  * Add the found bitmap buffer to the transaction.
1301  * Set the found bits to @new_state to change block's allocation state.
1302  *
1303  * This function never fails, because we wouldn't call it unless we
1304  * know (from reservation results, etc.) that a block is available.
1305  *
1306  * Scope of @goal and returned block is just within rgrp, not the whole
1307  * filesystem.
1308  *
1309  * Returns:  the block number allocated
1310  */
1311 
1312 static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal,
1313 			unsigned char old_state, unsigned char new_state,
1314 			unsigned int *n)
1315 {
1316 	struct gfs2_bitmap *bi = NULL;
1317 	const u32 length = rgd->rd_length;
1318 	u32 blk = 0;
1319 	unsigned int buf, x;
1320 	const unsigned int elen = *n;
1321 	const u8 *buffer;
1322 
1323 	*n = 0;
1324 	/* Find bitmap block that contains bits for goal block */
1325 	for (buf = 0; buf < length; buf++) {
1326 		bi = rgd->rd_bits + buf;
1327 		if (goal < (bi->bi_start + bi->bi_len) * GFS2_NBBY)
1328 			break;
1329 	}
1330 
1331 	gfs2_assert(rgd->rd_sbd, buf < length);
1332 
1333 	/* Convert scope of "goal" from rgrp-wide to within found bit block */
1334 	goal -= bi->bi_start * GFS2_NBBY;
1335 
1336 	/* Search (up to entire) bitmap in this rgrp for allocatable block.
1337 	   "x <= length", instead of "x < length", because we typically start
1338 	   the search in the middle of a bit block, but if we can't find an
1339 	   allocatable block anywhere else, we want to be able wrap around and
1340 	   search in the first part of our first-searched bit block.  */
1341 	for (x = 0; x <= length; x++) {
1342 		/* The GFS2_BLKST_UNLINKED state doesn't apply to the clone
1343 		   bitmaps, so we must search the originals for that. */
1344 		buffer = bi->bi_bh->b_data + bi->bi_offset;
1345 		if (old_state != GFS2_BLKST_UNLINKED && bi->bi_clone)
1346 			buffer = bi->bi_clone + bi->bi_offset;
1347 
1348 		blk = gfs2_bitfit(buffer, bi->bi_len, goal, old_state);
1349 		if (blk != BFITNOENT)
1350 			break;
1351 
1352 		/* Try next bitmap block (wrap back to rgrp header if at end) */
1353 		buf = (buf + 1) % length;
1354 		bi = rgd->rd_bits + buf;
1355 		goal = 0;
1356 	}
1357 
1358 	if (blk != BFITNOENT && old_state != new_state) {
1359 		*n = 1;
1360 		gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1);
1361 		gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, bi->bi_offset,
1362 			    bi->bi_len, blk, new_state);
1363 		goal = blk;
1364 		while (*n < elen) {
1365 			goal++;
1366 			if (goal >= (bi->bi_len * GFS2_NBBY))
1367 				break;
1368 			if (gfs2_testbit(rgd, buffer, bi->bi_len, goal) !=
1369 			    GFS2_BLKST_FREE)
1370 				break;
1371 			gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone,
1372 				    bi->bi_offset, bi->bi_len, goal,
1373 				    new_state);
1374 			(*n)++;
1375 		}
1376 	}
1377 
1378 	return (blk == BFITNOENT) ? blk : (bi->bi_start * GFS2_NBBY) + blk;
1379 }
1380 
1381 /**
1382  * rgblk_free - Change alloc state of given block(s)
1383  * @sdp: the filesystem
1384  * @bstart: the start of a run of blocks to free
1385  * @blen: the length of the block run (all must lie within ONE RG!)
1386  * @new_state: GFS2_BLKST_XXX the after-allocation block state
1387  *
1388  * Returns:  Resource group containing the block(s)
1389  */
1390 
1391 static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart,
1392 				     u32 blen, unsigned char new_state)
1393 {
1394 	struct gfs2_rgrpd *rgd;
1395 	struct gfs2_bitmap *bi = NULL;
1396 	u32 length, rgrp_blk, buf_blk;
1397 	unsigned int buf;
1398 
1399 	rgd = gfs2_blk2rgrpd(sdp, bstart);
1400 	if (!rgd) {
1401 		if (gfs2_consist(sdp))
1402 			fs_err(sdp, "block = %llu\n", (unsigned long long)bstart);
1403 		return NULL;
1404 	}
1405 
1406 	length = rgd->rd_length;
1407 
1408 	rgrp_blk = bstart - rgd->rd_data0;
1409 
1410 	while (blen--) {
1411 		for (buf = 0; buf < length; buf++) {
1412 			bi = rgd->rd_bits + buf;
1413 			if (rgrp_blk < (bi->bi_start + bi->bi_len) * GFS2_NBBY)
1414 				break;
1415 		}
1416 
1417 		gfs2_assert(rgd->rd_sbd, buf < length);
1418 
1419 		buf_blk = rgrp_blk - bi->bi_start * GFS2_NBBY;
1420 		rgrp_blk++;
1421 
1422 		if (!bi->bi_clone) {
1423 			bi->bi_clone = kmalloc(bi->bi_bh->b_size,
1424 					       GFP_NOFS | __GFP_NOFAIL);
1425 			memcpy(bi->bi_clone + bi->bi_offset,
1426 			       bi->bi_bh->b_data + bi->bi_offset,
1427 			       bi->bi_len);
1428 		}
1429 		gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1);
1430 		gfs2_setbit(rgd, bi->bi_bh->b_data, NULL, bi->bi_offset,
1431 			    bi->bi_len, buf_blk, new_state);
1432 	}
1433 
1434 	return rgd;
1435 }
1436 
1437 /**
1438  * gfs2_alloc_block - Allocate a block
1439  * @ip: the inode to allocate the block for
1440  *
1441  * Returns: the allocated block
1442  */
1443 
1444 u64 gfs2_alloc_block(struct gfs2_inode *ip, unsigned int *n)
1445 {
1446 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1447 	struct buffer_head *dibh;
1448 	struct gfs2_alloc *al = ip->i_alloc;
1449 	struct gfs2_rgrpd *rgd = al->al_rgd;
1450 	u32 goal, blk;
1451 	u64 block;
1452 	int error;
1453 
1454 	if (rgrp_contains_block(rgd, ip->i_goal))
1455 		goal = ip->i_goal - rgd->rd_data0;
1456 	else
1457 		goal = rgd->rd_last_alloc;
1458 
1459 	blk = rgblk_search(rgd, goal, GFS2_BLKST_FREE, GFS2_BLKST_USED, n);
1460 	BUG_ON(blk == BFITNOENT);
1461 
1462 	rgd->rd_last_alloc = blk;
1463 	block = rgd->rd_data0 + blk;
1464 	ip->i_goal = block;
1465 	error = gfs2_meta_inode_buffer(ip, &dibh);
1466 	if (error == 0) {
1467 		struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
1468 		gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1469 		di->di_goal_meta = di->di_goal_data = cpu_to_be64(ip->i_goal);
1470 		brelse(dibh);
1471 	}
1472 	gfs2_assert_withdraw(sdp, rgd->rd_free >= *n);
1473 	rgd->rd_free -= *n;
1474 
1475 	gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1476 	gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1477 
1478 	al->al_alloced += *n;
1479 
1480 	gfs2_statfs_change(sdp, 0, -(s64)*n, 0);
1481 	gfs2_quota_change(ip, *n, ip->i_inode.i_uid, ip->i_inode.i_gid);
1482 
1483 	spin_lock(&sdp->sd_rindex_spin);
1484 	rgd->rd_free_clone -= *n;
1485 	spin_unlock(&sdp->sd_rindex_spin);
1486 
1487 	return block;
1488 }
1489 
1490 /**
1491  * gfs2_alloc_di - Allocate a dinode
1492  * @dip: the directory that the inode is going in
1493  *
1494  * Returns: the block allocated
1495  */
1496 
1497 u64 gfs2_alloc_di(struct gfs2_inode *dip, u64 *generation)
1498 {
1499 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1500 	struct gfs2_alloc *al = dip->i_alloc;
1501 	struct gfs2_rgrpd *rgd = al->al_rgd;
1502 	u32 blk;
1503 	u64 block;
1504 	unsigned int n = 1;
1505 
1506 	blk = rgblk_search(rgd, rgd->rd_last_alloc,
1507 			   GFS2_BLKST_FREE, GFS2_BLKST_DINODE, &n);
1508 	BUG_ON(blk == BFITNOENT);
1509 
1510 	rgd->rd_last_alloc = blk;
1511 
1512 	block = rgd->rd_data0 + blk;
1513 
1514 	gfs2_assert_withdraw(sdp, rgd->rd_free);
1515 	rgd->rd_free--;
1516 	rgd->rd_dinodes++;
1517 	*generation = rgd->rd_igeneration++;
1518 	gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1519 	gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1520 
1521 	al->al_alloced++;
1522 
1523 	gfs2_statfs_change(sdp, 0, -1, +1);
1524 	gfs2_trans_add_unrevoke(sdp, block, 1);
1525 
1526 	spin_lock(&sdp->sd_rindex_spin);
1527 	rgd->rd_free_clone--;
1528 	spin_unlock(&sdp->sd_rindex_spin);
1529 
1530 	return block;
1531 }
1532 
1533 /**
1534  * gfs2_free_data - free a contiguous run of data block(s)
1535  * @ip: the inode these blocks are being freed from
1536  * @bstart: first block of a run of contiguous blocks
1537  * @blen: the length of the block run
1538  *
1539  */
1540 
1541 void gfs2_free_data(struct gfs2_inode *ip, u64 bstart, u32 blen)
1542 {
1543 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1544 	struct gfs2_rgrpd *rgd;
1545 
1546 	rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE);
1547 	if (!rgd)
1548 		return;
1549 
1550 	rgd->rd_free += blen;
1551 
1552 	gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1553 	gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1554 
1555 	gfs2_trans_add_rg(rgd);
1556 
1557 	gfs2_statfs_change(sdp, 0, +blen, 0);
1558 	gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid);
1559 }
1560 
1561 /**
1562  * gfs2_free_meta - free a contiguous run of data block(s)
1563  * @ip: the inode these blocks are being freed from
1564  * @bstart: first block of a run of contiguous blocks
1565  * @blen: the length of the block run
1566  *
1567  */
1568 
1569 void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen)
1570 {
1571 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1572 	struct gfs2_rgrpd *rgd;
1573 
1574 	rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE);
1575 	if (!rgd)
1576 		return;
1577 
1578 	rgd->rd_free += blen;
1579 
1580 	gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1581 	gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1582 
1583 	gfs2_trans_add_rg(rgd);
1584 
1585 	gfs2_statfs_change(sdp, 0, +blen, 0);
1586 	gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid);
1587 	gfs2_meta_wipe(ip, bstart, blen);
1588 }
1589 
1590 void gfs2_unlink_di(struct inode *inode)
1591 {
1592 	struct gfs2_inode *ip = GFS2_I(inode);
1593 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1594 	struct gfs2_rgrpd *rgd;
1595 	u64 blkno = ip->i_no_addr;
1596 
1597 	rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED);
1598 	if (!rgd)
1599 		return;
1600 	gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1601 	gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1602 	gfs2_trans_add_rg(rgd);
1603 }
1604 
1605 static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno)
1606 {
1607 	struct gfs2_sbd *sdp = rgd->rd_sbd;
1608 	struct gfs2_rgrpd *tmp_rgd;
1609 
1610 	tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE);
1611 	if (!tmp_rgd)
1612 		return;
1613 	gfs2_assert_withdraw(sdp, rgd == tmp_rgd);
1614 
1615 	if (!rgd->rd_dinodes)
1616 		gfs2_consist_rgrpd(rgd);
1617 	rgd->rd_dinodes--;
1618 	rgd->rd_free++;
1619 
1620 	gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1621 	gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1622 
1623 	gfs2_statfs_change(sdp, 0, +1, -1);
1624 	gfs2_trans_add_rg(rgd);
1625 }
1626 
1627 
1628 void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip)
1629 {
1630 	gfs2_free_uninit_di(rgd, ip->i_no_addr);
1631 	gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid);
1632 	gfs2_meta_wipe(ip, ip->i_no_addr, 1);
1633 }
1634 
1635 /**
1636  * gfs2_rlist_add - add a RG to a list of RGs
1637  * @sdp: the filesystem
1638  * @rlist: the list of resource groups
1639  * @block: the block
1640  *
1641  * Figure out what RG a block belongs to and add that RG to the list
1642  *
1643  * FIXME: Don't use NOFAIL
1644  *
1645  */
1646 
1647 void gfs2_rlist_add(struct gfs2_sbd *sdp, struct gfs2_rgrp_list *rlist,
1648 		    u64 block)
1649 {
1650 	struct gfs2_rgrpd *rgd;
1651 	struct gfs2_rgrpd **tmp;
1652 	unsigned int new_space;
1653 	unsigned int x;
1654 
1655 	if (gfs2_assert_warn(sdp, !rlist->rl_ghs))
1656 		return;
1657 
1658 	rgd = gfs2_blk2rgrpd(sdp, block);
1659 	if (!rgd) {
1660 		if (gfs2_consist(sdp))
1661 			fs_err(sdp, "block = %llu\n", (unsigned long long)block);
1662 		return;
1663 	}
1664 
1665 	for (x = 0; x < rlist->rl_rgrps; x++)
1666 		if (rlist->rl_rgd[x] == rgd)
1667 			return;
1668 
1669 	if (rlist->rl_rgrps == rlist->rl_space) {
1670 		new_space = rlist->rl_space + 10;
1671 
1672 		tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *),
1673 			      GFP_NOFS | __GFP_NOFAIL);
1674 
1675 		if (rlist->rl_rgd) {
1676 			memcpy(tmp, rlist->rl_rgd,
1677 			       rlist->rl_space * sizeof(struct gfs2_rgrpd *));
1678 			kfree(rlist->rl_rgd);
1679 		}
1680 
1681 		rlist->rl_space = new_space;
1682 		rlist->rl_rgd = tmp;
1683 	}
1684 
1685 	rlist->rl_rgd[rlist->rl_rgrps++] = rgd;
1686 }
1687 
1688 /**
1689  * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate
1690  *      and initialize an array of glock holders for them
1691  * @rlist: the list of resource groups
1692  * @state: the lock state to acquire the RG lock in
1693  * @flags: the modifier flags for the holder structures
1694  *
1695  * FIXME: Don't use NOFAIL
1696  *
1697  */
1698 
1699 void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state)
1700 {
1701 	unsigned int x;
1702 
1703 	rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder),
1704 				GFP_NOFS | __GFP_NOFAIL);
1705 	for (x = 0; x < rlist->rl_rgrps; x++)
1706 		gfs2_holder_init(rlist->rl_rgd[x]->rd_gl,
1707 				state, 0,
1708 				&rlist->rl_ghs[x]);
1709 }
1710 
1711 /**
1712  * gfs2_rlist_free - free a resource group list
1713  * @list: the list of resource groups
1714  *
1715  */
1716 
1717 void gfs2_rlist_free(struct gfs2_rgrp_list *rlist)
1718 {
1719 	unsigned int x;
1720 
1721 	kfree(rlist->rl_rgd);
1722 
1723 	if (rlist->rl_ghs) {
1724 		for (x = 0; x < rlist->rl_rgrps; x++)
1725 			gfs2_holder_uninit(&rlist->rl_ghs[x]);
1726 		kfree(rlist->rl_ghs);
1727 	}
1728 }
1729 
1730