xref: /linux/fs/gfs2/rgrp.c (revision 2277ab4a1df50e05bc732fe9488d4e902bb8399a)
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 "trace_gfs2.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 		bi->bi_flags = 0;
446 		/* small rgrp; bitmap stored completely in header block */
447 		if (length == 1) {
448 			bytes = bytes_left;
449 			bi->bi_offset = sizeof(struct gfs2_rgrp);
450 			bi->bi_start = 0;
451 			bi->bi_len = bytes;
452 		/* header block */
453 		} else if (x == 0) {
454 			bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp);
455 			bi->bi_offset = sizeof(struct gfs2_rgrp);
456 			bi->bi_start = 0;
457 			bi->bi_len = bytes;
458 		/* last block */
459 		} else if (x + 1 == length) {
460 			bytes = bytes_left;
461 			bi->bi_offset = sizeof(struct gfs2_meta_header);
462 			bi->bi_start = rgd->rd_bitbytes - bytes_left;
463 			bi->bi_len = bytes;
464 		/* other blocks */
465 		} else {
466 			bytes = sdp->sd_sb.sb_bsize -
467 				sizeof(struct gfs2_meta_header);
468 			bi->bi_offset = sizeof(struct gfs2_meta_header);
469 			bi->bi_start = rgd->rd_bitbytes - bytes_left;
470 			bi->bi_len = bytes;
471 		}
472 
473 		bytes_left -= bytes;
474 	}
475 
476 	if (bytes_left) {
477 		gfs2_consist_rgrpd(rgd);
478 		return -EIO;
479 	}
480 	bi = rgd->rd_bits + (length - 1);
481 	if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) {
482 		if (gfs2_consist_rgrpd(rgd)) {
483 			gfs2_rindex_print(rgd);
484 			fs_err(sdp, "start=%u len=%u offset=%u\n",
485 			       bi->bi_start, bi->bi_len, bi->bi_offset);
486 		}
487 		return -EIO;
488 	}
489 
490 	return 0;
491 }
492 
493 /**
494  * gfs2_ri_total - Total up the file system space, according to the rindex.
495  *
496  */
497 u64 gfs2_ri_total(struct gfs2_sbd *sdp)
498 {
499 	u64 total_data = 0;
500 	struct inode *inode = sdp->sd_rindex;
501 	struct gfs2_inode *ip = GFS2_I(inode);
502 	char buf[sizeof(struct gfs2_rindex)];
503 	struct file_ra_state ra_state;
504 	int error, rgrps;
505 
506 	mutex_lock(&sdp->sd_rindex_mutex);
507 	file_ra_state_init(&ra_state, inode->i_mapping);
508 	for (rgrps = 0;; rgrps++) {
509 		loff_t pos = rgrps * sizeof(struct gfs2_rindex);
510 
511 		if (pos + sizeof(struct gfs2_rindex) >= ip->i_disksize)
512 			break;
513 		error = gfs2_internal_read(ip, &ra_state, buf, &pos,
514 					   sizeof(struct gfs2_rindex));
515 		if (error != sizeof(struct gfs2_rindex))
516 			break;
517 		total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data);
518 	}
519 	mutex_unlock(&sdp->sd_rindex_mutex);
520 	return total_data;
521 }
522 
523 static void gfs2_rindex_in(struct gfs2_rgrpd *rgd, const void *buf)
524 {
525 	const struct gfs2_rindex *str = buf;
526 
527 	rgd->rd_addr = be64_to_cpu(str->ri_addr);
528 	rgd->rd_length = be32_to_cpu(str->ri_length);
529 	rgd->rd_data0 = be64_to_cpu(str->ri_data0);
530 	rgd->rd_data = be32_to_cpu(str->ri_data);
531 	rgd->rd_bitbytes = be32_to_cpu(str->ri_bitbytes);
532 }
533 
534 /**
535  * read_rindex_entry - Pull in a new resource index entry from the disk
536  * @gl: The glock covering the rindex inode
537  *
538  * Returns: 0 on success, error code otherwise
539  */
540 
541 static int read_rindex_entry(struct gfs2_inode *ip,
542 			     struct file_ra_state *ra_state)
543 {
544 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
545 	loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex);
546 	char buf[sizeof(struct gfs2_rindex)];
547 	int error;
548 	struct gfs2_rgrpd *rgd;
549 
550 	error = gfs2_internal_read(ip, ra_state, buf, &pos,
551 				   sizeof(struct gfs2_rindex));
552 	if (!error)
553 		return 0;
554 	if (error != sizeof(struct gfs2_rindex)) {
555 		if (error > 0)
556 			error = -EIO;
557 		return error;
558 	}
559 
560 	rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS);
561 	error = -ENOMEM;
562 	if (!rgd)
563 		return error;
564 
565 	mutex_init(&rgd->rd_mutex);
566 	lops_init_le(&rgd->rd_le, &gfs2_rg_lops);
567 	rgd->rd_sbd = sdp;
568 
569 	list_add_tail(&rgd->rd_list, &sdp->sd_rindex_list);
570 	list_add_tail(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list);
571 
572 	gfs2_rindex_in(rgd, buf);
573 	error = compute_bitstructs(rgd);
574 	if (error)
575 		return error;
576 
577 	error = gfs2_glock_get(sdp, rgd->rd_addr,
578 			       &gfs2_rgrp_glops, CREATE, &rgd->rd_gl);
579 	if (error)
580 		return error;
581 
582 	rgd->rd_gl->gl_object = rgd;
583 	rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
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 	rg_flags &= ~GFS2_RDF_MASK;
705 	rgd->rd_flags &= GFS2_RDF_MASK;
706 	rgd->rd_flags |= rg_flags;
707 	rgd->rd_free = be32_to_cpu(str->rg_free);
708 	rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes);
709 	rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration);
710 }
711 
712 static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf)
713 {
714 	struct gfs2_rgrp *str = buf;
715 
716 	str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK);
717 	str->rg_free = cpu_to_be32(rgd->rd_free);
718 	str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes);
719 	str->__pad = cpu_to_be32(0);
720 	str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration);
721 	memset(&str->rg_reserved, 0, sizeof(str->rg_reserved));
722 }
723 
724 /**
725  * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps
726  * @rgd: the struct gfs2_rgrpd describing the RG to read in
727  *
728  * Read in all of a Resource Group's header and bitmap blocks.
729  * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps.
730  *
731  * Returns: errno
732  */
733 
734 int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd)
735 {
736 	struct gfs2_sbd *sdp = rgd->rd_sbd;
737 	struct gfs2_glock *gl = rgd->rd_gl;
738 	unsigned int length = rgd->rd_length;
739 	struct gfs2_bitmap *bi;
740 	unsigned int x, y;
741 	int error;
742 
743 	mutex_lock(&rgd->rd_mutex);
744 
745 	spin_lock(&sdp->sd_rindex_spin);
746 	if (rgd->rd_bh_count) {
747 		rgd->rd_bh_count++;
748 		spin_unlock(&sdp->sd_rindex_spin);
749 		mutex_unlock(&rgd->rd_mutex);
750 		return 0;
751 	}
752 	spin_unlock(&sdp->sd_rindex_spin);
753 
754 	for (x = 0; x < length; x++) {
755 		bi = rgd->rd_bits + x;
756 		error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, &bi->bi_bh);
757 		if (error)
758 			goto fail;
759 	}
760 
761 	for (y = length; y--;) {
762 		bi = rgd->rd_bits + y;
763 		error = gfs2_meta_wait(sdp, bi->bi_bh);
764 		if (error)
765 			goto fail;
766 		if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB :
767 					      GFS2_METATYPE_RG)) {
768 			error = -EIO;
769 			goto fail;
770 		}
771 	}
772 
773 	if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) {
774 		for (x = 0; x < length; x++)
775 			clear_bit(GBF_FULL, &rgd->rd_bits[x].bi_flags);
776 		gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data);
777 		rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK);
778 	}
779 
780 	spin_lock(&sdp->sd_rindex_spin);
781 	rgd->rd_free_clone = rgd->rd_free;
782 	rgd->rd_bh_count++;
783 	spin_unlock(&sdp->sd_rindex_spin);
784 
785 	mutex_unlock(&rgd->rd_mutex);
786 
787 	return 0;
788 
789 fail:
790 	while (x--) {
791 		bi = rgd->rd_bits + x;
792 		brelse(bi->bi_bh);
793 		bi->bi_bh = NULL;
794 		gfs2_assert_warn(sdp, !bi->bi_clone);
795 	}
796 	mutex_unlock(&rgd->rd_mutex);
797 
798 	return error;
799 }
800 
801 void gfs2_rgrp_bh_hold(struct gfs2_rgrpd *rgd)
802 {
803 	struct gfs2_sbd *sdp = rgd->rd_sbd;
804 
805 	spin_lock(&sdp->sd_rindex_spin);
806 	gfs2_assert_warn(rgd->rd_sbd, rgd->rd_bh_count);
807 	rgd->rd_bh_count++;
808 	spin_unlock(&sdp->sd_rindex_spin);
809 }
810 
811 /**
812  * gfs2_rgrp_bh_put - Release RG bitmaps read in with gfs2_rgrp_bh_get()
813  * @rgd: the struct gfs2_rgrpd describing the RG to read in
814  *
815  */
816 
817 void gfs2_rgrp_bh_put(struct gfs2_rgrpd *rgd)
818 {
819 	struct gfs2_sbd *sdp = rgd->rd_sbd;
820 	int x, length = rgd->rd_length;
821 
822 	spin_lock(&sdp->sd_rindex_spin);
823 	gfs2_assert_warn(rgd->rd_sbd, rgd->rd_bh_count);
824 	if (--rgd->rd_bh_count) {
825 		spin_unlock(&sdp->sd_rindex_spin);
826 		return;
827 	}
828 
829 	for (x = 0; x < length; x++) {
830 		struct gfs2_bitmap *bi = rgd->rd_bits + x;
831 		kfree(bi->bi_clone);
832 		bi->bi_clone = NULL;
833 		brelse(bi->bi_bh);
834 		bi->bi_bh = NULL;
835 	}
836 
837 	spin_unlock(&sdp->sd_rindex_spin);
838 }
839 
840 static void gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset,
841 				    const struct gfs2_bitmap *bi)
842 {
843 	struct super_block *sb = sdp->sd_vfs;
844 	struct block_device *bdev = sb->s_bdev;
845 	const unsigned int sects_per_blk = sdp->sd_sb.sb_bsize /
846 					   bdev_logical_block_size(sb->s_bdev);
847 	u64 blk;
848 	sector_t start = 0;
849 	sector_t nr_sects = 0;
850 	int rv;
851 	unsigned int x;
852 
853 	for (x = 0; x < bi->bi_len; x++) {
854 		const u8 *orig = bi->bi_bh->b_data + bi->bi_offset + x;
855 		const u8 *clone = bi->bi_clone + bi->bi_offset + x;
856 		u8 diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1));
857 		diff &= 0x55;
858 		if (diff == 0)
859 			continue;
860 		blk = offset + ((bi->bi_start + x) * GFS2_NBBY);
861 		blk *= sects_per_blk; /* convert to sectors */
862 		while(diff) {
863 			if (diff & 1) {
864 				if (nr_sects == 0)
865 					goto start_new_extent;
866 				if ((start + nr_sects) != blk) {
867 					rv = blkdev_issue_discard(bdev, start,
868 							    nr_sects, GFP_NOFS);
869 					if (rv)
870 						goto fail;
871 					nr_sects = 0;
872 start_new_extent:
873 					start = blk;
874 				}
875 				nr_sects += sects_per_blk;
876 			}
877 			diff >>= 2;
878 			blk += sects_per_blk;
879 		}
880 	}
881 	if (nr_sects) {
882 		rv = blkdev_issue_discard(bdev, start, nr_sects, GFP_NOFS);
883 		if (rv)
884 			goto fail;
885 	}
886 	return;
887 fail:
888 	fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv);
889 	sdp->sd_args.ar_discard = 0;
890 }
891 
892 void gfs2_rgrp_repolish_clones(struct gfs2_rgrpd *rgd)
893 {
894 	struct gfs2_sbd *sdp = rgd->rd_sbd;
895 	unsigned int length = rgd->rd_length;
896 	unsigned int x;
897 
898 	for (x = 0; x < length; x++) {
899 		struct gfs2_bitmap *bi = rgd->rd_bits + x;
900 		if (!bi->bi_clone)
901 			continue;
902 		if (sdp->sd_args.ar_discard)
903 			gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bi);
904 		clear_bit(GBF_FULL, &bi->bi_flags);
905 		memcpy(bi->bi_clone + bi->bi_offset,
906 		       bi->bi_bh->b_data + bi->bi_offset, bi->bi_len);
907 	}
908 
909 	spin_lock(&sdp->sd_rindex_spin);
910 	rgd->rd_free_clone = rgd->rd_free;
911 	spin_unlock(&sdp->sd_rindex_spin);
912 }
913 
914 /**
915  * gfs2_alloc_get - get the struct gfs2_alloc structure for an inode
916  * @ip: the incore GFS2 inode structure
917  *
918  * Returns: the struct gfs2_alloc
919  */
920 
921 struct gfs2_alloc *gfs2_alloc_get(struct gfs2_inode *ip)
922 {
923 	BUG_ON(ip->i_alloc != NULL);
924 	ip->i_alloc = kzalloc(sizeof(struct gfs2_alloc), GFP_KERNEL);
925 	return ip->i_alloc;
926 }
927 
928 /**
929  * try_rgrp_fit - See if a given reservation will fit in a given RG
930  * @rgd: the RG data
931  * @al: the struct gfs2_alloc structure describing the reservation
932  *
933  * If there's room for the requested blocks to be allocated from the RG:
934  *   Sets the $al_rgd field in @al.
935  *
936  * Returns: 1 on success (it fits), 0 on failure (it doesn't fit)
937  */
938 
939 static int try_rgrp_fit(struct gfs2_rgrpd *rgd, struct gfs2_alloc *al)
940 {
941 	struct gfs2_sbd *sdp = rgd->rd_sbd;
942 	int ret = 0;
943 
944 	if (rgd->rd_flags & (GFS2_RGF_NOALLOC | GFS2_RDF_ERROR))
945 		return 0;
946 
947 	spin_lock(&sdp->sd_rindex_spin);
948 	if (rgd->rd_free_clone >= al->al_requested) {
949 		al->al_rgd = rgd;
950 		ret = 1;
951 	}
952 	spin_unlock(&sdp->sd_rindex_spin);
953 
954 	return ret;
955 }
956 
957 /**
958  * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes
959  * @rgd: The rgrp
960  *
961  * Returns: The inode, if one has been found
962  */
963 
964 static struct inode *try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked)
965 {
966 	struct inode *inode;
967 	u32 goal = 0, block;
968 	u64 no_addr;
969 	struct gfs2_sbd *sdp = rgd->rd_sbd;
970 	unsigned int n;
971 
972 	for(;;) {
973 		if (goal >= rgd->rd_data)
974 			break;
975 		down_write(&sdp->sd_log_flush_lock);
976 		n = 1;
977 		block = rgblk_search(rgd, goal, GFS2_BLKST_UNLINKED,
978 				     GFS2_BLKST_UNLINKED, &n);
979 		up_write(&sdp->sd_log_flush_lock);
980 		if (block == BFITNOENT)
981 			break;
982 		/* rgblk_search can return a block < goal, so we need to
983 		   keep it marching forward. */
984 		no_addr = block + rgd->rd_data0;
985 		goal++;
986 		if (*last_unlinked != NO_BLOCK && no_addr <= *last_unlinked)
987 			continue;
988 		*last_unlinked = no_addr;
989 		inode = gfs2_inode_lookup(rgd->rd_sbd->sd_vfs, DT_UNKNOWN,
990 					  no_addr, -1, 1);
991 		if (!IS_ERR(inode))
992 			return inode;
993 	}
994 
995 	rgd->rd_flags &= ~GFS2_RDF_CHECK;
996 	return NULL;
997 }
998 
999 /**
1000  * recent_rgrp_next - get next RG from "recent" list
1001  * @cur_rgd: current rgrp
1002  *
1003  * Returns: The next rgrp in the recent list
1004  */
1005 
1006 static struct gfs2_rgrpd *recent_rgrp_next(struct gfs2_rgrpd *cur_rgd)
1007 {
1008 	struct gfs2_sbd *sdp = cur_rgd->rd_sbd;
1009 	struct list_head *head;
1010 	struct gfs2_rgrpd *rgd;
1011 
1012 	spin_lock(&sdp->sd_rindex_spin);
1013 	head = &sdp->sd_rindex_mru_list;
1014 	if (unlikely(cur_rgd->rd_list_mru.next == head)) {
1015 		spin_unlock(&sdp->sd_rindex_spin);
1016 		return NULL;
1017 	}
1018 	rgd = list_entry(cur_rgd->rd_list_mru.next, struct gfs2_rgrpd, rd_list_mru);
1019 	spin_unlock(&sdp->sd_rindex_spin);
1020 	return rgd;
1021 }
1022 
1023 /**
1024  * forward_rgrp_get - get an rgrp to try next from full list
1025  * @sdp: The GFS2 superblock
1026  *
1027  * Returns: The rgrp to try next
1028  */
1029 
1030 static struct gfs2_rgrpd *forward_rgrp_get(struct gfs2_sbd *sdp)
1031 {
1032 	struct gfs2_rgrpd *rgd;
1033 	unsigned int journals = gfs2_jindex_size(sdp);
1034 	unsigned int rg = 0, x;
1035 
1036 	spin_lock(&sdp->sd_rindex_spin);
1037 
1038 	rgd = sdp->sd_rindex_forward;
1039 	if (!rgd) {
1040 		if (sdp->sd_rgrps >= journals)
1041 			rg = sdp->sd_rgrps * sdp->sd_jdesc->jd_jid / journals;
1042 
1043 		for (x = 0, rgd = gfs2_rgrpd_get_first(sdp); x < rg;
1044 		     x++, rgd = gfs2_rgrpd_get_next(rgd))
1045 			/* Do Nothing */;
1046 
1047 		sdp->sd_rindex_forward = rgd;
1048 	}
1049 
1050 	spin_unlock(&sdp->sd_rindex_spin);
1051 
1052 	return rgd;
1053 }
1054 
1055 /**
1056  * forward_rgrp_set - set the forward rgrp pointer
1057  * @sdp: the filesystem
1058  * @rgd: The new forward rgrp
1059  *
1060  */
1061 
1062 static void forward_rgrp_set(struct gfs2_sbd *sdp, struct gfs2_rgrpd *rgd)
1063 {
1064 	spin_lock(&sdp->sd_rindex_spin);
1065 	sdp->sd_rindex_forward = rgd;
1066 	spin_unlock(&sdp->sd_rindex_spin);
1067 }
1068 
1069 /**
1070  * get_local_rgrp - Choose and lock a rgrp for allocation
1071  * @ip: the inode to reserve space for
1072  * @rgp: the chosen and locked rgrp
1073  *
1074  * Try to acquire rgrp in way which avoids contending with others.
1075  *
1076  * Returns: errno
1077  */
1078 
1079 static struct inode *get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked)
1080 {
1081 	struct inode *inode = NULL;
1082 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1083 	struct gfs2_rgrpd *rgd, *begin = NULL;
1084 	struct gfs2_alloc *al = ip->i_alloc;
1085 	int flags = LM_FLAG_TRY;
1086 	int skipped = 0;
1087 	int loops = 0;
1088 	int error, rg_locked;
1089 
1090 	rgd = gfs2_blk2rgrpd(sdp, ip->i_goal);
1091 
1092 	while (rgd) {
1093 		rg_locked = 0;
1094 
1095 		if (gfs2_glock_is_locked_by_me(rgd->rd_gl)) {
1096 			rg_locked = 1;
1097 			error = 0;
1098 		} else {
1099 			error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1100 						   LM_FLAG_TRY, &al->al_rgd_gh);
1101 		}
1102 		switch (error) {
1103 		case 0:
1104 			if (try_rgrp_fit(rgd, al))
1105 				goto out;
1106 			if (rgd->rd_flags & GFS2_RDF_CHECK)
1107 				inode = try_rgrp_unlink(rgd, last_unlinked);
1108 			if (!rg_locked)
1109 				gfs2_glock_dq_uninit(&al->al_rgd_gh);
1110 			if (inode)
1111 				return inode;
1112 			/* fall through */
1113 		case GLR_TRYFAILED:
1114 			rgd = recent_rgrp_next(rgd);
1115 			break;
1116 
1117 		default:
1118 			return ERR_PTR(error);
1119 		}
1120 	}
1121 
1122 	/* Go through full list of rgrps */
1123 
1124 	begin = rgd = forward_rgrp_get(sdp);
1125 
1126 	for (;;) {
1127 		rg_locked = 0;
1128 
1129 		if (gfs2_glock_is_locked_by_me(rgd->rd_gl)) {
1130 			rg_locked = 1;
1131 			error = 0;
1132 		} else {
1133 			error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, flags,
1134 						   &al->al_rgd_gh);
1135 		}
1136 		switch (error) {
1137 		case 0:
1138 			if (try_rgrp_fit(rgd, al))
1139 				goto out;
1140 			if (rgd->rd_flags & GFS2_RDF_CHECK)
1141 				inode = try_rgrp_unlink(rgd, last_unlinked);
1142 			if (!rg_locked)
1143 				gfs2_glock_dq_uninit(&al->al_rgd_gh);
1144 			if (inode)
1145 				return inode;
1146 			break;
1147 
1148 		case GLR_TRYFAILED:
1149 			skipped++;
1150 			break;
1151 
1152 		default:
1153 			return ERR_PTR(error);
1154 		}
1155 
1156 		rgd = gfs2_rgrpd_get_next(rgd);
1157 		if (!rgd)
1158 			rgd = gfs2_rgrpd_get_first(sdp);
1159 
1160 		if (rgd == begin) {
1161 			if (++loops >= 3)
1162 				return ERR_PTR(-ENOSPC);
1163 			if (!skipped)
1164 				loops++;
1165 			flags = 0;
1166 			if (loops == 2)
1167 				gfs2_log_flush(sdp, NULL);
1168 		}
1169 	}
1170 
1171 out:
1172 	if (begin) {
1173 		spin_lock(&sdp->sd_rindex_spin);
1174 		list_move(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list);
1175 		spin_unlock(&sdp->sd_rindex_spin);
1176 		rgd = gfs2_rgrpd_get_next(rgd);
1177 		if (!rgd)
1178 			rgd = gfs2_rgrpd_get_first(sdp);
1179 		forward_rgrp_set(sdp, rgd);
1180 	}
1181 
1182 	return NULL;
1183 }
1184 
1185 /**
1186  * gfs2_inplace_reserve_i - Reserve space in the filesystem
1187  * @ip: the inode to reserve space for
1188  *
1189  * Returns: errno
1190  */
1191 
1192 int gfs2_inplace_reserve_i(struct gfs2_inode *ip, char *file, unsigned int line)
1193 {
1194 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1195 	struct gfs2_alloc *al = ip->i_alloc;
1196 	struct inode *inode;
1197 	int error = 0;
1198 	u64 last_unlinked = NO_BLOCK;
1199 
1200 	if (gfs2_assert_warn(sdp, al->al_requested))
1201 		return -EINVAL;
1202 
1203 try_again:
1204 	/* We need to hold the rindex unless the inode we're using is
1205 	   the rindex itself, in which case it's already held. */
1206 	if (ip != GFS2_I(sdp->sd_rindex))
1207 		error = gfs2_rindex_hold(sdp, &al->al_ri_gh);
1208 	else if (!sdp->sd_rgrps) /* We may not have the rindex read in, so: */
1209 		error = gfs2_ri_update_special(ip);
1210 
1211 	if (error)
1212 		return error;
1213 
1214 	inode = get_local_rgrp(ip, &last_unlinked);
1215 	if (inode) {
1216 		if (ip != GFS2_I(sdp->sd_rindex))
1217 			gfs2_glock_dq_uninit(&al->al_ri_gh);
1218 		if (IS_ERR(inode))
1219 			return PTR_ERR(inode);
1220 		iput(inode);
1221 		gfs2_log_flush(sdp, NULL);
1222 		goto try_again;
1223 	}
1224 
1225 	al->al_file = file;
1226 	al->al_line = line;
1227 
1228 	return 0;
1229 }
1230 
1231 /**
1232  * gfs2_inplace_release - release an inplace reservation
1233  * @ip: the inode the reservation was taken out on
1234  *
1235  * Release a reservation made by gfs2_inplace_reserve().
1236  */
1237 
1238 void gfs2_inplace_release(struct gfs2_inode *ip)
1239 {
1240 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1241 	struct gfs2_alloc *al = ip->i_alloc;
1242 
1243 	if (gfs2_assert_warn(sdp, al->al_alloced <= al->al_requested) == -1)
1244 		fs_warn(sdp, "al_alloced = %u, al_requested = %u "
1245 			     "al_file = %s, al_line = %u\n",
1246 		             al->al_alloced, al->al_requested, al->al_file,
1247 			     al->al_line);
1248 
1249 	al->al_rgd = NULL;
1250 	if (al->al_rgd_gh.gh_gl)
1251 		gfs2_glock_dq_uninit(&al->al_rgd_gh);
1252 	if (ip != GFS2_I(sdp->sd_rindex))
1253 		gfs2_glock_dq_uninit(&al->al_ri_gh);
1254 }
1255 
1256 /**
1257  * gfs2_get_block_type - Check a block in a RG is of given type
1258  * @rgd: the resource group holding the block
1259  * @block: the block number
1260  *
1261  * Returns: The block type (GFS2_BLKST_*)
1262  */
1263 
1264 unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block)
1265 {
1266 	struct gfs2_bitmap *bi = NULL;
1267 	u32 length, rgrp_block, buf_block;
1268 	unsigned int buf;
1269 	unsigned char type;
1270 
1271 	length = rgd->rd_length;
1272 	rgrp_block = block - rgd->rd_data0;
1273 
1274 	for (buf = 0; buf < length; buf++) {
1275 		bi = rgd->rd_bits + buf;
1276 		if (rgrp_block < (bi->bi_start + bi->bi_len) * GFS2_NBBY)
1277 			break;
1278 	}
1279 
1280 	gfs2_assert(rgd->rd_sbd, buf < length);
1281 	buf_block = rgrp_block - bi->bi_start * GFS2_NBBY;
1282 
1283 	type = gfs2_testbit(rgd, bi->bi_bh->b_data + bi->bi_offset,
1284 			   bi->bi_len, buf_block);
1285 
1286 	return type;
1287 }
1288 
1289 /**
1290  * rgblk_search - find a block in @old_state, change allocation
1291  *           state to @new_state
1292  * @rgd: the resource group descriptor
1293  * @goal: the goal block within the RG (start here to search for avail block)
1294  * @old_state: GFS2_BLKST_XXX the before-allocation state to find
1295  * @new_state: GFS2_BLKST_XXX the after-allocation block state
1296  * @n: The extent length
1297  *
1298  * Walk rgrp's bitmap to find bits that represent a block in @old_state.
1299  * Add the found bitmap buffer to the transaction.
1300  * Set the found bits to @new_state to change block's allocation state.
1301  *
1302  * This function never fails, because we wouldn't call it unless we
1303  * know (from reservation results, etc.) that a block is available.
1304  *
1305  * Scope of @goal and returned block is just within rgrp, not the whole
1306  * filesystem.
1307  *
1308  * Returns:  the block number allocated
1309  */
1310 
1311 static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal,
1312 			unsigned char old_state, unsigned char new_state,
1313 			unsigned int *n)
1314 {
1315 	struct gfs2_bitmap *bi = NULL;
1316 	const u32 length = rgd->rd_length;
1317 	u32 blk = BFITNOENT;
1318 	unsigned int buf, x;
1319 	const unsigned int elen = *n;
1320 	const u8 *buffer = NULL;
1321 
1322 	*n = 0;
1323 	/* Find bitmap block that contains bits for goal block */
1324 	for (buf = 0; buf < length; buf++) {
1325 		bi = rgd->rd_bits + buf;
1326 		/* Convert scope of "goal" from rgrp-wide to within found bit block */
1327 		if (goal < (bi->bi_start + bi->bi_len) * GFS2_NBBY) {
1328 			goal -= bi->bi_start * GFS2_NBBY;
1329 			goto do_search;
1330 		}
1331 	}
1332 	buf = 0;
1333 	goal = 0;
1334 
1335 do_search:
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 		bi = rgd->rd_bits + buf;
1343 
1344 		if (test_bit(GBF_FULL, &bi->bi_flags) &&
1345 		    (old_state == GFS2_BLKST_FREE))
1346 			goto skip;
1347 
1348 		/* The GFS2_BLKST_UNLINKED state doesn't apply to the clone
1349 		   bitmaps, so we must search the originals for that. */
1350 		buffer = bi->bi_bh->b_data + bi->bi_offset;
1351 		if (old_state != GFS2_BLKST_UNLINKED && bi->bi_clone)
1352 			buffer = bi->bi_clone + bi->bi_offset;
1353 
1354 		blk = gfs2_bitfit(buffer, bi->bi_len, goal, old_state);
1355 		if (blk != BFITNOENT)
1356 			break;
1357 
1358 		if ((goal == 0) && (old_state == GFS2_BLKST_FREE))
1359 			set_bit(GBF_FULL, &bi->bi_flags);
1360 
1361 		/* Try next bitmap block (wrap back to rgrp header if at end) */
1362 skip:
1363 		buf++;
1364 		buf %= length;
1365 		goal = 0;
1366 	}
1367 
1368 	if (blk == BFITNOENT)
1369 		return blk;
1370 	*n = 1;
1371 	if (old_state == new_state)
1372 		goto out;
1373 
1374 	gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1);
1375 	gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, bi->bi_offset,
1376 		    bi->bi_len, blk, new_state);
1377 	goal = blk;
1378 	while (*n < elen) {
1379 		goal++;
1380 		if (goal >= (bi->bi_len * GFS2_NBBY))
1381 			break;
1382 		if (gfs2_testbit(rgd, buffer, bi->bi_len, goal) !=
1383 		    GFS2_BLKST_FREE)
1384 			break;
1385 		gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, bi->bi_offset,
1386 			    bi->bi_len, goal, new_state);
1387 		(*n)++;
1388 	}
1389 out:
1390 	return (bi->bi_start * GFS2_NBBY) + blk;
1391 }
1392 
1393 /**
1394  * rgblk_free - Change alloc state of given block(s)
1395  * @sdp: the filesystem
1396  * @bstart: the start of a run of blocks to free
1397  * @blen: the length of the block run (all must lie within ONE RG!)
1398  * @new_state: GFS2_BLKST_XXX the after-allocation block state
1399  *
1400  * Returns:  Resource group containing the block(s)
1401  */
1402 
1403 static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart,
1404 				     u32 blen, unsigned char new_state)
1405 {
1406 	struct gfs2_rgrpd *rgd;
1407 	struct gfs2_bitmap *bi = NULL;
1408 	u32 length, rgrp_blk, buf_blk;
1409 	unsigned int buf;
1410 
1411 	rgd = gfs2_blk2rgrpd(sdp, bstart);
1412 	if (!rgd) {
1413 		if (gfs2_consist(sdp))
1414 			fs_err(sdp, "block = %llu\n", (unsigned long long)bstart);
1415 		return NULL;
1416 	}
1417 
1418 	length = rgd->rd_length;
1419 
1420 	rgrp_blk = bstart - rgd->rd_data0;
1421 
1422 	while (blen--) {
1423 		for (buf = 0; buf < length; buf++) {
1424 			bi = rgd->rd_bits + buf;
1425 			if (rgrp_blk < (bi->bi_start + bi->bi_len) * GFS2_NBBY)
1426 				break;
1427 		}
1428 
1429 		gfs2_assert(rgd->rd_sbd, buf < length);
1430 
1431 		buf_blk = rgrp_blk - bi->bi_start * GFS2_NBBY;
1432 		rgrp_blk++;
1433 
1434 		if (!bi->bi_clone) {
1435 			bi->bi_clone = kmalloc(bi->bi_bh->b_size,
1436 					       GFP_NOFS | __GFP_NOFAIL);
1437 			memcpy(bi->bi_clone + bi->bi_offset,
1438 			       bi->bi_bh->b_data + bi->bi_offset,
1439 			       bi->bi_len);
1440 		}
1441 		gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1);
1442 		gfs2_setbit(rgd, bi->bi_bh->b_data, NULL, bi->bi_offset,
1443 			    bi->bi_len, buf_blk, new_state);
1444 	}
1445 
1446 	return rgd;
1447 }
1448 
1449 /**
1450  * gfs2_rgrp_dump - print out an rgrp
1451  * @seq: The iterator
1452  * @gl: The glock in question
1453  *
1454  */
1455 
1456 int gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl)
1457 {
1458 	const struct gfs2_rgrpd *rgd = gl->gl_object;
1459 	if (rgd == NULL)
1460 		return 0;
1461 	gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u\n",
1462 		       (unsigned long long)rgd->rd_addr, rgd->rd_flags,
1463 		       rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes);
1464 	return 0;
1465 }
1466 
1467 /**
1468  * gfs2_alloc_block - Allocate one or more blocks
1469  * @ip: the inode to allocate the block for
1470  * @bn: Used to return the starting block number
1471  * @n: requested number of blocks/extent length (value/result)
1472  *
1473  * Returns: 0 or error
1474  */
1475 
1476 int gfs2_alloc_block(struct gfs2_inode *ip, u64 *bn, unsigned int *n)
1477 {
1478 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1479 	struct buffer_head *dibh;
1480 	struct gfs2_alloc *al = ip->i_alloc;
1481 	struct gfs2_rgrpd *rgd = al->al_rgd;
1482 	u32 goal, blk;
1483 	u64 block;
1484 	int error;
1485 
1486 	if (rgrp_contains_block(rgd, ip->i_goal))
1487 		goal = ip->i_goal - rgd->rd_data0;
1488 	else
1489 		goal = rgd->rd_last_alloc;
1490 
1491 	blk = rgblk_search(rgd, goal, GFS2_BLKST_FREE, GFS2_BLKST_USED, n);
1492 
1493 	/* Since all blocks are reserved in advance, this shouldn't happen */
1494 	if (blk == BFITNOENT)
1495 		goto rgrp_error;
1496 
1497 	rgd->rd_last_alloc = blk;
1498 	block = rgd->rd_data0 + blk;
1499 	ip->i_goal = block;
1500 	error = gfs2_meta_inode_buffer(ip, &dibh);
1501 	if (error == 0) {
1502 		struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
1503 		gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1504 		di->di_goal_meta = di->di_goal_data = cpu_to_be64(ip->i_goal);
1505 		brelse(dibh);
1506 	}
1507 	if (rgd->rd_free < *n)
1508 		goto rgrp_error;
1509 
1510 	rgd->rd_free -= *n;
1511 
1512 	gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1513 	gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1514 
1515 	al->al_alloced += *n;
1516 
1517 	gfs2_statfs_change(sdp, 0, -(s64)*n, 0);
1518 	gfs2_quota_change(ip, *n, ip->i_inode.i_uid, ip->i_inode.i_gid);
1519 
1520 	spin_lock(&sdp->sd_rindex_spin);
1521 	rgd->rd_free_clone -= *n;
1522 	spin_unlock(&sdp->sd_rindex_spin);
1523 	trace_gfs2_block_alloc(ip, block, *n, GFS2_BLKST_USED);
1524 	*bn = block;
1525 	return 0;
1526 
1527 rgrp_error:
1528 	fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n",
1529 	        (unsigned long long)rgd->rd_addr);
1530 	fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n");
1531 	gfs2_rgrp_dump(NULL, rgd->rd_gl);
1532 	rgd->rd_flags |= GFS2_RDF_ERROR;
1533 	return -EIO;
1534 }
1535 
1536 /**
1537  * gfs2_alloc_di - Allocate a dinode
1538  * @dip: the directory that the inode is going in
1539  *
1540  * Returns: the block allocated
1541  */
1542 
1543 u64 gfs2_alloc_di(struct gfs2_inode *dip, u64 *generation)
1544 {
1545 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1546 	struct gfs2_alloc *al = dip->i_alloc;
1547 	struct gfs2_rgrpd *rgd = al->al_rgd;
1548 	u32 blk;
1549 	u64 block;
1550 	unsigned int n = 1;
1551 
1552 	blk = rgblk_search(rgd, rgd->rd_last_alloc,
1553 			   GFS2_BLKST_FREE, GFS2_BLKST_DINODE, &n);
1554 	BUG_ON(blk == BFITNOENT);
1555 
1556 	rgd->rd_last_alloc = blk;
1557 
1558 	block = rgd->rd_data0 + blk;
1559 
1560 	gfs2_assert_withdraw(sdp, rgd->rd_free);
1561 	rgd->rd_free--;
1562 	rgd->rd_dinodes++;
1563 	*generation = rgd->rd_igeneration++;
1564 	gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1565 	gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1566 
1567 	al->al_alloced++;
1568 
1569 	gfs2_statfs_change(sdp, 0, -1, +1);
1570 	gfs2_trans_add_unrevoke(sdp, block, 1);
1571 
1572 	spin_lock(&sdp->sd_rindex_spin);
1573 	rgd->rd_free_clone--;
1574 	spin_unlock(&sdp->sd_rindex_spin);
1575 	trace_gfs2_block_alloc(dip, block, 1, GFS2_BLKST_DINODE);
1576 	return block;
1577 }
1578 
1579 /**
1580  * gfs2_free_data - free a contiguous run of data block(s)
1581  * @ip: the inode these blocks are being freed from
1582  * @bstart: first block of a run of contiguous blocks
1583  * @blen: the length of the block run
1584  *
1585  */
1586 
1587 void gfs2_free_data(struct gfs2_inode *ip, u64 bstart, u32 blen)
1588 {
1589 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1590 	struct gfs2_rgrpd *rgd;
1591 
1592 	rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE);
1593 	if (!rgd)
1594 		return;
1595 	trace_gfs2_block_alloc(ip, bstart, blen, GFS2_BLKST_FREE);
1596 	rgd->rd_free += blen;
1597 
1598 	gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1599 	gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1600 
1601 	gfs2_trans_add_rg(rgd);
1602 
1603 	gfs2_statfs_change(sdp, 0, +blen, 0);
1604 	gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid);
1605 }
1606 
1607 /**
1608  * gfs2_free_meta - free a contiguous run of data block(s)
1609  * @ip: the inode these blocks are being freed from
1610  * @bstart: first block of a run of contiguous blocks
1611  * @blen: the length of the block run
1612  *
1613  */
1614 
1615 void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen)
1616 {
1617 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1618 	struct gfs2_rgrpd *rgd;
1619 
1620 	rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE);
1621 	if (!rgd)
1622 		return;
1623 	trace_gfs2_block_alloc(ip, bstart, blen, GFS2_BLKST_FREE);
1624 	rgd->rd_free += blen;
1625 
1626 	gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1627 	gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1628 
1629 	gfs2_trans_add_rg(rgd);
1630 
1631 	gfs2_statfs_change(sdp, 0, +blen, 0);
1632 	gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid);
1633 	gfs2_meta_wipe(ip, bstart, blen);
1634 }
1635 
1636 void gfs2_unlink_di(struct inode *inode)
1637 {
1638 	struct gfs2_inode *ip = GFS2_I(inode);
1639 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1640 	struct gfs2_rgrpd *rgd;
1641 	u64 blkno = ip->i_no_addr;
1642 
1643 	rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED);
1644 	if (!rgd)
1645 		return;
1646 	trace_gfs2_block_alloc(ip, blkno, 1, GFS2_BLKST_UNLINKED);
1647 	gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1648 	gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1649 	gfs2_trans_add_rg(rgd);
1650 }
1651 
1652 static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno)
1653 {
1654 	struct gfs2_sbd *sdp = rgd->rd_sbd;
1655 	struct gfs2_rgrpd *tmp_rgd;
1656 
1657 	tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE);
1658 	if (!tmp_rgd)
1659 		return;
1660 	gfs2_assert_withdraw(sdp, rgd == tmp_rgd);
1661 
1662 	if (!rgd->rd_dinodes)
1663 		gfs2_consist_rgrpd(rgd);
1664 	rgd->rd_dinodes--;
1665 	rgd->rd_free++;
1666 
1667 	gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1668 	gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1669 
1670 	gfs2_statfs_change(sdp, 0, +1, -1);
1671 	gfs2_trans_add_rg(rgd);
1672 }
1673 
1674 
1675 void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip)
1676 {
1677 	gfs2_free_uninit_di(rgd, ip->i_no_addr);
1678 	trace_gfs2_block_alloc(ip, ip->i_no_addr, 1, GFS2_BLKST_FREE);
1679 	gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid);
1680 	gfs2_meta_wipe(ip, ip->i_no_addr, 1);
1681 }
1682 
1683 /**
1684  * gfs2_rlist_add - add a RG to a list of RGs
1685  * @sdp: the filesystem
1686  * @rlist: the list of resource groups
1687  * @block: the block
1688  *
1689  * Figure out what RG a block belongs to and add that RG to the list
1690  *
1691  * FIXME: Don't use NOFAIL
1692  *
1693  */
1694 
1695 void gfs2_rlist_add(struct gfs2_sbd *sdp, struct gfs2_rgrp_list *rlist,
1696 		    u64 block)
1697 {
1698 	struct gfs2_rgrpd *rgd;
1699 	struct gfs2_rgrpd **tmp;
1700 	unsigned int new_space;
1701 	unsigned int x;
1702 
1703 	if (gfs2_assert_warn(sdp, !rlist->rl_ghs))
1704 		return;
1705 
1706 	rgd = gfs2_blk2rgrpd(sdp, block);
1707 	if (!rgd) {
1708 		if (gfs2_consist(sdp))
1709 			fs_err(sdp, "block = %llu\n", (unsigned long long)block);
1710 		return;
1711 	}
1712 
1713 	for (x = 0; x < rlist->rl_rgrps; x++)
1714 		if (rlist->rl_rgd[x] == rgd)
1715 			return;
1716 
1717 	if (rlist->rl_rgrps == rlist->rl_space) {
1718 		new_space = rlist->rl_space + 10;
1719 
1720 		tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *),
1721 			      GFP_NOFS | __GFP_NOFAIL);
1722 
1723 		if (rlist->rl_rgd) {
1724 			memcpy(tmp, rlist->rl_rgd,
1725 			       rlist->rl_space * sizeof(struct gfs2_rgrpd *));
1726 			kfree(rlist->rl_rgd);
1727 		}
1728 
1729 		rlist->rl_space = new_space;
1730 		rlist->rl_rgd = tmp;
1731 	}
1732 
1733 	rlist->rl_rgd[rlist->rl_rgrps++] = rgd;
1734 }
1735 
1736 /**
1737  * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate
1738  *      and initialize an array of glock holders for them
1739  * @rlist: the list of resource groups
1740  * @state: the lock state to acquire the RG lock in
1741  * @flags: the modifier flags for the holder structures
1742  *
1743  * FIXME: Don't use NOFAIL
1744  *
1745  */
1746 
1747 void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state)
1748 {
1749 	unsigned int x;
1750 
1751 	rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder),
1752 				GFP_NOFS | __GFP_NOFAIL);
1753 	for (x = 0; x < rlist->rl_rgrps; x++)
1754 		gfs2_holder_init(rlist->rl_rgd[x]->rd_gl,
1755 				state, 0,
1756 				&rlist->rl_ghs[x]);
1757 }
1758 
1759 /**
1760  * gfs2_rlist_free - free a resource group list
1761  * @list: the list of resource groups
1762  *
1763  */
1764 
1765 void gfs2_rlist_free(struct gfs2_rgrp_list *rlist)
1766 {
1767 	unsigned int x;
1768 
1769 	kfree(rlist->rl_rgd);
1770 
1771 	if (rlist->rl_ghs) {
1772 		for (x = 0; x < rlist->rl_rgrps; x++)
1773 			gfs2_holder_uninit(&rlist->rl_ghs[x]);
1774 		kfree(rlist->rl_ghs);
1775 	}
1776 }
1777 
1778