xref: /openbmc/linux/fs/gfs2/rgrp.c (revision 9b9c2cd4)
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 
12 #include <linux/slab.h>
13 #include <linux/spinlock.h>
14 #include <linux/completion.h>
15 #include <linux/buffer_head.h>
16 #include <linux/fs.h>
17 #include <linux/gfs2_ondisk.h>
18 #include <linux/prefetch.h>
19 #include <linux/blkdev.h>
20 #include <linux/rbtree.h>
21 #include <linux/random.h>
22 
23 #include "gfs2.h"
24 #include "incore.h"
25 #include "glock.h"
26 #include "glops.h"
27 #include "lops.h"
28 #include "meta_io.h"
29 #include "quota.h"
30 #include "rgrp.h"
31 #include "super.h"
32 #include "trans.h"
33 #include "util.h"
34 #include "log.h"
35 #include "inode.h"
36 #include "trace_gfs2.h"
37 
38 #define BFITNOENT ((u32)~0)
39 #define NO_BLOCK ((u64)~0)
40 
41 #if BITS_PER_LONG == 32
42 #define LBITMASK   (0x55555555UL)
43 #define LBITSKIP55 (0x55555555UL)
44 #define LBITSKIP00 (0x00000000UL)
45 #else
46 #define LBITMASK   (0x5555555555555555UL)
47 #define LBITSKIP55 (0x5555555555555555UL)
48 #define LBITSKIP00 (0x0000000000000000UL)
49 #endif
50 
51 /*
52  * These routines are used by the resource group routines (rgrp.c)
53  * to keep track of block allocation.  Each block is represented by two
54  * bits.  So, each byte represents GFS2_NBBY (i.e. 4) blocks.
55  *
56  * 0 = Free
57  * 1 = Used (not metadata)
58  * 2 = Unlinked (still in use) inode
59  * 3 = Used (metadata)
60  */
61 
62 struct gfs2_extent {
63 	struct gfs2_rbm rbm;
64 	u32 len;
65 };
66 
67 static const char valid_change[16] = {
68 	        /* current */
69 	/* n */ 0, 1, 1, 1,
70 	/* e */ 1, 0, 0, 0,
71 	/* w */ 0, 0, 0, 1,
72 	        1, 0, 0, 0
73 };
74 
75 static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 *minext,
76 			 const struct gfs2_inode *ip, bool nowrap,
77 			 const struct gfs2_alloc_parms *ap);
78 
79 
80 /**
81  * gfs2_setbit - Set a bit in the bitmaps
82  * @rbm: The position of the bit to set
83  * @do_clone: Also set the clone bitmap, if it exists
84  * @new_state: the new state of the block
85  *
86  */
87 
88 static inline void gfs2_setbit(const struct gfs2_rbm *rbm, bool do_clone,
89 			       unsigned char new_state)
90 {
91 	unsigned char *byte1, *byte2, *end, cur_state;
92 	struct gfs2_bitmap *bi = rbm_bi(rbm);
93 	unsigned int buflen = bi->bi_len;
94 	const unsigned int bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE;
95 
96 	byte1 = bi->bi_bh->b_data + bi->bi_offset + (rbm->offset / GFS2_NBBY);
97 	end = bi->bi_bh->b_data + bi->bi_offset + buflen;
98 
99 	BUG_ON(byte1 >= end);
100 
101 	cur_state = (*byte1 >> bit) & GFS2_BIT_MASK;
102 
103 	if (unlikely(!valid_change[new_state * 4 + cur_state])) {
104 		pr_warn("buf_blk = 0x%x old_state=%d, new_state=%d\n",
105 			rbm->offset, cur_state, new_state);
106 		pr_warn("rgrp=0x%llx bi_start=0x%x\n",
107 			(unsigned long long)rbm->rgd->rd_addr, bi->bi_start);
108 		pr_warn("bi_offset=0x%x bi_len=0x%x\n",
109 			bi->bi_offset, bi->bi_len);
110 		dump_stack();
111 		gfs2_consist_rgrpd(rbm->rgd);
112 		return;
113 	}
114 	*byte1 ^= (cur_state ^ new_state) << bit;
115 
116 	if (do_clone && bi->bi_clone) {
117 		byte2 = bi->bi_clone + bi->bi_offset + (rbm->offset / GFS2_NBBY);
118 		cur_state = (*byte2 >> bit) & GFS2_BIT_MASK;
119 		*byte2 ^= (cur_state ^ new_state) << bit;
120 	}
121 }
122 
123 /**
124  * gfs2_testbit - test a bit in the bitmaps
125  * @rbm: The bit to test
126  *
127  * Returns: The two bit block state of the requested bit
128  */
129 
130 static inline u8 gfs2_testbit(const struct gfs2_rbm *rbm)
131 {
132 	struct gfs2_bitmap *bi = rbm_bi(rbm);
133 	const u8 *buffer = bi->bi_bh->b_data + bi->bi_offset;
134 	const u8 *byte;
135 	unsigned int bit;
136 
137 	byte = buffer + (rbm->offset / GFS2_NBBY);
138 	bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE;
139 
140 	return (*byte >> bit) & GFS2_BIT_MASK;
141 }
142 
143 /**
144  * gfs2_bit_search
145  * @ptr: Pointer to bitmap data
146  * @mask: Mask to use (normally 0x55555.... but adjusted for search start)
147  * @state: The state we are searching for
148  *
149  * We xor the bitmap data with a patter which is the bitwise opposite
150  * of what we are looking for, this gives rise to a pattern of ones
151  * wherever there is a match. Since we have two bits per entry, we
152  * take this pattern, shift it down by one place and then and it with
153  * the original. All the even bit positions (0,2,4, etc) then represent
154  * successful matches, so we mask with 0x55555..... to remove the unwanted
155  * odd bit positions.
156  *
157  * This allows searching of a whole u64 at once (32 blocks) with a
158  * single test (on 64 bit arches).
159  */
160 
161 static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state)
162 {
163 	u64 tmp;
164 	static const u64 search[] = {
165 		[0] = 0xffffffffffffffffULL,
166 		[1] = 0xaaaaaaaaaaaaaaaaULL,
167 		[2] = 0x5555555555555555ULL,
168 		[3] = 0x0000000000000000ULL,
169 	};
170 	tmp = le64_to_cpu(*ptr) ^ search[state];
171 	tmp &= (tmp >> 1);
172 	tmp &= mask;
173 	return tmp;
174 }
175 
176 /**
177  * rs_cmp - multi-block reservation range compare
178  * @blk: absolute file system block number of the new reservation
179  * @len: number of blocks in the new reservation
180  * @rs: existing reservation to compare against
181  *
182  * returns: 1 if the block range is beyond the reach of the reservation
183  *         -1 if the block range is before the start of the reservation
184  *          0 if the block range overlaps with the reservation
185  */
186 static inline int rs_cmp(u64 blk, u32 len, struct gfs2_blkreserv *rs)
187 {
188 	u64 startblk = gfs2_rbm_to_block(&rs->rs_rbm);
189 
190 	if (blk >= startblk + rs->rs_free)
191 		return 1;
192 	if (blk + len - 1 < startblk)
193 		return -1;
194 	return 0;
195 }
196 
197 /**
198  * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing
199  *       a block in a given allocation state.
200  * @buf: the buffer that holds the bitmaps
201  * @len: the length (in bytes) of the buffer
202  * @goal: start search at this block's bit-pair (within @buffer)
203  * @state: GFS2_BLKST_XXX the state of the block we're looking for.
204  *
205  * Scope of @goal and returned block number is only within this bitmap buffer,
206  * not entire rgrp or filesystem.  @buffer will be offset from the actual
207  * beginning of a bitmap block buffer, skipping any header structures, but
208  * headers are always a multiple of 64 bits long so that the buffer is
209  * always aligned to a 64 bit boundary.
210  *
211  * The size of the buffer is in bytes, but is it assumed that it is
212  * always ok to read a complete multiple of 64 bits at the end
213  * of the block in case the end is no aligned to a natural boundary.
214  *
215  * Return: the block number (bitmap buffer scope) that was found
216  */
217 
218 static u32 gfs2_bitfit(const u8 *buf, const unsigned int len,
219 		       u32 goal, u8 state)
220 {
221 	u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1);
222 	const __le64 *ptr = ((__le64 *)buf) + (goal >> 5);
223 	const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64)));
224 	u64 tmp;
225 	u64 mask = 0x5555555555555555ULL;
226 	u32 bit;
227 
228 	/* Mask off bits we don't care about at the start of the search */
229 	mask <<= spoint;
230 	tmp = gfs2_bit_search(ptr, mask, state);
231 	ptr++;
232 	while(tmp == 0 && ptr < end) {
233 		tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state);
234 		ptr++;
235 	}
236 	/* Mask off any bits which are more than len bytes from the start */
237 	if (ptr == end && (len & (sizeof(u64) - 1)))
238 		tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1))));
239 	/* Didn't find anything, so return */
240 	if (tmp == 0)
241 		return BFITNOENT;
242 	ptr--;
243 	bit = __ffs64(tmp);
244 	bit /= 2;	/* two bits per entry in the bitmap */
245 	return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit;
246 }
247 
248 /**
249  * gfs2_rbm_from_block - Set the rbm based upon rgd and block number
250  * @rbm: The rbm with rgd already set correctly
251  * @block: The block number (filesystem relative)
252  *
253  * This sets the bi and offset members of an rbm based on a
254  * resource group and a filesystem relative block number. The
255  * resource group must be set in the rbm on entry, the bi and
256  * offset members will be set by this function.
257  *
258  * Returns: 0 on success, or an error code
259  */
260 
261 static int gfs2_rbm_from_block(struct gfs2_rbm *rbm, u64 block)
262 {
263 	u64 rblock = block - rbm->rgd->rd_data0;
264 
265 	if (WARN_ON_ONCE(rblock > UINT_MAX))
266 		return -EINVAL;
267 	if (block >= rbm->rgd->rd_data0 + rbm->rgd->rd_data)
268 		return -E2BIG;
269 
270 	rbm->bii = 0;
271 	rbm->offset = (u32)(rblock);
272 	/* Check if the block is within the first block */
273 	if (rbm->offset < rbm_bi(rbm)->bi_blocks)
274 		return 0;
275 
276 	/* Adjust for the size diff between gfs2_meta_header and gfs2_rgrp */
277 	rbm->offset += (sizeof(struct gfs2_rgrp) -
278 			sizeof(struct gfs2_meta_header)) * GFS2_NBBY;
279 	rbm->bii = rbm->offset / rbm->rgd->rd_sbd->sd_blocks_per_bitmap;
280 	rbm->offset -= rbm->bii * rbm->rgd->rd_sbd->sd_blocks_per_bitmap;
281 	return 0;
282 }
283 
284 /**
285  * gfs2_rbm_incr - increment an rbm structure
286  * @rbm: The rbm with rgd already set correctly
287  *
288  * This function takes an existing rbm structure and increments it to the next
289  * viable block offset.
290  *
291  * Returns: If incrementing the offset would cause the rbm to go past the
292  *          end of the rgrp, true is returned, otherwise false.
293  *
294  */
295 
296 static bool gfs2_rbm_incr(struct gfs2_rbm *rbm)
297 {
298 	if (rbm->offset + 1 < rbm_bi(rbm)->bi_blocks) { /* in the same bitmap */
299 		rbm->offset++;
300 		return false;
301 	}
302 	if (rbm->bii == rbm->rgd->rd_length - 1) /* at the last bitmap */
303 		return true;
304 
305 	rbm->offset = 0;
306 	rbm->bii++;
307 	return false;
308 }
309 
310 /**
311  * gfs2_unaligned_extlen - Look for free blocks which are not byte aligned
312  * @rbm: Position to search (value/result)
313  * @n_unaligned: Number of unaligned blocks to check
314  * @len: Decremented for each block found (terminate on zero)
315  *
316  * Returns: true if a non-free block is encountered
317  */
318 
319 static bool gfs2_unaligned_extlen(struct gfs2_rbm *rbm, u32 n_unaligned, u32 *len)
320 {
321 	u32 n;
322 	u8 res;
323 
324 	for (n = 0; n < n_unaligned; n++) {
325 		res = gfs2_testbit(rbm);
326 		if (res != GFS2_BLKST_FREE)
327 			return true;
328 		(*len)--;
329 		if (*len == 0)
330 			return true;
331 		if (gfs2_rbm_incr(rbm))
332 			return true;
333 	}
334 
335 	return false;
336 }
337 
338 /**
339  * gfs2_free_extlen - Return extent length of free blocks
340  * @rrbm: Starting position
341  * @len: Max length to check
342  *
343  * Starting at the block specified by the rbm, see how many free blocks
344  * there are, not reading more than len blocks ahead. This can be done
345  * using memchr_inv when the blocks are byte aligned, but has to be done
346  * on a block by block basis in case of unaligned blocks. Also this
347  * function can cope with bitmap boundaries (although it must stop on
348  * a resource group boundary)
349  *
350  * Returns: Number of free blocks in the extent
351  */
352 
353 static u32 gfs2_free_extlen(const struct gfs2_rbm *rrbm, u32 len)
354 {
355 	struct gfs2_rbm rbm = *rrbm;
356 	u32 n_unaligned = rbm.offset & 3;
357 	u32 size = len;
358 	u32 bytes;
359 	u32 chunk_size;
360 	u8 *ptr, *start, *end;
361 	u64 block;
362 	struct gfs2_bitmap *bi;
363 
364 	if (n_unaligned &&
365 	    gfs2_unaligned_extlen(&rbm, 4 - n_unaligned, &len))
366 		goto out;
367 
368 	n_unaligned = len & 3;
369 	/* Start is now byte aligned */
370 	while (len > 3) {
371 		bi = rbm_bi(&rbm);
372 		start = bi->bi_bh->b_data;
373 		if (bi->bi_clone)
374 			start = bi->bi_clone;
375 		end = start + bi->bi_bh->b_size;
376 		start += bi->bi_offset;
377 		BUG_ON(rbm.offset & 3);
378 		start += (rbm.offset / GFS2_NBBY);
379 		bytes = min_t(u32, len / GFS2_NBBY, (end - start));
380 		ptr = memchr_inv(start, 0, bytes);
381 		chunk_size = ((ptr == NULL) ? bytes : (ptr - start));
382 		chunk_size *= GFS2_NBBY;
383 		BUG_ON(len < chunk_size);
384 		len -= chunk_size;
385 		block = gfs2_rbm_to_block(&rbm);
386 		if (gfs2_rbm_from_block(&rbm, block + chunk_size)) {
387 			n_unaligned = 0;
388 			break;
389 		}
390 		if (ptr) {
391 			n_unaligned = 3;
392 			break;
393 		}
394 		n_unaligned = len & 3;
395 	}
396 
397 	/* Deal with any bits left over at the end */
398 	if (n_unaligned)
399 		gfs2_unaligned_extlen(&rbm, n_unaligned, &len);
400 out:
401 	return size - len;
402 }
403 
404 /**
405  * gfs2_bitcount - count the number of bits in a certain state
406  * @rgd: the resource group descriptor
407  * @buffer: the buffer that holds the bitmaps
408  * @buflen: the length (in bytes) of the buffer
409  * @state: the state of the block we're looking for
410  *
411  * Returns: The number of bits
412  */
413 
414 static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer,
415 			 unsigned int buflen, u8 state)
416 {
417 	const u8 *byte = buffer;
418 	const u8 *end = buffer + buflen;
419 	const u8 state1 = state << 2;
420 	const u8 state2 = state << 4;
421 	const u8 state3 = state << 6;
422 	u32 count = 0;
423 
424 	for (; byte < end; byte++) {
425 		if (((*byte) & 0x03) == state)
426 			count++;
427 		if (((*byte) & 0x0C) == state1)
428 			count++;
429 		if (((*byte) & 0x30) == state2)
430 			count++;
431 		if (((*byte) & 0xC0) == state3)
432 			count++;
433 	}
434 
435 	return count;
436 }
437 
438 /**
439  * gfs2_rgrp_verify - Verify that a resource group is consistent
440  * @rgd: the rgrp
441  *
442  */
443 
444 void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd)
445 {
446 	struct gfs2_sbd *sdp = rgd->rd_sbd;
447 	struct gfs2_bitmap *bi = NULL;
448 	u32 length = rgd->rd_length;
449 	u32 count[4], tmp;
450 	int buf, x;
451 
452 	memset(count, 0, 4 * sizeof(u32));
453 
454 	/* Count # blocks in each of 4 possible allocation states */
455 	for (buf = 0; buf < length; buf++) {
456 		bi = rgd->rd_bits + buf;
457 		for (x = 0; x < 4; x++)
458 			count[x] += gfs2_bitcount(rgd,
459 						  bi->bi_bh->b_data +
460 						  bi->bi_offset,
461 						  bi->bi_len, x);
462 	}
463 
464 	if (count[0] != rgd->rd_free) {
465 		if (gfs2_consist_rgrpd(rgd))
466 			fs_err(sdp, "free data mismatch:  %u != %u\n",
467 			       count[0], rgd->rd_free);
468 		return;
469 	}
470 
471 	tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes;
472 	if (count[1] != tmp) {
473 		if (gfs2_consist_rgrpd(rgd))
474 			fs_err(sdp, "used data mismatch:  %u != %u\n",
475 			       count[1], tmp);
476 		return;
477 	}
478 
479 	if (count[2] + count[3] != rgd->rd_dinodes) {
480 		if (gfs2_consist_rgrpd(rgd))
481 			fs_err(sdp, "used metadata mismatch:  %u != %u\n",
482 			       count[2] + count[3], rgd->rd_dinodes);
483 		return;
484 	}
485 }
486 
487 static inline int rgrp_contains_block(struct gfs2_rgrpd *rgd, u64 block)
488 {
489 	u64 first = rgd->rd_data0;
490 	u64 last = first + rgd->rd_data;
491 	return first <= block && block < last;
492 }
493 
494 /**
495  * gfs2_blk2rgrpd - Find resource group for a given data/meta block number
496  * @sdp: The GFS2 superblock
497  * @blk: The data block number
498  * @exact: True if this needs to be an exact match
499  *
500  * Returns: The resource group, or NULL if not found
501  */
502 
503 struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk, bool exact)
504 {
505 	struct rb_node *n, *next;
506 	struct gfs2_rgrpd *cur;
507 
508 	spin_lock(&sdp->sd_rindex_spin);
509 	n = sdp->sd_rindex_tree.rb_node;
510 	while (n) {
511 		cur = rb_entry(n, struct gfs2_rgrpd, rd_node);
512 		next = NULL;
513 		if (blk < cur->rd_addr)
514 			next = n->rb_left;
515 		else if (blk >= cur->rd_data0 + cur->rd_data)
516 			next = n->rb_right;
517 		if (next == NULL) {
518 			spin_unlock(&sdp->sd_rindex_spin);
519 			if (exact) {
520 				if (blk < cur->rd_addr)
521 					return NULL;
522 				if (blk >= cur->rd_data0 + cur->rd_data)
523 					return NULL;
524 			}
525 			return cur;
526 		}
527 		n = next;
528 	}
529 	spin_unlock(&sdp->sd_rindex_spin);
530 
531 	return NULL;
532 }
533 
534 /**
535  * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem
536  * @sdp: The GFS2 superblock
537  *
538  * Returns: The first rgrp in the filesystem
539  */
540 
541 struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp)
542 {
543 	const struct rb_node *n;
544 	struct gfs2_rgrpd *rgd;
545 
546 	spin_lock(&sdp->sd_rindex_spin);
547 	n = rb_first(&sdp->sd_rindex_tree);
548 	rgd = rb_entry(n, struct gfs2_rgrpd, rd_node);
549 	spin_unlock(&sdp->sd_rindex_spin);
550 
551 	return rgd;
552 }
553 
554 /**
555  * gfs2_rgrpd_get_next - get the next RG
556  * @rgd: the resource group descriptor
557  *
558  * Returns: The next rgrp
559  */
560 
561 struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd)
562 {
563 	struct gfs2_sbd *sdp = rgd->rd_sbd;
564 	const struct rb_node *n;
565 
566 	spin_lock(&sdp->sd_rindex_spin);
567 	n = rb_next(&rgd->rd_node);
568 	if (n == NULL)
569 		n = rb_first(&sdp->sd_rindex_tree);
570 
571 	if (unlikely(&rgd->rd_node == n)) {
572 		spin_unlock(&sdp->sd_rindex_spin);
573 		return NULL;
574 	}
575 	rgd = rb_entry(n, struct gfs2_rgrpd, rd_node);
576 	spin_unlock(&sdp->sd_rindex_spin);
577 	return rgd;
578 }
579 
580 void check_and_update_goal(struct gfs2_inode *ip)
581 {
582 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
583 	if (!ip->i_goal || gfs2_blk2rgrpd(sdp, ip->i_goal, 1) == NULL)
584 		ip->i_goal = ip->i_no_addr;
585 }
586 
587 void gfs2_free_clones(struct gfs2_rgrpd *rgd)
588 {
589 	int x;
590 
591 	for (x = 0; x < rgd->rd_length; x++) {
592 		struct gfs2_bitmap *bi = rgd->rd_bits + x;
593 		kfree(bi->bi_clone);
594 		bi->bi_clone = NULL;
595 	}
596 }
597 
598 /**
599  * gfs2_rs_alloc - make sure we have a reservation assigned to the inode
600  * @ip: the inode for this reservation
601  */
602 int gfs2_rs_alloc(struct gfs2_inode *ip)
603 {
604 	int error = 0;
605 
606 	down_write(&ip->i_rw_mutex);
607 	if (ip->i_res)
608 		goto out;
609 
610 	ip->i_res = kmem_cache_zalloc(gfs2_rsrv_cachep, GFP_NOFS);
611 	if (!ip->i_res) {
612 		error = -ENOMEM;
613 		goto out;
614 	}
615 
616 	RB_CLEAR_NODE(&ip->i_res->rs_node);
617 out:
618 	up_write(&ip->i_rw_mutex);
619 	return error;
620 }
621 
622 static void dump_rs(struct seq_file *seq, const struct gfs2_blkreserv *rs)
623 {
624 	gfs2_print_dbg(seq, "  B: n:%llu s:%llu b:%u f:%u\n",
625 		       (unsigned long long)rs->rs_inum,
626 		       (unsigned long long)gfs2_rbm_to_block(&rs->rs_rbm),
627 		       rs->rs_rbm.offset, rs->rs_free);
628 }
629 
630 /**
631  * __rs_deltree - remove a multi-block reservation from the rgd tree
632  * @rs: The reservation to remove
633  *
634  */
635 static void __rs_deltree(struct gfs2_blkreserv *rs)
636 {
637 	struct gfs2_rgrpd *rgd;
638 
639 	if (!gfs2_rs_active(rs))
640 		return;
641 
642 	rgd = rs->rs_rbm.rgd;
643 	trace_gfs2_rs(rs, TRACE_RS_TREEDEL);
644 	rb_erase(&rs->rs_node, &rgd->rd_rstree);
645 	RB_CLEAR_NODE(&rs->rs_node);
646 
647 	if (rs->rs_free) {
648 		struct gfs2_bitmap *bi = rbm_bi(&rs->rs_rbm);
649 
650 		/* return reserved blocks to the rgrp */
651 		BUG_ON(rs->rs_rbm.rgd->rd_reserved < rs->rs_free);
652 		rs->rs_rbm.rgd->rd_reserved -= rs->rs_free;
653 		/* The rgrp extent failure point is likely not to increase;
654 		   it will only do so if the freed blocks are somehow
655 		   contiguous with a span of free blocks that follows. Still,
656 		   it will force the number to be recalculated later. */
657 		rgd->rd_extfail_pt += rs->rs_free;
658 		rs->rs_free = 0;
659 		clear_bit(GBF_FULL, &bi->bi_flags);
660 	}
661 }
662 
663 /**
664  * gfs2_rs_deltree - remove a multi-block reservation from the rgd tree
665  * @rs: The reservation to remove
666  *
667  */
668 void gfs2_rs_deltree(struct gfs2_blkreserv *rs)
669 {
670 	struct gfs2_rgrpd *rgd;
671 
672 	rgd = rs->rs_rbm.rgd;
673 	if (rgd) {
674 		spin_lock(&rgd->rd_rsspin);
675 		__rs_deltree(rs);
676 		spin_unlock(&rgd->rd_rsspin);
677 	}
678 }
679 
680 /**
681  * gfs2_rs_delete - delete a multi-block reservation
682  * @ip: The inode for this reservation
683  * @wcount: The inode's write count, or NULL
684  *
685  */
686 void gfs2_rs_delete(struct gfs2_inode *ip, atomic_t *wcount)
687 {
688 	down_write(&ip->i_rw_mutex);
689 	if (ip->i_res && ((wcount == NULL) || (atomic_read(wcount) <= 1))) {
690 		gfs2_rs_deltree(ip->i_res);
691 		BUG_ON(ip->i_res->rs_free);
692 		kmem_cache_free(gfs2_rsrv_cachep, ip->i_res);
693 		ip->i_res = NULL;
694 	}
695 	up_write(&ip->i_rw_mutex);
696 }
697 
698 /**
699  * return_all_reservations - return all reserved blocks back to the rgrp.
700  * @rgd: the rgrp that needs its space back
701  *
702  * We previously reserved a bunch of blocks for allocation. Now we need to
703  * give them back. This leave the reservation structures in tact, but removes
704  * all of their corresponding "no-fly zones".
705  */
706 static void return_all_reservations(struct gfs2_rgrpd *rgd)
707 {
708 	struct rb_node *n;
709 	struct gfs2_blkreserv *rs;
710 
711 	spin_lock(&rgd->rd_rsspin);
712 	while ((n = rb_first(&rgd->rd_rstree))) {
713 		rs = rb_entry(n, struct gfs2_blkreserv, rs_node);
714 		__rs_deltree(rs);
715 	}
716 	spin_unlock(&rgd->rd_rsspin);
717 }
718 
719 void gfs2_clear_rgrpd(struct gfs2_sbd *sdp)
720 {
721 	struct rb_node *n;
722 	struct gfs2_rgrpd *rgd;
723 	struct gfs2_glock *gl;
724 
725 	while ((n = rb_first(&sdp->sd_rindex_tree))) {
726 		rgd = rb_entry(n, struct gfs2_rgrpd, rd_node);
727 		gl = rgd->rd_gl;
728 
729 		rb_erase(n, &sdp->sd_rindex_tree);
730 
731 		if (gl) {
732 			spin_lock(&gl->gl_lockref.lock);
733 			gl->gl_object = NULL;
734 			spin_unlock(&gl->gl_lockref.lock);
735 			gfs2_glock_add_to_lru(gl);
736 			gfs2_glock_put(gl);
737 		}
738 
739 		gfs2_free_clones(rgd);
740 		kfree(rgd->rd_bits);
741 		return_all_reservations(rgd);
742 		kmem_cache_free(gfs2_rgrpd_cachep, rgd);
743 	}
744 }
745 
746 static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd)
747 {
748 	pr_info("ri_addr = %llu\n", (unsigned long long)rgd->rd_addr);
749 	pr_info("ri_length = %u\n", rgd->rd_length);
750 	pr_info("ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0);
751 	pr_info("ri_data = %u\n", rgd->rd_data);
752 	pr_info("ri_bitbytes = %u\n", rgd->rd_bitbytes);
753 }
754 
755 /**
756  * gfs2_compute_bitstructs - Compute the bitmap sizes
757  * @rgd: The resource group descriptor
758  *
759  * Calculates bitmap descriptors, one for each block that contains bitmap data
760  *
761  * Returns: errno
762  */
763 
764 static int compute_bitstructs(struct gfs2_rgrpd *rgd)
765 {
766 	struct gfs2_sbd *sdp = rgd->rd_sbd;
767 	struct gfs2_bitmap *bi;
768 	u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */
769 	u32 bytes_left, bytes;
770 	int x;
771 
772 	if (!length)
773 		return -EINVAL;
774 
775 	rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS);
776 	if (!rgd->rd_bits)
777 		return -ENOMEM;
778 
779 	bytes_left = rgd->rd_bitbytes;
780 
781 	for (x = 0; x < length; x++) {
782 		bi = rgd->rd_bits + x;
783 
784 		bi->bi_flags = 0;
785 		/* small rgrp; bitmap stored completely in header block */
786 		if (length == 1) {
787 			bytes = bytes_left;
788 			bi->bi_offset = sizeof(struct gfs2_rgrp);
789 			bi->bi_start = 0;
790 			bi->bi_len = bytes;
791 			bi->bi_blocks = bytes * GFS2_NBBY;
792 		/* header block */
793 		} else if (x == 0) {
794 			bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp);
795 			bi->bi_offset = sizeof(struct gfs2_rgrp);
796 			bi->bi_start = 0;
797 			bi->bi_len = bytes;
798 			bi->bi_blocks = bytes * GFS2_NBBY;
799 		/* last block */
800 		} else if (x + 1 == length) {
801 			bytes = bytes_left;
802 			bi->bi_offset = sizeof(struct gfs2_meta_header);
803 			bi->bi_start = rgd->rd_bitbytes - bytes_left;
804 			bi->bi_len = bytes;
805 			bi->bi_blocks = bytes * GFS2_NBBY;
806 		/* other blocks */
807 		} else {
808 			bytes = sdp->sd_sb.sb_bsize -
809 				sizeof(struct gfs2_meta_header);
810 			bi->bi_offset = sizeof(struct gfs2_meta_header);
811 			bi->bi_start = rgd->rd_bitbytes - bytes_left;
812 			bi->bi_len = bytes;
813 			bi->bi_blocks = bytes * GFS2_NBBY;
814 		}
815 
816 		bytes_left -= bytes;
817 	}
818 
819 	if (bytes_left) {
820 		gfs2_consist_rgrpd(rgd);
821 		return -EIO;
822 	}
823 	bi = rgd->rd_bits + (length - 1);
824 	if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) {
825 		if (gfs2_consist_rgrpd(rgd)) {
826 			gfs2_rindex_print(rgd);
827 			fs_err(sdp, "start=%u len=%u offset=%u\n",
828 			       bi->bi_start, bi->bi_len, bi->bi_offset);
829 		}
830 		return -EIO;
831 	}
832 
833 	return 0;
834 }
835 
836 /**
837  * gfs2_ri_total - Total up the file system space, according to the rindex.
838  * @sdp: the filesystem
839  *
840  */
841 u64 gfs2_ri_total(struct gfs2_sbd *sdp)
842 {
843 	u64 total_data = 0;
844 	struct inode *inode = sdp->sd_rindex;
845 	struct gfs2_inode *ip = GFS2_I(inode);
846 	char buf[sizeof(struct gfs2_rindex)];
847 	int error, rgrps;
848 
849 	for (rgrps = 0;; rgrps++) {
850 		loff_t pos = rgrps * sizeof(struct gfs2_rindex);
851 
852 		if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode))
853 			break;
854 		error = gfs2_internal_read(ip, buf, &pos,
855 					   sizeof(struct gfs2_rindex));
856 		if (error != sizeof(struct gfs2_rindex))
857 			break;
858 		total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data);
859 	}
860 	return total_data;
861 }
862 
863 static int rgd_insert(struct gfs2_rgrpd *rgd)
864 {
865 	struct gfs2_sbd *sdp = rgd->rd_sbd;
866 	struct rb_node **newn = &sdp->sd_rindex_tree.rb_node, *parent = NULL;
867 
868 	/* Figure out where to put new node */
869 	while (*newn) {
870 		struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd,
871 						  rd_node);
872 
873 		parent = *newn;
874 		if (rgd->rd_addr < cur->rd_addr)
875 			newn = &((*newn)->rb_left);
876 		else if (rgd->rd_addr > cur->rd_addr)
877 			newn = &((*newn)->rb_right);
878 		else
879 			return -EEXIST;
880 	}
881 
882 	rb_link_node(&rgd->rd_node, parent, newn);
883 	rb_insert_color(&rgd->rd_node, &sdp->sd_rindex_tree);
884 	sdp->sd_rgrps++;
885 	return 0;
886 }
887 
888 /**
889  * read_rindex_entry - Pull in a new resource index entry from the disk
890  * @ip: Pointer to the rindex inode
891  *
892  * Returns: 0 on success, > 0 on EOF, error code otherwise
893  */
894 
895 static int read_rindex_entry(struct gfs2_inode *ip)
896 {
897 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
898 	const unsigned bsize = sdp->sd_sb.sb_bsize;
899 	loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex);
900 	struct gfs2_rindex buf;
901 	int error;
902 	struct gfs2_rgrpd *rgd;
903 
904 	if (pos >= i_size_read(&ip->i_inode))
905 		return 1;
906 
907 	error = gfs2_internal_read(ip, (char *)&buf, &pos,
908 				   sizeof(struct gfs2_rindex));
909 
910 	if (error != sizeof(struct gfs2_rindex))
911 		return (error == 0) ? 1 : error;
912 
913 	rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS);
914 	error = -ENOMEM;
915 	if (!rgd)
916 		return error;
917 
918 	rgd->rd_sbd = sdp;
919 	rgd->rd_addr = be64_to_cpu(buf.ri_addr);
920 	rgd->rd_length = be32_to_cpu(buf.ri_length);
921 	rgd->rd_data0 = be64_to_cpu(buf.ri_data0);
922 	rgd->rd_data = be32_to_cpu(buf.ri_data);
923 	rgd->rd_bitbytes = be32_to_cpu(buf.ri_bitbytes);
924 	spin_lock_init(&rgd->rd_rsspin);
925 
926 	error = compute_bitstructs(rgd);
927 	if (error)
928 		goto fail;
929 
930 	error = gfs2_glock_get(sdp, rgd->rd_addr,
931 			       &gfs2_rgrp_glops, CREATE, &rgd->rd_gl);
932 	if (error)
933 		goto fail;
934 
935 	rgd->rd_gl->gl_object = rgd;
936 	rgd->rd_gl->gl_vm.start = (rgd->rd_addr * bsize) & PAGE_CACHE_MASK;
937 	rgd->rd_gl->gl_vm.end = PAGE_CACHE_ALIGN((rgd->rd_addr +
938 						  rgd->rd_length) * bsize) - 1;
939 	rgd->rd_rgl = (struct gfs2_rgrp_lvb *)rgd->rd_gl->gl_lksb.sb_lvbptr;
940 	rgd->rd_flags &= ~(GFS2_RDF_UPTODATE | GFS2_RDF_PREFERRED);
941 	if (rgd->rd_data > sdp->sd_max_rg_data)
942 		sdp->sd_max_rg_data = rgd->rd_data;
943 	spin_lock(&sdp->sd_rindex_spin);
944 	error = rgd_insert(rgd);
945 	spin_unlock(&sdp->sd_rindex_spin);
946 	if (!error)
947 		return 0;
948 
949 	error = 0; /* someone else read in the rgrp; free it and ignore it */
950 	gfs2_glock_put(rgd->rd_gl);
951 
952 fail:
953 	kfree(rgd->rd_bits);
954 	kmem_cache_free(gfs2_rgrpd_cachep, rgd);
955 	return error;
956 }
957 
958 /**
959  * set_rgrp_preferences - Run all the rgrps, selecting some we prefer to use
960  * @sdp: the GFS2 superblock
961  *
962  * The purpose of this function is to select a subset of the resource groups
963  * and mark them as PREFERRED. We do it in such a way that each node prefers
964  * to use a unique set of rgrps to minimize glock contention.
965  */
966 static void set_rgrp_preferences(struct gfs2_sbd *sdp)
967 {
968 	struct gfs2_rgrpd *rgd, *first;
969 	int i;
970 
971 	/* Skip an initial number of rgrps, based on this node's journal ID.
972 	   That should start each node out on its own set. */
973 	rgd = gfs2_rgrpd_get_first(sdp);
974 	for (i = 0; i < sdp->sd_lockstruct.ls_jid; i++)
975 		rgd = gfs2_rgrpd_get_next(rgd);
976 	first = rgd;
977 
978 	do {
979 		rgd->rd_flags |= GFS2_RDF_PREFERRED;
980 		for (i = 0; i < sdp->sd_journals; i++) {
981 			rgd = gfs2_rgrpd_get_next(rgd);
982 			if (!rgd || rgd == first)
983 				break;
984 		}
985 	} while (rgd && rgd != first);
986 }
987 
988 /**
989  * gfs2_ri_update - Pull in a new resource index from the disk
990  * @ip: pointer to the rindex inode
991  *
992  * Returns: 0 on successful update, error code otherwise
993  */
994 
995 static int gfs2_ri_update(struct gfs2_inode *ip)
996 {
997 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
998 	int error;
999 
1000 	do {
1001 		error = read_rindex_entry(ip);
1002 	} while (error == 0);
1003 
1004 	if (error < 0)
1005 		return error;
1006 
1007 	set_rgrp_preferences(sdp);
1008 
1009 	sdp->sd_rindex_uptodate = 1;
1010 	return 0;
1011 }
1012 
1013 /**
1014  * gfs2_rindex_update - Update the rindex if required
1015  * @sdp: The GFS2 superblock
1016  *
1017  * We grab a lock on the rindex inode to make sure that it doesn't
1018  * change whilst we are performing an operation. We keep this lock
1019  * for quite long periods of time compared to other locks. This
1020  * doesn't matter, since it is shared and it is very, very rarely
1021  * accessed in the exclusive mode (i.e. only when expanding the filesystem).
1022  *
1023  * This makes sure that we're using the latest copy of the resource index
1024  * special file, which might have been updated if someone expanded the
1025  * filesystem (via gfs2_grow utility), which adds new resource groups.
1026  *
1027  * Returns: 0 on succeess, error code otherwise
1028  */
1029 
1030 int gfs2_rindex_update(struct gfs2_sbd *sdp)
1031 {
1032 	struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex);
1033 	struct gfs2_glock *gl = ip->i_gl;
1034 	struct gfs2_holder ri_gh;
1035 	int error = 0;
1036 	int unlock_required = 0;
1037 
1038 	/* Read new copy from disk if we don't have the latest */
1039 	if (!sdp->sd_rindex_uptodate) {
1040 		if (!gfs2_glock_is_locked_by_me(gl)) {
1041 			error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &ri_gh);
1042 			if (error)
1043 				return error;
1044 			unlock_required = 1;
1045 		}
1046 		if (!sdp->sd_rindex_uptodate)
1047 			error = gfs2_ri_update(ip);
1048 		if (unlock_required)
1049 			gfs2_glock_dq_uninit(&ri_gh);
1050 	}
1051 
1052 	return error;
1053 }
1054 
1055 static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf)
1056 {
1057 	const struct gfs2_rgrp *str = buf;
1058 	u32 rg_flags;
1059 
1060 	rg_flags = be32_to_cpu(str->rg_flags);
1061 	rg_flags &= ~GFS2_RDF_MASK;
1062 	rgd->rd_flags &= GFS2_RDF_MASK;
1063 	rgd->rd_flags |= rg_flags;
1064 	rgd->rd_free = be32_to_cpu(str->rg_free);
1065 	rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes);
1066 	rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration);
1067 }
1068 
1069 static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf)
1070 {
1071 	struct gfs2_rgrp *str = buf;
1072 
1073 	str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK);
1074 	str->rg_free = cpu_to_be32(rgd->rd_free);
1075 	str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes);
1076 	str->__pad = cpu_to_be32(0);
1077 	str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration);
1078 	memset(&str->rg_reserved, 0, sizeof(str->rg_reserved));
1079 }
1080 
1081 static int gfs2_rgrp_lvb_valid(struct gfs2_rgrpd *rgd)
1082 {
1083 	struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl;
1084 	struct gfs2_rgrp *str = (struct gfs2_rgrp *)rgd->rd_bits[0].bi_bh->b_data;
1085 
1086 	if (rgl->rl_flags != str->rg_flags || rgl->rl_free != str->rg_free ||
1087 	    rgl->rl_dinodes != str->rg_dinodes ||
1088 	    rgl->rl_igeneration != str->rg_igeneration)
1089 		return 0;
1090 	return 1;
1091 }
1092 
1093 static void gfs2_rgrp_ondisk2lvb(struct gfs2_rgrp_lvb *rgl, const void *buf)
1094 {
1095 	const struct gfs2_rgrp *str = buf;
1096 
1097 	rgl->rl_magic = cpu_to_be32(GFS2_MAGIC);
1098 	rgl->rl_flags = str->rg_flags;
1099 	rgl->rl_free = str->rg_free;
1100 	rgl->rl_dinodes = str->rg_dinodes;
1101 	rgl->rl_igeneration = str->rg_igeneration;
1102 	rgl->__pad = 0UL;
1103 }
1104 
1105 static void update_rgrp_lvb_unlinked(struct gfs2_rgrpd *rgd, u32 change)
1106 {
1107 	struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl;
1108 	u32 unlinked = be32_to_cpu(rgl->rl_unlinked) + change;
1109 	rgl->rl_unlinked = cpu_to_be32(unlinked);
1110 }
1111 
1112 static u32 count_unlinked(struct gfs2_rgrpd *rgd)
1113 {
1114 	struct gfs2_bitmap *bi;
1115 	const u32 length = rgd->rd_length;
1116 	const u8 *buffer = NULL;
1117 	u32 i, goal, count = 0;
1118 
1119 	for (i = 0, bi = rgd->rd_bits; i < length; i++, bi++) {
1120 		goal = 0;
1121 		buffer = bi->bi_bh->b_data + bi->bi_offset;
1122 		WARN_ON(!buffer_uptodate(bi->bi_bh));
1123 		while (goal < bi->bi_len * GFS2_NBBY) {
1124 			goal = gfs2_bitfit(buffer, bi->bi_len, goal,
1125 					   GFS2_BLKST_UNLINKED);
1126 			if (goal == BFITNOENT)
1127 				break;
1128 			count++;
1129 			goal++;
1130 		}
1131 	}
1132 
1133 	return count;
1134 }
1135 
1136 
1137 /**
1138  * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps
1139  * @rgd: the struct gfs2_rgrpd describing the RG to read in
1140  *
1141  * Read in all of a Resource Group's header and bitmap blocks.
1142  * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps.
1143  *
1144  * Returns: errno
1145  */
1146 
1147 static int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd)
1148 {
1149 	struct gfs2_sbd *sdp = rgd->rd_sbd;
1150 	struct gfs2_glock *gl = rgd->rd_gl;
1151 	unsigned int length = rgd->rd_length;
1152 	struct gfs2_bitmap *bi;
1153 	unsigned int x, y;
1154 	int error;
1155 
1156 	if (rgd->rd_bits[0].bi_bh != NULL)
1157 		return 0;
1158 
1159 	for (x = 0; x < length; x++) {
1160 		bi = rgd->rd_bits + x;
1161 		error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, &bi->bi_bh);
1162 		if (error)
1163 			goto fail;
1164 	}
1165 
1166 	for (y = length; y--;) {
1167 		bi = rgd->rd_bits + y;
1168 		error = gfs2_meta_wait(sdp, bi->bi_bh);
1169 		if (error)
1170 			goto fail;
1171 		if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB :
1172 					      GFS2_METATYPE_RG)) {
1173 			error = -EIO;
1174 			goto fail;
1175 		}
1176 	}
1177 
1178 	if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) {
1179 		for (x = 0; x < length; x++)
1180 			clear_bit(GBF_FULL, &rgd->rd_bits[x].bi_flags);
1181 		gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data);
1182 		rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK);
1183 		rgd->rd_free_clone = rgd->rd_free;
1184 		/* max out the rgrp allocation failure point */
1185 		rgd->rd_extfail_pt = rgd->rd_free;
1186 	}
1187 	if (cpu_to_be32(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) {
1188 		rgd->rd_rgl->rl_unlinked = cpu_to_be32(count_unlinked(rgd));
1189 		gfs2_rgrp_ondisk2lvb(rgd->rd_rgl,
1190 				     rgd->rd_bits[0].bi_bh->b_data);
1191 	}
1192 	else if (sdp->sd_args.ar_rgrplvb) {
1193 		if (!gfs2_rgrp_lvb_valid(rgd)){
1194 			gfs2_consist_rgrpd(rgd);
1195 			error = -EIO;
1196 			goto fail;
1197 		}
1198 		if (rgd->rd_rgl->rl_unlinked == 0)
1199 			rgd->rd_flags &= ~GFS2_RDF_CHECK;
1200 	}
1201 	return 0;
1202 
1203 fail:
1204 	while (x--) {
1205 		bi = rgd->rd_bits + x;
1206 		brelse(bi->bi_bh);
1207 		bi->bi_bh = NULL;
1208 		gfs2_assert_warn(sdp, !bi->bi_clone);
1209 	}
1210 
1211 	return error;
1212 }
1213 
1214 static int update_rgrp_lvb(struct gfs2_rgrpd *rgd)
1215 {
1216 	u32 rl_flags;
1217 
1218 	if (rgd->rd_flags & GFS2_RDF_UPTODATE)
1219 		return 0;
1220 
1221 	if (cpu_to_be32(GFS2_MAGIC) != rgd->rd_rgl->rl_magic)
1222 		return gfs2_rgrp_bh_get(rgd);
1223 
1224 	rl_flags = be32_to_cpu(rgd->rd_rgl->rl_flags);
1225 	rl_flags &= ~GFS2_RDF_MASK;
1226 	rgd->rd_flags &= GFS2_RDF_MASK;
1227 	rgd->rd_flags |= (rl_flags | GFS2_RDF_UPTODATE | GFS2_RDF_CHECK);
1228 	if (rgd->rd_rgl->rl_unlinked == 0)
1229 		rgd->rd_flags &= ~GFS2_RDF_CHECK;
1230 	rgd->rd_free = be32_to_cpu(rgd->rd_rgl->rl_free);
1231 	rgd->rd_free_clone = rgd->rd_free;
1232 	rgd->rd_dinodes = be32_to_cpu(rgd->rd_rgl->rl_dinodes);
1233 	rgd->rd_igeneration = be64_to_cpu(rgd->rd_rgl->rl_igeneration);
1234 	return 0;
1235 }
1236 
1237 int gfs2_rgrp_go_lock(struct gfs2_holder *gh)
1238 {
1239 	struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object;
1240 	struct gfs2_sbd *sdp = rgd->rd_sbd;
1241 
1242 	if (gh->gh_flags & GL_SKIP && sdp->sd_args.ar_rgrplvb)
1243 		return 0;
1244 	return gfs2_rgrp_bh_get(rgd);
1245 }
1246 
1247 /**
1248  * gfs2_rgrp_brelse - Release RG bitmaps read in with gfs2_rgrp_bh_get()
1249  * @rgd: The resource group
1250  *
1251  */
1252 
1253 void gfs2_rgrp_brelse(struct gfs2_rgrpd *rgd)
1254 {
1255 	int x, length = rgd->rd_length;
1256 
1257 	for (x = 0; x < length; x++) {
1258 		struct gfs2_bitmap *bi = rgd->rd_bits + x;
1259 		if (bi->bi_bh) {
1260 			brelse(bi->bi_bh);
1261 			bi->bi_bh = NULL;
1262 		}
1263 	}
1264 
1265 }
1266 
1267 /**
1268  * gfs2_rgrp_go_unlock - Unlock a rgrp glock
1269  * @gh: The glock holder for the resource group
1270  *
1271  */
1272 
1273 void gfs2_rgrp_go_unlock(struct gfs2_holder *gh)
1274 {
1275 	struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object;
1276 	int demote_requested = test_bit(GLF_DEMOTE, &gh->gh_gl->gl_flags) |
1277 		test_bit(GLF_PENDING_DEMOTE, &gh->gh_gl->gl_flags);
1278 
1279 	if (rgd && demote_requested)
1280 		gfs2_rgrp_brelse(rgd);
1281 }
1282 
1283 int gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset,
1284 			     struct buffer_head *bh,
1285 			     const struct gfs2_bitmap *bi, unsigned minlen, u64 *ptrimmed)
1286 {
1287 	struct super_block *sb = sdp->sd_vfs;
1288 	u64 blk;
1289 	sector_t start = 0;
1290 	sector_t nr_blks = 0;
1291 	int rv;
1292 	unsigned int x;
1293 	u32 trimmed = 0;
1294 	u8 diff;
1295 
1296 	for (x = 0; x < bi->bi_len; x++) {
1297 		const u8 *clone = bi->bi_clone ? bi->bi_clone : bi->bi_bh->b_data;
1298 		clone += bi->bi_offset;
1299 		clone += x;
1300 		if (bh) {
1301 			const u8 *orig = bh->b_data + bi->bi_offset + x;
1302 			diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1));
1303 		} else {
1304 			diff = ~(*clone | (*clone >> 1));
1305 		}
1306 		diff &= 0x55;
1307 		if (diff == 0)
1308 			continue;
1309 		blk = offset + ((bi->bi_start + x) * GFS2_NBBY);
1310 		while(diff) {
1311 			if (diff & 1) {
1312 				if (nr_blks == 0)
1313 					goto start_new_extent;
1314 				if ((start + nr_blks) != blk) {
1315 					if (nr_blks >= minlen) {
1316 						rv = sb_issue_discard(sb,
1317 							start, nr_blks,
1318 							GFP_NOFS, 0);
1319 						if (rv)
1320 							goto fail;
1321 						trimmed += nr_blks;
1322 					}
1323 					nr_blks = 0;
1324 start_new_extent:
1325 					start = blk;
1326 				}
1327 				nr_blks++;
1328 			}
1329 			diff >>= 2;
1330 			blk++;
1331 		}
1332 	}
1333 	if (nr_blks >= minlen) {
1334 		rv = sb_issue_discard(sb, start, nr_blks, GFP_NOFS, 0);
1335 		if (rv)
1336 			goto fail;
1337 		trimmed += nr_blks;
1338 	}
1339 	if (ptrimmed)
1340 		*ptrimmed = trimmed;
1341 	return 0;
1342 
1343 fail:
1344 	if (sdp->sd_args.ar_discard)
1345 		fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv);
1346 	sdp->sd_args.ar_discard = 0;
1347 	return -EIO;
1348 }
1349 
1350 /**
1351  * gfs2_fitrim - Generate discard requests for unused bits of the filesystem
1352  * @filp: Any file on the filesystem
1353  * @argp: Pointer to the arguments (also used to pass result)
1354  *
1355  * Returns: 0 on success, otherwise error code
1356  */
1357 
1358 int gfs2_fitrim(struct file *filp, void __user *argp)
1359 {
1360 	struct inode *inode = file_inode(filp);
1361 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1362 	struct request_queue *q = bdev_get_queue(sdp->sd_vfs->s_bdev);
1363 	struct buffer_head *bh;
1364 	struct gfs2_rgrpd *rgd;
1365 	struct gfs2_rgrpd *rgd_end;
1366 	struct gfs2_holder gh;
1367 	struct fstrim_range r;
1368 	int ret = 0;
1369 	u64 amt;
1370 	u64 trimmed = 0;
1371 	u64 start, end, minlen;
1372 	unsigned int x;
1373 	unsigned bs_shift = sdp->sd_sb.sb_bsize_shift;
1374 
1375 	if (!capable(CAP_SYS_ADMIN))
1376 		return -EPERM;
1377 
1378 	if (!blk_queue_discard(q))
1379 		return -EOPNOTSUPP;
1380 
1381 	if (copy_from_user(&r, argp, sizeof(r)))
1382 		return -EFAULT;
1383 
1384 	ret = gfs2_rindex_update(sdp);
1385 	if (ret)
1386 		return ret;
1387 
1388 	start = r.start >> bs_shift;
1389 	end = start + (r.len >> bs_shift);
1390 	minlen = max_t(u64, r.minlen,
1391 		       q->limits.discard_granularity) >> bs_shift;
1392 
1393 	if (end <= start || minlen > sdp->sd_max_rg_data)
1394 		return -EINVAL;
1395 
1396 	rgd = gfs2_blk2rgrpd(sdp, start, 0);
1397 	rgd_end = gfs2_blk2rgrpd(sdp, end, 0);
1398 
1399 	if ((gfs2_rgrpd_get_first(sdp) == gfs2_rgrpd_get_next(rgd_end))
1400 	    && (start > rgd_end->rd_data0 + rgd_end->rd_data))
1401 		return -EINVAL; /* start is beyond the end of the fs */
1402 
1403 	while (1) {
1404 
1405 		ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1406 		if (ret)
1407 			goto out;
1408 
1409 		if (!(rgd->rd_flags & GFS2_RGF_TRIMMED)) {
1410 			/* Trim each bitmap in the rgrp */
1411 			for (x = 0; x < rgd->rd_length; x++) {
1412 				struct gfs2_bitmap *bi = rgd->rd_bits + x;
1413 				ret = gfs2_rgrp_send_discards(sdp,
1414 						rgd->rd_data0, NULL, bi, minlen,
1415 						&amt);
1416 				if (ret) {
1417 					gfs2_glock_dq_uninit(&gh);
1418 					goto out;
1419 				}
1420 				trimmed += amt;
1421 			}
1422 
1423 			/* Mark rgrp as having been trimmed */
1424 			ret = gfs2_trans_begin(sdp, RES_RG_HDR, 0);
1425 			if (ret == 0) {
1426 				bh = rgd->rd_bits[0].bi_bh;
1427 				rgd->rd_flags |= GFS2_RGF_TRIMMED;
1428 				gfs2_trans_add_meta(rgd->rd_gl, bh);
1429 				gfs2_rgrp_out(rgd, bh->b_data);
1430 				gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, bh->b_data);
1431 				gfs2_trans_end(sdp);
1432 			}
1433 		}
1434 		gfs2_glock_dq_uninit(&gh);
1435 
1436 		if (rgd == rgd_end)
1437 			break;
1438 
1439 		rgd = gfs2_rgrpd_get_next(rgd);
1440 	}
1441 
1442 out:
1443 	r.len = trimmed << bs_shift;
1444 	if (copy_to_user(argp, &r, sizeof(r)))
1445 		return -EFAULT;
1446 
1447 	return ret;
1448 }
1449 
1450 /**
1451  * rs_insert - insert a new multi-block reservation into the rgrp's rb_tree
1452  * @ip: the inode structure
1453  *
1454  */
1455 static void rs_insert(struct gfs2_inode *ip)
1456 {
1457 	struct rb_node **newn, *parent = NULL;
1458 	int rc;
1459 	struct gfs2_blkreserv *rs = ip->i_res;
1460 	struct gfs2_rgrpd *rgd = rs->rs_rbm.rgd;
1461 	u64 fsblock = gfs2_rbm_to_block(&rs->rs_rbm);
1462 
1463 	BUG_ON(gfs2_rs_active(rs));
1464 
1465 	spin_lock(&rgd->rd_rsspin);
1466 	newn = &rgd->rd_rstree.rb_node;
1467 	while (*newn) {
1468 		struct gfs2_blkreserv *cur =
1469 			rb_entry(*newn, struct gfs2_blkreserv, rs_node);
1470 
1471 		parent = *newn;
1472 		rc = rs_cmp(fsblock, rs->rs_free, cur);
1473 		if (rc > 0)
1474 			newn = &((*newn)->rb_right);
1475 		else if (rc < 0)
1476 			newn = &((*newn)->rb_left);
1477 		else {
1478 			spin_unlock(&rgd->rd_rsspin);
1479 			WARN_ON(1);
1480 			return;
1481 		}
1482 	}
1483 
1484 	rb_link_node(&rs->rs_node, parent, newn);
1485 	rb_insert_color(&rs->rs_node, &rgd->rd_rstree);
1486 
1487 	/* Do our rgrp accounting for the reservation */
1488 	rgd->rd_reserved += rs->rs_free; /* blocks reserved */
1489 	spin_unlock(&rgd->rd_rsspin);
1490 	trace_gfs2_rs(rs, TRACE_RS_INSERT);
1491 }
1492 
1493 /**
1494  * rg_mblk_search - find a group of multiple free blocks to form a reservation
1495  * @rgd: the resource group descriptor
1496  * @ip: pointer to the inode for which we're reserving blocks
1497  * @ap: the allocation parameters
1498  *
1499  */
1500 
1501 static void rg_mblk_search(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip,
1502 			   const struct gfs2_alloc_parms *ap)
1503 {
1504 	struct gfs2_rbm rbm = { .rgd = rgd, };
1505 	u64 goal;
1506 	struct gfs2_blkreserv *rs = ip->i_res;
1507 	u32 extlen;
1508 	u32 free_blocks = rgd->rd_free_clone - rgd->rd_reserved;
1509 	int ret;
1510 	struct inode *inode = &ip->i_inode;
1511 
1512 	if (S_ISDIR(inode->i_mode))
1513 		extlen = 1;
1514 	else {
1515 		extlen = max_t(u32, atomic_read(&rs->rs_sizehint), ap->target);
1516 		extlen = clamp(extlen, RGRP_RSRV_MINBLKS, free_blocks);
1517 	}
1518 	if ((rgd->rd_free_clone < rgd->rd_reserved) || (free_blocks < extlen))
1519 		return;
1520 
1521 	/* Find bitmap block that contains bits for goal block */
1522 	if (rgrp_contains_block(rgd, ip->i_goal))
1523 		goal = ip->i_goal;
1524 	else
1525 		goal = rgd->rd_last_alloc + rgd->rd_data0;
1526 
1527 	if (WARN_ON(gfs2_rbm_from_block(&rbm, goal)))
1528 		return;
1529 
1530 	ret = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, &extlen, ip, true, ap);
1531 	if (ret == 0) {
1532 		rs->rs_rbm = rbm;
1533 		rs->rs_free = extlen;
1534 		rs->rs_inum = ip->i_no_addr;
1535 		rs_insert(ip);
1536 	} else {
1537 		if (goal == rgd->rd_last_alloc + rgd->rd_data0)
1538 			rgd->rd_last_alloc = 0;
1539 	}
1540 }
1541 
1542 /**
1543  * gfs2_next_unreserved_block - Return next block that is not reserved
1544  * @rgd: The resource group
1545  * @block: The starting block
1546  * @length: The required length
1547  * @ip: Ignore any reservations for this inode
1548  *
1549  * If the block does not appear in any reservation, then return the
1550  * block number unchanged. If it does appear in the reservation, then
1551  * keep looking through the tree of reservations in order to find the
1552  * first block number which is not reserved.
1553  */
1554 
1555 static u64 gfs2_next_unreserved_block(struct gfs2_rgrpd *rgd, u64 block,
1556 				      u32 length,
1557 				      const struct gfs2_inode *ip)
1558 {
1559 	struct gfs2_blkreserv *rs;
1560 	struct rb_node *n;
1561 	int rc;
1562 
1563 	spin_lock(&rgd->rd_rsspin);
1564 	n = rgd->rd_rstree.rb_node;
1565 	while (n) {
1566 		rs = rb_entry(n, struct gfs2_blkreserv, rs_node);
1567 		rc = rs_cmp(block, length, rs);
1568 		if (rc < 0)
1569 			n = n->rb_left;
1570 		else if (rc > 0)
1571 			n = n->rb_right;
1572 		else
1573 			break;
1574 	}
1575 
1576 	if (n) {
1577 		while ((rs_cmp(block, length, rs) == 0) && (ip->i_res != rs)) {
1578 			block = gfs2_rbm_to_block(&rs->rs_rbm) + rs->rs_free;
1579 			n = n->rb_right;
1580 			if (n == NULL)
1581 				break;
1582 			rs = rb_entry(n, struct gfs2_blkreserv, rs_node);
1583 		}
1584 	}
1585 
1586 	spin_unlock(&rgd->rd_rsspin);
1587 	return block;
1588 }
1589 
1590 /**
1591  * gfs2_reservation_check_and_update - Check for reservations during block alloc
1592  * @rbm: The current position in the resource group
1593  * @ip: The inode for which we are searching for blocks
1594  * @minext: The minimum extent length
1595  * @maxext: A pointer to the maximum extent structure
1596  *
1597  * This checks the current position in the rgrp to see whether there is
1598  * a reservation covering this block. If not then this function is a
1599  * no-op. If there is, then the position is moved to the end of the
1600  * contiguous reservation(s) so that we are pointing at the first
1601  * non-reserved block.
1602  *
1603  * Returns: 0 if no reservation, 1 if @rbm has changed, otherwise an error
1604  */
1605 
1606 static int gfs2_reservation_check_and_update(struct gfs2_rbm *rbm,
1607 					     const struct gfs2_inode *ip,
1608 					     u32 minext,
1609 					     struct gfs2_extent *maxext)
1610 {
1611 	u64 block = gfs2_rbm_to_block(rbm);
1612 	u32 extlen = 1;
1613 	u64 nblock;
1614 	int ret;
1615 
1616 	/*
1617 	 * If we have a minimum extent length, then skip over any extent
1618 	 * which is less than the min extent length in size.
1619 	 */
1620 	if (minext) {
1621 		extlen = gfs2_free_extlen(rbm, minext);
1622 		if (extlen <= maxext->len)
1623 			goto fail;
1624 	}
1625 
1626 	/*
1627 	 * Check the extent which has been found against the reservations
1628 	 * and skip if parts of it are already reserved
1629 	 */
1630 	nblock = gfs2_next_unreserved_block(rbm->rgd, block, extlen, ip);
1631 	if (nblock == block) {
1632 		if (!minext || extlen >= minext)
1633 			return 0;
1634 
1635 		if (extlen > maxext->len) {
1636 			maxext->len = extlen;
1637 			maxext->rbm = *rbm;
1638 		}
1639 fail:
1640 		nblock = block + extlen;
1641 	}
1642 	ret = gfs2_rbm_from_block(rbm, nblock);
1643 	if (ret < 0)
1644 		return ret;
1645 	return 1;
1646 }
1647 
1648 /**
1649  * gfs2_rbm_find - Look for blocks of a particular state
1650  * @rbm: Value/result starting position and final position
1651  * @state: The state which we want to find
1652  * @minext: Pointer to the requested extent length (NULL for a single block)
1653  *          This is updated to be the actual reservation size.
1654  * @ip: If set, check for reservations
1655  * @nowrap: Stop looking at the end of the rgrp, rather than wrapping
1656  *          around until we've reached the starting point.
1657  * @ap: the allocation parameters
1658  *
1659  * Side effects:
1660  * - If looking for free blocks, we set GBF_FULL on each bitmap which
1661  *   has no free blocks in it.
1662  * - If looking for free blocks, we set rd_extfail_pt on each rgrp which
1663  *   has come up short on a free block search.
1664  *
1665  * Returns: 0 on success, -ENOSPC if there is no block of the requested state
1666  */
1667 
1668 static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 *minext,
1669 			 const struct gfs2_inode *ip, bool nowrap,
1670 			 const struct gfs2_alloc_parms *ap)
1671 {
1672 	struct buffer_head *bh;
1673 	int initial_bii;
1674 	u32 initial_offset;
1675 	int first_bii = rbm->bii;
1676 	u32 first_offset = rbm->offset;
1677 	u32 offset;
1678 	u8 *buffer;
1679 	int n = 0;
1680 	int iters = rbm->rgd->rd_length;
1681 	int ret;
1682 	struct gfs2_bitmap *bi;
1683 	struct gfs2_extent maxext = { .rbm.rgd = rbm->rgd, };
1684 
1685 	/* If we are not starting at the beginning of a bitmap, then we
1686 	 * need to add one to the bitmap count to ensure that we search
1687 	 * the starting bitmap twice.
1688 	 */
1689 	if (rbm->offset != 0)
1690 		iters++;
1691 
1692 	while(1) {
1693 		bi = rbm_bi(rbm);
1694 		if (test_bit(GBF_FULL, &bi->bi_flags) &&
1695 		    (state == GFS2_BLKST_FREE))
1696 			goto next_bitmap;
1697 
1698 		bh = bi->bi_bh;
1699 		buffer = bh->b_data + bi->bi_offset;
1700 		WARN_ON(!buffer_uptodate(bh));
1701 		if (state != GFS2_BLKST_UNLINKED && bi->bi_clone)
1702 			buffer = bi->bi_clone + bi->bi_offset;
1703 		initial_offset = rbm->offset;
1704 		offset = gfs2_bitfit(buffer, bi->bi_len, rbm->offset, state);
1705 		if (offset == BFITNOENT)
1706 			goto bitmap_full;
1707 		rbm->offset = offset;
1708 		if (ip == NULL)
1709 			return 0;
1710 
1711 		initial_bii = rbm->bii;
1712 		ret = gfs2_reservation_check_and_update(rbm, ip,
1713 							minext ? *minext : 0,
1714 							&maxext);
1715 		if (ret == 0)
1716 			return 0;
1717 		if (ret > 0) {
1718 			n += (rbm->bii - initial_bii);
1719 			goto next_iter;
1720 		}
1721 		if (ret == -E2BIG) {
1722 			rbm->bii = 0;
1723 			rbm->offset = 0;
1724 			n += (rbm->bii - initial_bii);
1725 			goto res_covered_end_of_rgrp;
1726 		}
1727 		return ret;
1728 
1729 bitmap_full:	/* Mark bitmap as full and fall through */
1730 		if ((state == GFS2_BLKST_FREE) && initial_offset == 0)
1731 			set_bit(GBF_FULL, &bi->bi_flags);
1732 
1733 next_bitmap:	/* Find next bitmap in the rgrp */
1734 		rbm->offset = 0;
1735 		rbm->bii++;
1736 		if (rbm->bii == rbm->rgd->rd_length)
1737 			rbm->bii = 0;
1738 res_covered_end_of_rgrp:
1739 		if ((rbm->bii == 0) && nowrap)
1740 			break;
1741 		n++;
1742 next_iter:
1743 		if (n >= iters)
1744 			break;
1745 	}
1746 
1747 	if (minext == NULL || state != GFS2_BLKST_FREE)
1748 		return -ENOSPC;
1749 
1750 	/* If the extent was too small, and it's smaller than the smallest
1751 	   to have failed before, remember for future reference that it's
1752 	   useless to search this rgrp again for this amount or more. */
1753 	if ((first_offset == 0) && (first_bii == 0) &&
1754 	    (*minext < rbm->rgd->rd_extfail_pt))
1755 		rbm->rgd->rd_extfail_pt = *minext;
1756 
1757 	/* If the maximum extent we found is big enough to fulfill the
1758 	   minimum requirements, use it anyway. */
1759 	if (maxext.len) {
1760 		*rbm = maxext.rbm;
1761 		*minext = maxext.len;
1762 		return 0;
1763 	}
1764 
1765 	return -ENOSPC;
1766 }
1767 
1768 /**
1769  * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes
1770  * @rgd: The rgrp
1771  * @last_unlinked: block address of the last dinode we unlinked
1772  * @skip: block address we should explicitly not unlink
1773  *
1774  * Returns: 0 if no error
1775  *          The inode, if one has been found, in inode.
1776  */
1777 
1778 static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip)
1779 {
1780 	u64 block;
1781 	struct gfs2_sbd *sdp = rgd->rd_sbd;
1782 	struct gfs2_glock *gl;
1783 	struct gfs2_inode *ip;
1784 	int error;
1785 	int found = 0;
1786 	struct gfs2_rbm rbm = { .rgd = rgd, .bii = 0, .offset = 0 };
1787 
1788 	while (1) {
1789 		down_write(&sdp->sd_log_flush_lock);
1790 		error = gfs2_rbm_find(&rbm, GFS2_BLKST_UNLINKED, NULL, NULL,
1791 				      true, NULL);
1792 		up_write(&sdp->sd_log_flush_lock);
1793 		if (error == -ENOSPC)
1794 			break;
1795 		if (WARN_ON_ONCE(error))
1796 			break;
1797 
1798 		block = gfs2_rbm_to_block(&rbm);
1799 		if (gfs2_rbm_from_block(&rbm, block + 1))
1800 			break;
1801 		if (*last_unlinked != NO_BLOCK && block <= *last_unlinked)
1802 			continue;
1803 		if (block == skip)
1804 			continue;
1805 		*last_unlinked = block;
1806 
1807 		error = gfs2_glock_get(sdp, block, &gfs2_inode_glops, CREATE, &gl);
1808 		if (error)
1809 			continue;
1810 
1811 		/* If the inode is already in cache, we can ignore it here
1812 		 * because the existing inode disposal code will deal with
1813 		 * it when all refs have gone away. Accessing gl_object like
1814 		 * this is not safe in general. Here it is ok because we do
1815 		 * not dereference the pointer, and we only need an approx
1816 		 * answer to whether it is NULL or not.
1817 		 */
1818 		ip = gl->gl_object;
1819 
1820 		if (ip || queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
1821 			gfs2_glock_put(gl);
1822 		else
1823 			found++;
1824 
1825 		/* Limit reclaim to sensible number of tasks */
1826 		if (found > NR_CPUS)
1827 			return;
1828 	}
1829 
1830 	rgd->rd_flags &= ~GFS2_RDF_CHECK;
1831 	return;
1832 }
1833 
1834 /**
1835  * gfs2_rgrp_congested - Use stats to figure out whether an rgrp is congested
1836  * @rgd: The rgrp in question
1837  * @loops: An indication of how picky we can be (0=very, 1=less so)
1838  *
1839  * This function uses the recently added glock statistics in order to
1840  * figure out whether a parciular resource group is suffering from
1841  * contention from multiple nodes. This is done purely on the basis
1842  * of timings, since this is the only data we have to work with and
1843  * our aim here is to reject a resource group which is highly contended
1844  * but (very important) not to do this too often in order to ensure that
1845  * we do not land up introducing fragmentation by changing resource
1846  * groups when not actually required.
1847  *
1848  * The calculation is fairly simple, we want to know whether the SRTTB
1849  * (i.e. smoothed round trip time for blocking operations) to acquire
1850  * the lock for this rgrp's glock is significantly greater than the
1851  * time taken for resource groups on average. We introduce a margin in
1852  * the form of the variable @var which is computed as the sum of the two
1853  * respective variences, and multiplied by a factor depending on @loops
1854  * and whether we have a lot of data to base the decision on. This is
1855  * then tested against the square difference of the means in order to
1856  * decide whether the result is statistically significant or not.
1857  *
1858  * Returns: A boolean verdict on the congestion status
1859  */
1860 
1861 static bool gfs2_rgrp_congested(const struct gfs2_rgrpd *rgd, int loops)
1862 {
1863 	const struct gfs2_glock *gl = rgd->rd_gl;
1864 	const struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1865 	struct gfs2_lkstats *st;
1866 	u64 r_dcount, l_dcount;
1867 	u64 l_srttb, a_srttb = 0;
1868 	s64 srttb_diff;
1869 	u64 sqr_diff;
1870 	u64 var;
1871 	int cpu, nonzero = 0;
1872 
1873 	preempt_disable();
1874 	for_each_present_cpu(cpu) {
1875 		st = &per_cpu_ptr(sdp->sd_lkstats, cpu)->lkstats[LM_TYPE_RGRP];
1876 		if (st->stats[GFS2_LKS_SRTTB]) {
1877 			a_srttb += st->stats[GFS2_LKS_SRTTB];
1878 			nonzero++;
1879 		}
1880 	}
1881 	st = &this_cpu_ptr(sdp->sd_lkstats)->lkstats[LM_TYPE_RGRP];
1882 	if (nonzero)
1883 		do_div(a_srttb, nonzero);
1884 	r_dcount = st->stats[GFS2_LKS_DCOUNT];
1885 	var = st->stats[GFS2_LKS_SRTTVARB] +
1886 	      gl->gl_stats.stats[GFS2_LKS_SRTTVARB];
1887 	preempt_enable();
1888 
1889 	l_srttb = gl->gl_stats.stats[GFS2_LKS_SRTTB];
1890 	l_dcount = gl->gl_stats.stats[GFS2_LKS_DCOUNT];
1891 
1892 	if ((l_dcount < 1) || (r_dcount < 1) || (a_srttb == 0))
1893 		return false;
1894 
1895 	srttb_diff = a_srttb - l_srttb;
1896 	sqr_diff = srttb_diff * srttb_diff;
1897 
1898 	var *= 2;
1899 	if (l_dcount < 8 || r_dcount < 8)
1900 		var *= 2;
1901 	if (loops == 1)
1902 		var *= 2;
1903 
1904 	return ((srttb_diff < 0) && (sqr_diff > var));
1905 }
1906 
1907 /**
1908  * gfs2_rgrp_used_recently
1909  * @rs: The block reservation with the rgrp to test
1910  * @msecs: The time limit in milliseconds
1911  *
1912  * Returns: True if the rgrp glock has been used within the time limit
1913  */
1914 static bool gfs2_rgrp_used_recently(const struct gfs2_blkreserv *rs,
1915 				    u64 msecs)
1916 {
1917 	u64 tdiff;
1918 
1919 	tdiff = ktime_to_ns(ktime_sub(ktime_get_real(),
1920                             rs->rs_rbm.rgd->rd_gl->gl_dstamp));
1921 
1922 	return tdiff > (msecs * 1000 * 1000);
1923 }
1924 
1925 static u32 gfs2_orlov_skip(const struct gfs2_inode *ip)
1926 {
1927 	const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1928 	u32 skip;
1929 
1930 	get_random_bytes(&skip, sizeof(skip));
1931 	return skip % sdp->sd_rgrps;
1932 }
1933 
1934 static bool gfs2_select_rgrp(struct gfs2_rgrpd **pos, const struct gfs2_rgrpd *begin)
1935 {
1936 	struct gfs2_rgrpd *rgd = *pos;
1937 	struct gfs2_sbd *sdp = rgd->rd_sbd;
1938 
1939 	rgd = gfs2_rgrpd_get_next(rgd);
1940 	if (rgd == NULL)
1941 		rgd = gfs2_rgrpd_get_first(sdp);
1942 	*pos = rgd;
1943 	if (rgd != begin) /* If we didn't wrap */
1944 		return true;
1945 	return false;
1946 }
1947 
1948 /**
1949  * fast_to_acquire - determine if a resource group will be fast to acquire
1950  *
1951  * If this is one of our preferred rgrps, it should be quicker to acquire,
1952  * because we tried to set ourselves up as dlm lock master.
1953  */
1954 static inline int fast_to_acquire(struct gfs2_rgrpd *rgd)
1955 {
1956 	struct gfs2_glock *gl = rgd->rd_gl;
1957 
1958 	if (gl->gl_state != LM_ST_UNLOCKED && list_empty(&gl->gl_holders) &&
1959 	    !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
1960 	    !test_bit(GLF_DEMOTE, &gl->gl_flags))
1961 		return 1;
1962 	if (rgd->rd_flags & GFS2_RDF_PREFERRED)
1963 		return 1;
1964 	return 0;
1965 }
1966 
1967 /**
1968  * gfs2_inplace_reserve - Reserve space in the filesystem
1969  * @ip: the inode to reserve space for
1970  * @ap: the allocation parameters
1971  *
1972  * We try our best to find an rgrp that has at least ap->target blocks
1973  * available. After a couple of passes (loops == 2), the prospects of finding
1974  * such an rgrp diminish. At this stage, we return the first rgrp that has
1975  * atleast ap->min_target blocks available. Either way, we set ap->allowed to
1976  * the number of blocks available in the chosen rgrp.
1977  *
1978  * Returns: 0 on success,
1979  *          -ENOMEM if a suitable rgrp can't be found
1980  *          errno otherwise
1981  */
1982 
1983 int gfs2_inplace_reserve(struct gfs2_inode *ip, struct gfs2_alloc_parms *ap)
1984 {
1985 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1986 	struct gfs2_rgrpd *begin = NULL;
1987 	struct gfs2_blkreserv *rs = ip->i_res;
1988 	int error = 0, rg_locked, flags = 0;
1989 	u64 last_unlinked = NO_BLOCK;
1990 	int loops = 0;
1991 	u32 skip = 0;
1992 
1993 	if (sdp->sd_args.ar_rgrplvb)
1994 		flags |= GL_SKIP;
1995 	if (gfs2_assert_warn(sdp, ap->target))
1996 		return -EINVAL;
1997 	if (gfs2_rs_active(rs)) {
1998 		begin = rs->rs_rbm.rgd;
1999 	} else if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, ip->i_goal)) {
2000 		rs->rs_rbm.rgd = begin = ip->i_rgd;
2001 	} else {
2002 		check_and_update_goal(ip);
2003 		rs->rs_rbm.rgd = begin = gfs2_blk2rgrpd(sdp, ip->i_goal, 1);
2004 	}
2005 	if (S_ISDIR(ip->i_inode.i_mode) && (ap->aflags & GFS2_AF_ORLOV))
2006 		skip = gfs2_orlov_skip(ip);
2007 	if (rs->rs_rbm.rgd == NULL)
2008 		return -EBADSLT;
2009 
2010 	while (loops < 3) {
2011 		rg_locked = 1;
2012 
2013 		if (!gfs2_glock_is_locked_by_me(rs->rs_rbm.rgd->rd_gl)) {
2014 			rg_locked = 0;
2015 			if (skip && skip--)
2016 				goto next_rgrp;
2017 			if (!gfs2_rs_active(rs)) {
2018 				if (loops == 0 &&
2019 				    !fast_to_acquire(rs->rs_rbm.rgd))
2020 					goto next_rgrp;
2021 				if ((loops < 2) &&
2022 				    gfs2_rgrp_used_recently(rs, 1000) &&
2023 				    gfs2_rgrp_congested(rs->rs_rbm.rgd, loops))
2024 					goto next_rgrp;
2025 			}
2026 			error = gfs2_glock_nq_init(rs->rs_rbm.rgd->rd_gl,
2027 						   LM_ST_EXCLUSIVE, flags,
2028 						   &rs->rs_rgd_gh);
2029 			if (unlikely(error))
2030 				return error;
2031 			if (!gfs2_rs_active(rs) && (loops < 2) &&
2032 			    gfs2_rgrp_congested(rs->rs_rbm.rgd, loops))
2033 				goto skip_rgrp;
2034 			if (sdp->sd_args.ar_rgrplvb) {
2035 				error = update_rgrp_lvb(rs->rs_rbm.rgd);
2036 				if (unlikely(error)) {
2037 					gfs2_glock_dq_uninit(&rs->rs_rgd_gh);
2038 					return error;
2039 				}
2040 			}
2041 		}
2042 
2043 		/* Skip unuseable resource groups */
2044 		if ((rs->rs_rbm.rgd->rd_flags & (GFS2_RGF_NOALLOC |
2045 						 GFS2_RDF_ERROR)) ||
2046 		    (loops == 0 && ap->target > rs->rs_rbm.rgd->rd_extfail_pt))
2047 			goto skip_rgrp;
2048 
2049 		if (sdp->sd_args.ar_rgrplvb)
2050 			gfs2_rgrp_bh_get(rs->rs_rbm.rgd);
2051 
2052 		/* Get a reservation if we don't already have one */
2053 		if (!gfs2_rs_active(rs))
2054 			rg_mblk_search(rs->rs_rbm.rgd, ip, ap);
2055 
2056 		/* Skip rgrps when we can't get a reservation on first pass */
2057 		if (!gfs2_rs_active(rs) && (loops < 1))
2058 			goto check_rgrp;
2059 
2060 		/* If rgrp has enough free space, use it */
2061 		if (rs->rs_rbm.rgd->rd_free_clone >= ap->target ||
2062 		    (loops == 2 && ap->min_target &&
2063 		     rs->rs_rbm.rgd->rd_free_clone >= ap->min_target)) {
2064 			ip->i_rgd = rs->rs_rbm.rgd;
2065 			ap->allowed = ip->i_rgd->rd_free_clone;
2066 			return 0;
2067 		}
2068 check_rgrp:
2069 		/* Check for unlinked inodes which can be reclaimed */
2070 		if (rs->rs_rbm.rgd->rd_flags & GFS2_RDF_CHECK)
2071 			try_rgrp_unlink(rs->rs_rbm.rgd, &last_unlinked,
2072 					ip->i_no_addr);
2073 skip_rgrp:
2074 		/* Drop reservation, if we couldn't use reserved rgrp */
2075 		if (gfs2_rs_active(rs))
2076 			gfs2_rs_deltree(rs);
2077 
2078 		/* Unlock rgrp if required */
2079 		if (!rg_locked)
2080 			gfs2_glock_dq_uninit(&rs->rs_rgd_gh);
2081 next_rgrp:
2082 		/* Find the next rgrp, and continue looking */
2083 		if (gfs2_select_rgrp(&rs->rs_rbm.rgd, begin))
2084 			continue;
2085 		if (skip)
2086 			continue;
2087 
2088 		/* If we've scanned all the rgrps, but found no free blocks
2089 		 * then this checks for some less likely conditions before
2090 		 * trying again.
2091 		 */
2092 		loops++;
2093 		/* Check that fs hasn't grown if writing to rindex */
2094 		if (ip == GFS2_I(sdp->sd_rindex) && !sdp->sd_rindex_uptodate) {
2095 			error = gfs2_ri_update(ip);
2096 			if (error)
2097 				return error;
2098 		}
2099 		/* Flushing the log may release space */
2100 		if (loops == 2)
2101 			gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
2102 	}
2103 
2104 	return -ENOSPC;
2105 }
2106 
2107 /**
2108  * gfs2_inplace_release - release an inplace reservation
2109  * @ip: the inode the reservation was taken out on
2110  *
2111  * Release a reservation made by gfs2_inplace_reserve().
2112  */
2113 
2114 void gfs2_inplace_release(struct gfs2_inode *ip)
2115 {
2116 	struct gfs2_blkreserv *rs = ip->i_res;
2117 
2118 	if (rs->rs_rgd_gh.gh_gl)
2119 		gfs2_glock_dq_uninit(&rs->rs_rgd_gh);
2120 }
2121 
2122 /**
2123  * gfs2_get_block_type - Check a block in a RG is of given type
2124  * @rgd: the resource group holding the block
2125  * @block: the block number
2126  *
2127  * Returns: The block type (GFS2_BLKST_*)
2128  */
2129 
2130 static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block)
2131 {
2132 	struct gfs2_rbm rbm = { .rgd = rgd, };
2133 	int ret;
2134 
2135 	ret = gfs2_rbm_from_block(&rbm, block);
2136 	WARN_ON_ONCE(ret != 0);
2137 
2138 	return gfs2_testbit(&rbm);
2139 }
2140 
2141 
2142 /**
2143  * gfs2_alloc_extent - allocate an extent from a given bitmap
2144  * @rbm: the resource group information
2145  * @dinode: TRUE if the first block we allocate is for a dinode
2146  * @n: The extent length (value/result)
2147  *
2148  * Add the bitmap buffer to the transaction.
2149  * Set the found bits to @new_state to change block's allocation state.
2150  */
2151 static void gfs2_alloc_extent(const struct gfs2_rbm *rbm, bool dinode,
2152 			     unsigned int *n)
2153 {
2154 	struct gfs2_rbm pos = { .rgd = rbm->rgd, };
2155 	const unsigned int elen = *n;
2156 	u64 block;
2157 	int ret;
2158 
2159 	*n = 1;
2160 	block = gfs2_rbm_to_block(rbm);
2161 	gfs2_trans_add_meta(rbm->rgd->rd_gl, rbm_bi(rbm)->bi_bh);
2162 	gfs2_setbit(rbm, true, dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED);
2163 	block++;
2164 	while (*n < elen) {
2165 		ret = gfs2_rbm_from_block(&pos, block);
2166 		if (ret || gfs2_testbit(&pos) != GFS2_BLKST_FREE)
2167 			break;
2168 		gfs2_trans_add_meta(pos.rgd->rd_gl, rbm_bi(&pos)->bi_bh);
2169 		gfs2_setbit(&pos, true, GFS2_BLKST_USED);
2170 		(*n)++;
2171 		block++;
2172 	}
2173 }
2174 
2175 /**
2176  * rgblk_free - Change alloc state of given block(s)
2177  * @sdp: the filesystem
2178  * @bstart: the start of a run of blocks to free
2179  * @blen: the length of the block run (all must lie within ONE RG!)
2180  * @new_state: GFS2_BLKST_XXX the after-allocation block state
2181  *
2182  * Returns:  Resource group containing the block(s)
2183  */
2184 
2185 static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart,
2186 				     u32 blen, unsigned char new_state)
2187 {
2188 	struct gfs2_rbm rbm;
2189 	struct gfs2_bitmap *bi, *bi_prev = NULL;
2190 
2191 	rbm.rgd = gfs2_blk2rgrpd(sdp, bstart, 1);
2192 	if (!rbm.rgd) {
2193 		if (gfs2_consist(sdp))
2194 			fs_err(sdp, "block = %llu\n", (unsigned long long)bstart);
2195 		return NULL;
2196 	}
2197 
2198 	gfs2_rbm_from_block(&rbm, bstart);
2199 	while (blen--) {
2200 		bi = rbm_bi(&rbm);
2201 		if (bi != bi_prev) {
2202 			if (!bi->bi_clone) {
2203 				bi->bi_clone = kmalloc(bi->bi_bh->b_size,
2204 						      GFP_NOFS | __GFP_NOFAIL);
2205 				memcpy(bi->bi_clone + bi->bi_offset,
2206 				       bi->bi_bh->b_data + bi->bi_offset,
2207 				       bi->bi_len);
2208 			}
2209 			gfs2_trans_add_meta(rbm.rgd->rd_gl, bi->bi_bh);
2210 			bi_prev = bi;
2211 		}
2212 		gfs2_setbit(&rbm, false, new_state);
2213 		gfs2_rbm_incr(&rbm);
2214 	}
2215 
2216 	return rbm.rgd;
2217 }
2218 
2219 /**
2220  * gfs2_rgrp_dump - print out an rgrp
2221  * @seq: The iterator
2222  * @gl: The glock in question
2223  *
2224  */
2225 
2226 void gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl)
2227 {
2228 	struct gfs2_rgrpd *rgd = gl->gl_object;
2229 	struct gfs2_blkreserv *trs;
2230 	const struct rb_node *n;
2231 
2232 	if (rgd == NULL)
2233 		return;
2234 	gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u r:%u e:%u\n",
2235 		       (unsigned long long)rgd->rd_addr, rgd->rd_flags,
2236 		       rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes,
2237 		       rgd->rd_reserved, rgd->rd_extfail_pt);
2238 	spin_lock(&rgd->rd_rsspin);
2239 	for (n = rb_first(&rgd->rd_rstree); n; n = rb_next(&trs->rs_node)) {
2240 		trs = rb_entry(n, struct gfs2_blkreserv, rs_node);
2241 		dump_rs(seq, trs);
2242 	}
2243 	spin_unlock(&rgd->rd_rsspin);
2244 }
2245 
2246 static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd)
2247 {
2248 	struct gfs2_sbd *sdp = rgd->rd_sbd;
2249 	fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n",
2250 		(unsigned long long)rgd->rd_addr);
2251 	fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n");
2252 	gfs2_rgrp_dump(NULL, rgd->rd_gl);
2253 	rgd->rd_flags |= GFS2_RDF_ERROR;
2254 }
2255 
2256 /**
2257  * gfs2_adjust_reservation - Adjust (or remove) a reservation after allocation
2258  * @ip: The inode we have just allocated blocks for
2259  * @rbm: The start of the allocated blocks
2260  * @len: The extent length
2261  *
2262  * Adjusts a reservation after an allocation has taken place. If the
2263  * reservation does not match the allocation, or if it is now empty
2264  * then it is removed.
2265  */
2266 
2267 static void gfs2_adjust_reservation(struct gfs2_inode *ip,
2268 				    const struct gfs2_rbm *rbm, unsigned len)
2269 {
2270 	struct gfs2_blkreserv *rs = ip->i_res;
2271 	struct gfs2_rgrpd *rgd = rbm->rgd;
2272 	unsigned rlen;
2273 	u64 block;
2274 	int ret;
2275 
2276 	spin_lock(&rgd->rd_rsspin);
2277 	if (gfs2_rs_active(rs)) {
2278 		if (gfs2_rbm_eq(&rs->rs_rbm, rbm)) {
2279 			block = gfs2_rbm_to_block(rbm);
2280 			ret = gfs2_rbm_from_block(&rs->rs_rbm, block + len);
2281 			rlen = min(rs->rs_free, len);
2282 			rs->rs_free -= rlen;
2283 			rgd->rd_reserved -= rlen;
2284 			trace_gfs2_rs(rs, TRACE_RS_CLAIM);
2285 			if (rs->rs_free && !ret)
2286 				goto out;
2287 			/* We used up our block reservation, so we should
2288 			   reserve more blocks next time. */
2289 			atomic_add(RGRP_RSRV_ADDBLKS, &rs->rs_sizehint);
2290 		}
2291 		__rs_deltree(rs);
2292 	}
2293 out:
2294 	spin_unlock(&rgd->rd_rsspin);
2295 }
2296 
2297 /**
2298  * gfs2_set_alloc_start - Set starting point for block allocation
2299  * @rbm: The rbm which will be set to the required location
2300  * @ip: The gfs2 inode
2301  * @dinode: Flag to say if allocation includes a new inode
2302  *
2303  * This sets the starting point from the reservation if one is active
2304  * otherwise it falls back to guessing a start point based on the
2305  * inode's goal block or the last allocation point in the rgrp.
2306  */
2307 
2308 static void gfs2_set_alloc_start(struct gfs2_rbm *rbm,
2309 				 const struct gfs2_inode *ip, bool dinode)
2310 {
2311 	u64 goal;
2312 
2313 	if (gfs2_rs_active(ip->i_res)) {
2314 		*rbm = ip->i_res->rs_rbm;
2315 		return;
2316 	}
2317 
2318 	if (!dinode && rgrp_contains_block(rbm->rgd, ip->i_goal))
2319 		goal = ip->i_goal;
2320 	else
2321 		goal = rbm->rgd->rd_last_alloc + rbm->rgd->rd_data0;
2322 
2323 	gfs2_rbm_from_block(rbm, goal);
2324 }
2325 
2326 /**
2327  * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode
2328  * @ip: the inode to allocate the block for
2329  * @bn: Used to return the starting block number
2330  * @nblocks: requested number of blocks/extent length (value/result)
2331  * @dinode: 1 if we're allocating a dinode block, else 0
2332  * @generation: the generation number of the inode
2333  *
2334  * Returns: 0 or error
2335  */
2336 
2337 int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *nblocks,
2338 		      bool dinode, u64 *generation)
2339 {
2340 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2341 	struct buffer_head *dibh;
2342 	struct gfs2_rbm rbm = { .rgd = ip->i_rgd, };
2343 	unsigned int ndata;
2344 	u64 block; /* block, within the file system scope */
2345 	int error;
2346 
2347 	gfs2_set_alloc_start(&rbm, ip, dinode);
2348 	error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, NULL, ip, false, NULL);
2349 
2350 	if (error == -ENOSPC) {
2351 		gfs2_set_alloc_start(&rbm, ip, dinode);
2352 		error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, NULL, NULL, false,
2353 				      NULL);
2354 	}
2355 
2356 	/* Since all blocks are reserved in advance, this shouldn't happen */
2357 	if (error) {
2358 		fs_warn(sdp, "inum=%llu error=%d, nblocks=%u, full=%d fail_pt=%d\n",
2359 			(unsigned long long)ip->i_no_addr, error, *nblocks,
2360 			test_bit(GBF_FULL, &rbm.rgd->rd_bits->bi_flags),
2361 			rbm.rgd->rd_extfail_pt);
2362 		goto rgrp_error;
2363 	}
2364 
2365 	gfs2_alloc_extent(&rbm, dinode, nblocks);
2366 	block = gfs2_rbm_to_block(&rbm);
2367 	rbm.rgd->rd_last_alloc = block - rbm.rgd->rd_data0;
2368 	if (gfs2_rs_active(ip->i_res))
2369 		gfs2_adjust_reservation(ip, &rbm, *nblocks);
2370 	ndata = *nblocks;
2371 	if (dinode)
2372 		ndata--;
2373 
2374 	if (!dinode) {
2375 		ip->i_goal = block + ndata - 1;
2376 		error = gfs2_meta_inode_buffer(ip, &dibh);
2377 		if (error == 0) {
2378 			struct gfs2_dinode *di =
2379 				(struct gfs2_dinode *)dibh->b_data;
2380 			gfs2_trans_add_meta(ip->i_gl, dibh);
2381 			di->di_goal_meta = di->di_goal_data =
2382 				cpu_to_be64(ip->i_goal);
2383 			brelse(dibh);
2384 		}
2385 	}
2386 	if (rbm.rgd->rd_free < *nblocks) {
2387 		pr_warn("nblocks=%u\n", *nblocks);
2388 		goto rgrp_error;
2389 	}
2390 
2391 	rbm.rgd->rd_free -= *nblocks;
2392 	if (dinode) {
2393 		rbm.rgd->rd_dinodes++;
2394 		*generation = rbm.rgd->rd_igeneration++;
2395 		if (*generation == 0)
2396 			*generation = rbm.rgd->rd_igeneration++;
2397 	}
2398 
2399 	gfs2_trans_add_meta(rbm.rgd->rd_gl, rbm.rgd->rd_bits[0].bi_bh);
2400 	gfs2_rgrp_out(rbm.rgd, rbm.rgd->rd_bits[0].bi_bh->b_data);
2401 	gfs2_rgrp_ondisk2lvb(rbm.rgd->rd_rgl, rbm.rgd->rd_bits[0].bi_bh->b_data);
2402 
2403 	gfs2_statfs_change(sdp, 0, -(s64)*nblocks, dinode ? 1 : 0);
2404 	if (dinode)
2405 		gfs2_trans_add_unrevoke(sdp, block, *nblocks);
2406 
2407 	gfs2_quota_change(ip, *nblocks, ip->i_inode.i_uid, ip->i_inode.i_gid);
2408 
2409 	rbm.rgd->rd_free_clone -= *nblocks;
2410 	trace_gfs2_block_alloc(ip, rbm.rgd, block, *nblocks,
2411 			       dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED);
2412 	*bn = block;
2413 	return 0;
2414 
2415 rgrp_error:
2416 	gfs2_rgrp_error(rbm.rgd);
2417 	return -EIO;
2418 }
2419 
2420 /**
2421  * __gfs2_free_blocks - free a contiguous run of block(s)
2422  * @ip: the inode these blocks are being freed from
2423  * @bstart: first block of a run of contiguous blocks
2424  * @blen: the length of the block run
2425  * @meta: 1 if the blocks represent metadata
2426  *
2427  */
2428 
2429 void __gfs2_free_blocks(struct gfs2_inode *ip, u64 bstart, u32 blen, int meta)
2430 {
2431 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2432 	struct gfs2_rgrpd *rgd;
2433 
2434 	rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE);
2435 	if (!rgd)
2436 		return;
2437 	trace_gfs2_block_alloc(ip, rgd, bstart, blen, GFS2_BLKST_FREE);
2438 	rgd->rd_free += blen;
2439 	rgd->rd_flags &= ~GFS2_RGF_TRIMMED;
2440 	gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh);
2441 	gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
2442 	gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data);
2443 
2444 	/* Directories keep their data in the metadata address space */
2445 	if (meta || ip->i_depth)
2446 		gfs2_meta_wipe(ip, bstart, blen);
2447 }
2448 
2449 /**
2450  * gfs2_free_meta - free a contiguous run of data block(s)
2451  * @ip: the inode these blocks are being freed from
2452  * @bstart: first block of a run of contiguous blocks
2453  * @blen: the length of the block run
2454  *
2455  */
2456 
2457 void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen)
2458 {
2459 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2460 
2461 	__gfs2_free_blocks(ip, bstart, blen, 1);
2462 	gfs2_statfs_change(sdp, 0, +blen, 0);
2463 	gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid);
2464 }
2465 
2466 void gfs2_unlink_di(struct inode *inode)
2467 {
2468 	struct gfs2_inode *ip = GFS2_I(inode);
2469 	struct gfs2_sbd *sdp = GFS2_SB(inode);
2470 	struct gfs2_rgrpd *rgd;
2471 	u64 blkno = ip->i_no_addr;
2472 
2473 	rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED);
2474 	if (!rgd)
2475 		return;
2476 	trace_gfs2_block_alloc(ip, rgd, blkno, 1, GFS2_BLKST_UNLINKED);
2477 	gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh);
2478 	gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
2479 	gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data);
2480 	update_rgrp_lvb_unlinked(rgd, 1);
2481 }
2482 
2483 static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno)
2484 {
2485 	struct gfs2_sbd *sdp = rgd->rd_sbd;
2486 	struct gfs2_rgrpd *tmp_rgd;
2487 
2488 	tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE);
2489 	if (!tmp_rgd)
2490 		return;
2491 	gfs2_assert_withdraw(sdp, rgd == tmp_rgd);
2492 
2493 	if (!rgd->rd_dinodes)
2494 		gfs2_consist_rgrpd(rgd);
2495 	rgd->rd_dinodes--;
2496 	rgd->rd_free++;
2497 
2498 	gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh);
2499 	gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
2500 	gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data);
2501 	update_rgrp_lvb_unlinked(rgd, -1);
2502 
2503 	gfs2_statfs_change(sdp, 0, +1, -1);
2504 }
2505 
2506 
2507 void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip)
2508 {
2509 	gfs2_free_uninit_di(rgd, ip->i_no_addr);
2510 	trace_gfs2_block_alloc(ip, rgd, ip->i_no_addr, 1, GFS2_BLKST_FREE);
2511 	gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid);
2512 	gfs2_meta_wipe(ip, ip->i_no_addr, 1);
2513 }
2514 
2515 /**
2516  * gfs2_check_blk_type - Check the type of a block
2517  * @sdp: The superblock
2518  * @no_addr: The block number to check
2519  * @type: The block type we are looking for
2520  *
2521  * Returns: 0 if the block type matches the expected type
2522  *          -ESTALE if it doesn't match
2523  *          or -ve errno if something went wrong while checking
2524  */
2525 
2526 int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type)
2527 {
2528 	struct gfs2_rgrpd *rgd;
2529 	struct gfs2_holder rgd_gh;
2530 	int error = -EINVAL;
2531 
2532 	rgd = gfs2_blk2rgrpd(sdp, no_addr, 1);
2533 	if (!rgd)
2534 		goto fail;
2535 
2536 	error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh);
2537 	if (error)
2538 		goto fail;
2539 
2540 	if (gfs2_get_block_type(rgd, no_addr) != type)
2541 		error = -ESTALE;
2542 
2543 	gfs2_glock_dq_uninit(&rgd_gh);
2544 fail:
2545 	return error;
2546 }
2547 
2548 /**
2549  * gfs2_rlist_add - add a RG to a list of RGs
2550  * @ip: the inode
2551  * @rlist: the list of resource groups
2552  * @block: the block
2553  *
2554  * Figure out what RG a block belongs to and add that RG to the list
2555  *
2556  * FIXME: Don't use NOFAIL
2557  *
2558  */
2559 
2560 void gfs2_rlist_add(struct gfs2_inode *ip, struct gfs2_rgrp_list *rlist,
2561 		    u64 block)
2562 {
2563 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2564 	struct gfs2_rgrpd *rgd;
2565 	struct gfs2_rgrpd **tmp;
2566 	unsigned int new_space;
2567 	unsigned int x;
2568 
2569 	if (gfs2_assert_warn(sdp, !rlist->rl_ghs))
2570 		return;
2571 
2572 	if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, block))
2573 		rgd = ip->i_rgd;
2574 	else
2575 		rgd = gfs2_blk2rgrpd(sdp, block, 1);
2576 	if (!rgd) {
2577 		fs_err(sdp, "rlist_add: no rgrp for block %llu\n", (unsigned long long)block);
2578 		return;
2579 	}
2580 	ip->i_rgd = rgd;
2581 
2582 	for (x = 0; x < rlist->rl_rgrps; x++)
2583 		if (rlist->rl_rgd[x] == rgd)
2584 			return;
2585 
2586 	if (rlist->rl_rgrps == rlist->rl_space) {
2587 		new_space = rlist->rl_space + 10;
2588 
2589 		tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *),
2590 			      GFP_NOFS | __GFP_NOFAIL);
2591 
2592 		if (rlist->rl_rgd) {
2593 			memcpy(tmp, rlist->rl_rgd,
2594 			       rlist->rl_space * sizeof(struct gfs2_rgrpd *));
2595 			kfree(rlist->rl_rgd);
2596 		}
2597 
2598 		rlist->rl_space = new_space;
2599 		rlist->rl_rgd = tmp;
2600 	}
2601 
2602 	rlist->rl_rgd[rlist->rl_rgrps++] = rgd;
2603 }
2604 
2605 /**
2606  * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate
2607  *      and initialize an array of glock holders for them
2608  * @rlist: the list of resource groups
2609  * @state: the lock state to acquire the RG lock in
2610  *
2611  * FIXME: Don't use NOFAIL
2612  *
2613  */
2614 
2615 void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state)
2616 {
2617 	unsigned int x;
2618 
2619 	rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder),
2620 				GFP_NOFS | __GFP_NOFAIL);
2621 	for (x = 0; x < rlist->rl_rgrps; x++)
2622 		gfs2_holder_init(rlist->rl_rgd[x]->rd_gl,
2623 				state, 0,
2624 				&rlist->rl_ghs[x]);
2625 }
2626 
2627 /**
2628  * gfs2_rlist_free - free a resource group list
2629  * @rlist: the list of resource groups
2630  *
2631  */
2632 
2633 void gfs2_rlist_free(struct gfs2_rgrp_list *rlist)
2634 {
2635 	unsigned int x;
2636 
2637 	kfree(rlist->rl_rgd);
2638 
2639 	if (rlist->rl_ghs) {
2640 		for (x = 0; x < rlist->rl_rgrps; x++)
2641 			gfs2_holder_uninit(&rlist->rl_ghs[x]);
2642 		kfree(rlist->rl_ghs);
2643 		rlist->rl_ghs = NULL;
2644 	}
2645 }
2646 
2647