2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
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.
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
15 #include <linux/gfs2_ondisk.h>
16 #include <linux/prefetch.h>
17 #include <linux/blkdev.h>
18 #include <linux/rbtree.h>
19 #include <linux/random.h>
34 #include "trace_gfs2.h"
36 #define BFITNOENT ((u32)~0)
37 #define NO_BLOCK ((u64)~0)
39 #if BITS_PER_LONG == 32
40 #define LBITMASK (0x55555555UL)
41 #define LBITSKIP55 (0x55555555UL)
42 #define LBITSKIP00 (0x00000000UL)
44 #define LBITMASK (0x5555555555555555UL)
45 #define LBITSKIP55 (0x5555555555555555UL)
46 #define LBITSKIP00 (0x0000000000000000UL)
50 * These routines are used by the resource group routines (rgrp.c)
51 * to keep track of block allocation. Each block is represented by two
52 * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks.
55 * 1 = Used (not metadata)
56 * 2 = Unlinked (still in use) inode
65 static const char valid_change[16] = {
73 static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 *minext,
74 const struct gfs2_inode *ip, bool nowrap,
75 const struct gfs2_alloc_parms *ap);
79 * gfs2_setbit - Set a bit in the bitmaps
80 * @rbm: The position of the bit to set
81 * @do_clone: Also set the clone bitmap, if it exists
82 * @new_state: the new state of the block
86 static inline void gfs2_setbit(const struct gfs2_rbm *rbm, bool do_clone,
87 unsigned char new_state)
89 unsigned char *byte1, *byte2, *end, cur_state;
90 struct gfs2_bitmap *bi = rbm_bi(rbm);
91 unsigned int buflen = bi->bi_len;
92 const unsigned int bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE;
94 byte1 = bi->bi_bh->b_data + bi->bi_offset + (rbm->offset / GFS2_NBBY);
95 end = bi->bi_bh->b_data + bi->bi_offset + buflen;
99 cur_state = (*byte1 >> bit) & GFS2_BIT_MASK;
101 if (unlikely(!valid_change[new_state * 4 + cur_state])) {
102 printk(KERN_WARNING "GFS2: buf_blk = 0x%x old_state=%d, "
103 "new_state=%d\n", rbm->offset, cur_state, new_state);
104 printk(KERN_WARNING "GFS2: rgrp=0x%llx bi_start=0x%x\n",
105 (unsigned long long)rbm->rgd->rd_addr, bi->bi_start);
106 printk(KERN_WARNING "GFS2: bi_offset=0x%x bi_len=0x%x\n",
107 bi->bi_offset, bi->bi_len);
109 gfs2_consist_rgrpd(rbm->rgd);
112 *byte1 ^= (cur_state ^ new_state) << bit;
114 if (do_clone && bi->bi_clone) {
115 byte2 = bi->bi_clone + bi->bi_offset + (rbm->offset / GFS2_NBBY);
116 cur_state = (*byte2 >> bit) & GFS2_BIT_MASK;
117 *byte2 ^= (cur_state ^ new_state) << bit;
122 * gfs2_testbit - test a bit in the bitmaps
123 * @rbm: The bit to test
125 * Returns: The two bit block state of the requested bit
128 static inline u8 gfs2_testbit(const struct gfs2_rbm *rbm)
130 struct gfs2_bitmap *bi = rbm_bi(rbm);
131 const u8 *buffer = bi->bi_bh->b_data + bi->bi_offset;
135 byte = buffer + (rbm->offset / GFS2_NBBY);
136 bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE;
138 return (*byte >> bit) & GFS2_BIT_MASK;
143 * @ptr: Pointer to bitmap data
144 * @mask: Mask to use (normally 0x55555.... but adjusted for search start)
145 * @state: The state we are searching for
147 * We xor the bitmap data with a patter which is the bitwise opposite
148 * of what we are looking for, this gives rise to a pattern of ones
149 * wherever there is a match. Since we have two bits per entry, we
150 * take this pattern, shift it down by one place and then and it with
151 * the original. All the even bit positions (0,2,4, etc) then represent
152 * successful matches, so we mask with 0x55555..... to remove the unwanted
155 * This allows searching of a whole u64 at once (32 blocks) with a
156 * single test (on 64 bit arches).
159 static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state)
162 static const u64 search[] = {
163 [0] = 0xffffffffffffffffULL,
164 [1] = 0xaaaaaaaaaaaaaaaaULL,
165 [2] = 0x5555555555555555ULL,
166 [3] = 0x0000000000000000ULL,
168 tmp = le64_to_cpu(*ptr) ^ search[state];
175 * rs_cmp - multi-block reservation range compare
176 * @blk: absolute file system block number of the new reservation
177 * @len: number of blocks in the new reservation
178 * @rs: existing reservation to compare against
180 * returns: 1 if the block range is beyond the reach of the reservation
181 * -1 if the block range is before the start of the reservation
182 * 0 if the block range overlaps with the reservation
184 static inline int rs_cmp(u64 blk, u32 len, struct gfs2_blkreserv *rs)
186 u64 startblk = gfs2_rbm_to_block(&rs->rs_rbm);
188 if (blk >= startblk + rs->rs_free)
190 if (blk + len - 1 < startblk)
196 * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing
197 * a block in a given allocation state.
198 * @buf: the buffer that holds the bitmaps
199 * @len: the length (in bytes) of the buffer
200 * @goal: start search at this block's bit-pair (within @buffer)
201 * @state: GFS2_BLKST_XXX the state of the block we're looking for.
203 * Scope of @goal and returned block number is only within this bitmap buffer,
204 * not entire rgrp or filesystem. @buffer will be offset from the actual
205 * beginning of a bitmap block buffer, skipping any header structures, but
206 * headers are always a multiple of 64 bits long so that the buffer is
207 * always aligned to a 64 bit boundary.
209 * The size of the buffer is in bytes, but is it assumed that it is
210 * always ok to read a complete multiple of 64 bits at the end
211 * of the block in case the end is no aligned to a natural boundary.
213 * Return: the block number (bitmap buffer scope) that was found
216 static u32 gfs2_bitfit(const u8 *buf, const unsigned int len,
219 u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1);
220 const __le64 *ptr = ((__le64 *)buf) + (goal >> 5);
221 const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64)));
223 u64 mask = 0x5555555555555555ULL;
226 /* Mask off bits we don't care about at the start of the search */
228 tmp = gfs2_bit_search(ptr, mask, state);
230 while(tmp == 0 && ptr < end) {
231 tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state);
234 /* Mask off any bits which are more than len bytes from the start */
235 if (ptr == end && (len & (sizeof(u64) - 1)))
236 tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1))));
237 /* Didn't find anything, so return */
242 bit /= 2; /* two bits per entry in the bitmap */
243 return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit;
247 * gfs2_rbm_from_block - Set the rbm based upon rgd and block number
248 * @rbm: The rbm with rgd already set correctly
249 * @block: The block number (filesystem relative)
251 * This sets the bi and offset members of an rbm based on a
252 * resource group and a filesystem relative block number. The
253 * resource group must be set in the rbm on entry, the bi and
254 * offset members will be set by this function.
256 * Returns: 0 on success, or an error code
259 static int gfs2_rbm_from_block(struct gfs2_rbm *rbm, u64 block)
261 u64 rblock = block - rbm->rgd->rd_data0;
263 if (WARN_ON_ONCE(rblock > UINT_MAX))
265 if (block >= rbm->rgd->rd_data0 + rbm->rgd->rd_data)
269 rbm->offset = (u32)(rblock);
270 /* Check if the block is within the first block */
271 if (rbm->offset < rbm_bi(rbm)->bi_blocks)
274 /* Adjust for the size diff between gfs2_meta_header and gfs2_rgrp */
275 rbm->offset += (sizeof(struct gfs2_rgrp) -
276 sizeof(struct gfs2_meta_header)) * GFS2_NBBY;
277 rbm->bii = rbm->offset / rbm->rgd->rd_sbd->sd_blocks_per_bitmap;
278 rbm->offset -= rbm->bii * rbm->rgd->rd_sbd->sd_blocks_per_bitmap;
283 * gfs2_rbm_incr - increment an rbm structure
284 * @rbm: The rbm with rgd already set correctly
286 * This function takes an existing rbm structure and increments it to the next
287 * viable block offset.
289 * Returns: If incrementing the offset would cause the rbm to go past the
290 * end of the rgrp, true is returned, otherwise false.
294 static bool gfs2_rbm_incr(struct gfs2_rbm *rbm)
296 if (rbm->offset + 1 < rbm_bi(rbm)->bi_blocks) { /* in the same bitmap */
300 if (rbm->bii == rbm->rgd->rd_length - 1) /* at the last bitmap */
309 * gfs2_unaligned_extlen - Look for free blocks which are not byte aligned
310 * @rbm: Position to search (value/result)
311 * @n_unaligned: Number of unaligned blocks to check
312 * @len: Decremented for each block found (terminate on zero)
314 * Returns: true if a non-free block is encountered
317 static bool gfs2_unaligned_extlen(struct gfs2_rbm *rbm, u32 n_unaligned, u32 *len)
322 for (n = 0; n < n_unaligned; n++) {
323 res = gfs2_testbit(rbm);
324 if (res != GFS2_BLKST_FREE)
329 if (gfs2_rbm_incr(rbm))
337 * gfs2_free_extlen - Return extent length of free blocks
338 * @rbm: Starting position
339 * @len: Max length to check
341 * Starting at the block specified by the rbm, see how many free blocks
342 * there are, not reading more than len blocks ahead. This can be done
343 * using memchr_inv when the blocks are byte aligned, but has to be done
344 * on a block by block basis in case of unaligned blocks. Also this
345 * function can cope with bitmap boundaries (although it must stop on
346 * a resource group boundary)
348 * Returns: Number of free blocks in the extent
351 static u32 gfs2_free_extlen(const struct gfs2_rbm *rrbm, u32 len)
353 struct gfs2_rbm rbm = *rrbm;
354 u32 n_unaligned = rbm.offset & 3;
358 u8 *ptr, *start, *end;
360 struct gfs2_bitmap *bi;
363 gfs2_unaligned_extlen(&rbm, 4 - n_unaligned, &len))
366 n_unaligned = len & 3;
367 /* Start is now byte aligned */
370 start = bi->bi_bh->b_data;
372 start = bi->bi_clone;
373 end = start + bi->bi_bh->b_size;
374 start += bi->bi_offset;
375 BUG_ON(rbm.offset & 3);
376 start += (rbm.offset / GFS2_NBBY);
377 bytes = min_t(u32, len / GFS2_NBBY, (end - start));
378 ptr = memchr_inv(start, 0, bytes);
379 chunk_size = ((ptr == NULL) ? bytes : (ptr - start));
380 chunk_size *= GFS2_NBBY;
381 BUG_ON(len < chunk_size);
383 block = gfs2_rbm_to_block(&rbm);
384 if (gfs2_rbm_from_block(&rbm, block + chunk_size)) {
392 n_unaligned = len & 3;
395 /* Deal with any bits left over at the end */
397 gfs2_unaligned_extlen(&rbm, n_unaligned, &len);
403 * gfs2_bitcount - count the number of bits in a certain state
404 * @rgd: the resource group descriptor
405 * @buffer: the buffer that holds the bitmaps
406 * @buflen: the length (in bytes) of the buffer
407 * @state: the state of the block we're looking for
409 * Returns: The number of bits
412 static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer,
413 unsigned int buflen, u8 state)
415 const u8 *byte = buffer;
416 const u8 *end = buffer + buflen;
417 const u8 state1 = state << 2;
418 const u8 state2 = state << 4;
419 const u8 state3 = state << 6;
422 for (; byte < end; byte++) {
423 if (((*byte) & 0x03) == state)
425 if (((*byte) & 0x0C) == state1)
427 if (((*byte) & 0x30) == state2)
429 if (((*byte) & 0xC0) == state3)
437 * gfs2_rgrp_verify - Verify that a resource group is consistent
442 void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd)
444 struct gfs2_sbd *sdp = rgd->rd_sbd;
445 struct gfs2_bitmap *bi = NULL;
446 u32 length = rgd->rd_length;
450 memset(count, 0, 4 * sizeof(u32));
452 /* Count # blocks in each of 4 possible allocation states */
453 for (buf = 0; buf < length; buf++) {
454 bi = rgd->rd_bits + buf;
455 for (x = 0; x < 4; x++)
456 count[x] += gfs2_bitcount(rgd,
462 if (count[0] != rgd->rd_free) {
463 if (gfs2_consist_rgrpd(rgd))
464 fs_err(sdp, "free data mismatch: %u != %u\n",
465 count[0], rgd->rd_free);
469 tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes;
470 if (count[1] != tmp) {
471 if (gfs2_consist_rgrpd(rgd))
472 fs_err(sdp, "used data mismatch: %u != %u\n",
477 if (count[2] + count[3] != rgd->rd_dinodes) {
478 if (gfs2_consist_rgrpd(rgd))
479 fs_err(sdp, "used metadata mismatch: %u != %u\n",
480 count[2] + count[3], rgd->rd_dinodes);
485 static inline int rgrp_contains_block(struct gfs2_rgrpd *rgd, u64 block)
487 u64 first = rgd->rd_data0;
488 u64 last = first + rgd->rd_data;
489 return first <= block && block < last;
493 * gfs2_blk2rgrpd - Find resource group for a given data/meta block number
494 * @sdp: The GFS2 superblock
495 * @blk: The data block number
496 * @exact: True if this needs to be an exact match
498 * Returns: The resource group, or NULL if not found
501 struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk, bool exact)
503 struct rb_node *n, *next;
504 struct gfs2_rgrpd *cur;
506 spin_lock(&sdp->sd_rindex_spin);
507 n = sdp->sd_rindex_tree.rb_node;
509 cur = rb_entry(n, struct gfs2_rgrpd, rd_node);
511 if (blk < cur->rd_addr)
513 else if (blk >= cur->rd_data0 + cur->rd_data)
516 spin_unlock(&sdp->sd_rindex_spin);
518 if (blk < cur->rd_addr)
520 if (blk >= cur->rd_data0 + cur->rd_data)
527 spin_unlock(&sdp->sd_rindex_spin);
533 * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem
534 * @sdp: The GFS2 superblock
536 * Returns: The first rgrp in the filesystem
539 struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp)
541 const struct rb_node *n;
542 struct gfs2_rgrpd *rgd;
544 spin_lock(&sdp->sd_rindex_spin);
545 n = rb_first(&sdp->sd_rindex_tree);
546 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node);
547 spin_unlock(&sdp->sd_rindex_spin);
553 * gfs2_rgrpd_get_next - get the next RG
554 * @rgd: the resource group descriptor
556 * Returns: The next rgrp
559 struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd)
561 struct gfs2_sbd *sdp = rgd->rd_sbd;
562 const struct rb_node *n;
564 spin_lock(&sdp->sd_rindex_spin);
565 n = rb_next(&rgd->rd_node);
567 n = rb_first(&sdp->sd_rindex_tree);
569 if (unlikely(&rgd->rd_node == n)) {
570 spin_unlock(&sdp->sd_rindex_spin);
573 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node);
574 spin_unlock(&sdp->sd_rindex_spin);
578 void gfs2_free_clones(struct gfs2_rgrpd *rgd)
582 for (x = 0; x < rgd->rd_length; x++) {
583 struct gfs2_bitmap *bi = rgd->rd_bits + x;
590 * gfs2_rs_alloc - make sure we have a reservation assigned to the inode
591 * @ip: the inode for this reservation
593 int gfs2_rs_alloc(struct gfs2_inode *ip)
597 down_write(&ip->i_rw_mutex);
601 ip->i_res = kmem_cache_zalloc(gfs2_rsrv_cachep, GFP_NOFS);
607 RB_CLEAR_NODE(&ip->i_res->rs_node);
609 up_write(&ip->i_rw_mutex);
613 static void dump_rs(struct seq_file *seq, const struct gfs2_blkreserv *rs)
615 gfs2_print_dbg(seq, " B: n:%llu s:%llu b:%u f:%u\n",
616 (unsigned long long)rs->rs_inum,
617 (unsigned long long)gfs2_rbm_to_block(&rs->rs_rbm),
618 rs->rs_rbm.offset, rs->rs_free);
622 * __rs_deltree - remove a multi-block reservation from the rgd tree
623 * @rs: The reservation to remove
626 static void __rs_deltree(struct gfs2_blkreserv *rs)
628 struct gfs2_rgrpd *rgd;
630 if (!gfs2_rs_active(rs))
633 rgd = rs->rs_rbm.rgd;
634 trace_gfs2_rs(rs, TRACE_RS_TREEDEL);
635 rb_erase(&rs->rs_node, &rgd->rd_rstree);
636 RB_CLEAR_NODE(&rs->rs_node);
639 struct gfs2_bitmap *bi = rbm_bi(&rs->rs_rbm);
641 /* return reserved blocks to the rgrp */
642 BUG_ON(rs->rs_rbm.rgd->rd_reserved < rs->rs_free);
643 rs->rs_rbm.rgd->rd_reserved -= rs->rs_free;
645 clear_bit(GBF_FULL, &bi->bi_flags);
646 smp_mb__after_clear_bit();
651 * gfs2_rs_deltree - remove a multi-block reservation from the rgd tree
652 * @rs: The reservation to remove
655 void gfs2_rs_deltree(struct gfs2_blkreserv *rs)
657 struct gfs2_rgrpd *rgd;
659 rgd = rs->rs_rbm.rgd;
661 spin_lock(&rgd->rd_rsspin);
663 spin_unlock(&rgd->rd_rsspin);
668 * gfs2_rs_delete - delete a multi-block reservation
669 * @ip: The inode for this reservation
670 * @wcount: The inode's write count, or NULL
673 void gfs2_rs_delete(struct gfs2_inode *ip, atomic_t *wcount)
675 down_write(&ip->i_rw_mutex);
676 if (ip->i_res && ((wcount == NULL) || (atomic_read(wcount) <= 1))) {
677 gfs2_rs_deltree(ip->i_res);
678 BUG_ON(ip->i_res->rs_free);
679 kmem_cache_free(gfs2_rsrv_cachep, ip->i_res);
682 up_write(&ip->i_rw_mutex);
686 * return_all_reservations - return all reserved blocks back to the rgrp.
687 * @rgd: the rgrp that needs its space back
689 * We previously reserved a bunch of blocks for allocation. Now we need to
690 * give them back. This leave the reservation structures in tact, but removes
691 * all of their corresponding "no-fly zones".
693 static void return_all_reservations(struct gfs2_rgrpd *rgd)
696 struct gfs2_blkreserv *rs;
698 spin_lock(&rgd->rd_rsspin);
699 while ((n = rb_first(&rgd->rd_rstree))) {
700 rs = rb_entry(n, struct gfs2_blkreserv, rs_node);
703 spin_unlock(&rgd->rd_rsspin);
706 void gfs2_clear_rgrpd(struct gfs2_sbd *sdp)
709 struct gfs2_rgrpd *rgd;
710 struct gfs2_glock *gl;
712 while ((n = rb_first(&sdp->sd_rindex_tree))) {
713 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node);
716 rb_erase(n, &sdp->sd_rindex_tree);
719 spin_lock(&gl->gl_spin);
720 gl->gl_object = NULL;
721 spin_unlock(&gl->gl_spin);
722 gfs2_glock_add_to_lru(gl);
726 gfs2_free_clones(rgd);
728 return_all_reservations(rgd);
729 kmem_cache_free(gfs2_rgrpd_cachep, rgd);
733 static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd)
735 printk(KERN_INFO " ri_addr = %llu\n", (unsigned long long)rgd->rd_addr);
736 printk(KERN_INFO " ri_length = %u\n", rgd->rd_length);
737 printk(KERN_INFO " ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0);
738 printk(KERN_INFO " ri_data = %u\n", rgd->rd_data);
739 printk(KERN_INFO " ri_bitbytes = %u\n", rgd->rd_bitbytes);
743 * gfs2_compute_bitstructs - Compute the bitmap sizes
744 * @rgd: The resource group descriptor
746 * Calculates bitmap descriptors, one for each block that contains bitmap data
751 static int compute_bitstructs(struct gfs2_rgrpd *rgd)
753 struct gfs2_sbd *sdp = rgd->rd_sbd;
754 struct gfs2_bitmap *bi;
755 u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */
756 u32 bytes_left, bytes;
762 rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS);
766 bytes_left = rgd->rd_bitbytes;
768 for (x = 0; x < length; x++) {
769 bi = rgd->rd_bits + x;
772 /* small rgrp; bitmap stored completely in header block */
775 bi->bi_offset = sizeof(struct gfs2_rgrp);
778 bi->bi_blocks = bytes * GFS2_NBBY;
781 bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp);
782 bi->bi_offset = sizeof(struct gfs2_rgrp);
785 bi->bi_blocks = bytes * GFS2_NBBY;
787 } else if (x + 1 == length) {
789 bi->bi_offset = sizeof(struct gfs2_meta_header);
790 bi->bi_start = rgd->rd_bitbytes - bytes_left;
792 bi->bi_blocks = bytes * GFS2_NBBY;
795 bytes = sdp->sd_sb.sb_bsize -
796 sizeof(struct gfs2_meta_header);
797 bi->bi_offset = sizeof(struct gfs2_meta_header);
798 bi->bi_start = rgd->rd_bitbytes - bytes_left;
800 bi->bi_blocks = bytes * GFS2_NBBY;
807 gfs2_consist_rgrpd(rgd);
810 bi = rgd->rd_bits + (length - 1);
811 if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) {
812 if (gfs2_consist_rgrpd(rgd)) {
813 gfs2_rindex_print(rgd);
814 fs_err(sdp, "start=%u len=%u offset=%u\n",
815 bi->bi_start, bi->bi_len, bi->bi_offset);
824 * gfs2_ri_total - Total up the file system space, according to the rindex.
825 * @sdp: the filesystem
828 u64 gfs2_ri_total(struct gfs2_sbd *sdp)
831 struct inode *inode = sdp->sd_rindex;
832 struct gfs2_inode *ip = GFS2_I(inode);
833 char buf[sizeof(struct gfs2_rindex)];
836 for (rgrps = 0;; rgrps++) {
837 loff_t pos = rgrps * sizeof(struct gfs2_rindex);
839 if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode))
841 error = gfs2_internal_read(ip, buf, &pos,
842 sizeof(struct gfs2_rindex));
843 if (error != sizeof(struct gfs2_rindex))
845 total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data);
850 static int rgd_insert(struct gfs2_rgrpd *rgd)
852 struct gfs2_sbd *sdp = rgd->rd_sbd;
853 struct rb_node **newn = &sdp->sd_rindex_tree.rb_node, *parent = NULL;
855 /* Figure out where to put new node */
857 struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd,
861 if (rgd->rd_addr < cur->rd_addr)
862 newn = &((*newn)->rb_left);
863 else if (rgd->rd_addr > cur->rd_addr)
864 newn = &((*newn)->rb_right);
869 rb_link_node(&rgd->rd_node, parent, newn);
870 rb_insert_color(&rgd->rd_node, &sdp->sd_rindex_tree);
876 * read_rindex_entry - Pull in a new resource index entry from the disk
877 * @ip: Pointer to the rindex inode
879 * Returns: 0 on success, > 0 on EOF, error code otherwise
882 static int read_rindex_entry(struct gfs2_inode *ip)
884 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
885 loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex);
886 struct gfs2_rindex buf;
888 struct gfs2_rgrpd *rgd;
890 if (pos >= i_size_read(&ip->i_inode))
893 error = gfs2_internal_read(ip, (char *)&buf, &pos,
894 sizeof(struct gfs2_rindex));
896 if (error != sizeof(struct gfs2_rindex))
897 return (error == 0) ? 1 : error;
899 rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS);
905 rgd->rd_addr = be64_to_cpu(buf.ri_addr);
906 rgd->rd_length = be32_to_cpu(buf.ri_length);
907 rgd->rd_data0 = be64_to_cpu(buf.ri_data0);
908 rgd->rd_data = be32_to_cpu(buf.ri_data);
909 rgd->rd_bitbytes = be32_to_cpu(buf.ri_bitbytes);
910 spin_lock_init(&rgd->rd_rsspin);
912 error = compute_bitstructs(rgd);
916 error = gfs2_glock_get(sdp, rgd->rd_addr,
917 &gfs2_rgrp_glops, CREATE, &rgd->rd_gl);
921 rgd->rd_gl->gl_object = rgd;
922 rgd->rd_rgl = (struct gfs2_rgrp_lvb *)rgd->rd_gl->gl_lksb.sb_lvbptr;
923 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
924 if (rgd->rd_data > sdp->sd_max_rg_data)
925 sdp->sd_max_rg_data = rgd->rd_data;
926 spin_lock(&sdp->sd_rindex_spin);
927 error = rgd_insert(rgd);
928 spin_unlock(&sdp->sd_rindex_spin);
932 error = 0; /* someone else read in the rgrp; free it and ignore it */
933 gfs2_glock_put(rgd->rd_gl);
937 kmem_cache_free(gfs2_rgrpd_cachep, rgd);
942 * gfs2_ri_update - Pull in a new resource index from the disk
943 * @ip: pointer to the rindex inode
945 * Returns: 0 on successful update, error code otherwise
948 static int gfs2_ri_update(struct gfs2_inode *ip)
950 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
954 error = read_rindex_entry(ip);
955 } while (error == 0);
960 sdp->sd_rindex_uptodate = 1;
965 * gfs2_rindex_update - Update the rindex if required
966 * @sdp: The GFS2 superblock
968 * We grab a lock on the rindex inode to make sure that it doesn't
969 * change whilst we are performing an operation. We keep this lock
970 * for quite long periods of time compared to other locks. This
971 * doesn't matter, since it is shared and it is very, very rarely
972 * accessed in the exclusive mode (i.e. only when expanding the filesystem).
974 * This makes sure that we're using the latest copy of the resource index
975 * special file, which might have been updated if someone expanded the
976 * filesystem (via gfs2_grow utility), which adds new resource groups.
978 * Returns: 0 on succeess, error code otherwise
981 int gfs2_rindex_update(struct gfs2_sbd *sdp)
983 struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex);
984 struct gfs2_glock *gl = ip->i_gl;
985 struct gfs2_holder ri_gh;
987 int unlock_required = 0;
989 /* Read new copy from disk if we don't have the latest */
990 if (!sdp->sd_rindex_uptodate) {
991 if (!gfs2_glock_is_locked_by_me(gl)) {
992 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &ri_gh);
997 if (!sdp->sd_rindex_uptodate)
998 error = gfs2_ri_update(ip);
1000 gfs2_glock_dq_uninit(&ri_gh);
1006 static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf)
1008 const struct gfs2_rgrp *str = buf;
1011 rg_flags = be32_to_cpu(str->rg_flags);
1012 rg_flags &= ~GFS2_RDF_MASK;
1013 rgd->rd_flags &= GFS2_RDF_MASK;
1014 rgd->rd_flags |= rg_flags;
1015 rgd->rd_free = be32_to_cpu(str->rg_free);
1016 rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes);
1017 rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration);
1020 static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf)
1022 struct gfs2_rgrp *str = buf;
1024 str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK);
1025 str->rg_free = cpu_to_be32(rgd->rd_free);
1026 str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes);
1027 str->__pad = cpu_to_be32(0);
1028 str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration);
1029 memset(&str->rg_reserved, 0, sizeof(str->rg_reserved));
1032 static int gfs2_rgrp_lvb_valid(struct gfs2_rgrpd *rgd)
1034 struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl;
1035 struct gfs2_rgrp *str = (struct gfs2_rgrp *)rgd->rd_bits[0].bi_bh->b_data;
1037 if (rgl->rl_flags != str->rg_flags || rgl->rl_free != str->rg_free ||
1038 rgl->rl_dinodes != str->rg_dinodes ||
1039 rgl->rl_igeneration != str->rg_igeneration)
1044 static void gfs2_rgrp_ondisk2lvb(struct gfs2_rgrp_lvb *rgl, const void *buf)
1046 const struct gfs2_rgrp *str = buf;
1048 rgl->rl_magic = cpu_to_be32(GFS2_MAGIC);
1049 rgl->rl_flags = str->rg_flags;
1050 rgl->rl_free = str->rg_free;
1051 rgl->rl_dinodes = str->rg_dinodes;
1052 rgl->rl_igeneration = str->rg_igeneration;
1056 static void update_rgrp_lvb_unlinked(struct gfs2_rgrpd *rgd, u32 change)
1058 struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl;
1059 u32 unlinked = be32_to_cpu(rgl->rl_unlinked) + change;
1060 rgl->rl_unlinked = cpu_to_be32(unlinked);
1063 static u32 count_unlinked(struct gfs2_rgrpd *rgd)
1065 struct gfs2_bitmap *bi;
1066 const u32 length = rgd->rd_length;
1067 const u8 *buffer = NULL;
1068 u32 i, goal, count = 0;
1070 for (i = 0, bi = rgd->rd_bits; i < length; i++, bi++) {
1072 buffer = bi->bi_bh->b_data + bi->bi_offset;
1073 WARN_ON(!buffer_uptodate(bi->bi_bh));
1074 while (goal < bi->bi_len * GFS2_NBBY) {
1075 goal = gfs2_bitfit(buffer, bi->bi_len, goal,
1076 GFS2_BLKST_UNLINKED);
1077 if (goal == BFITNOENT)
1089 * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps
1090 * @rgd: the struct gfs2_rgrpd describing the RG to read in
1092 * Read in all of a Resource Group's header and bitmap blocks.
1093 * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps.
1098 int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd)
1100 struct gfs2_sbd *sdp = rgd->rd_sbd;
1101 struct gfs2_glock *gl = rgd->rd_gl;
1102 unsigned int length = rgd->rd_length;
1103 struct gfs2_bitmap *bi;
1107 if (rgd->rd_bits[0].bi_bh != NULL)
1110 for (x = 0; x < length; x++) {
1111 bi = rgd->rd_bits + x;
1112 error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, &bi->bi_bh);
1117 for (y = length; y--;) {
1118 bi = rgd->rd_bits + y;
1119 error = gfs2_meta_wait(sdp, bi->bi_bh);
1122 if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB :
1123 GFS2_METATYPE_RG)) {
1129 if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) {
1130 for (x = 0; x < length; x++)
1131 clear_bit(GBF_FULL, &rgd->rd_bits[x].bi_flags);
1132 gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data);
1133 rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK);
1134 rgd->rd_free_clone = rgd->rd_free;
1136 if (cpu_to_be32(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) {
1137 rgd->rd_rgl->rl_unlinked = cpu_to_be32(count_unlinked(rgd));
1138 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl,
1139 rgd->rd_bits[0].bi_bh->b_data);
1141 else if (sdp->sd_args.ar_rgrplvb) {
1142 if (!gfs2_rgrp_lvb_valid(rgd)){
1143 gfs2_consist_rgrpd(rgd);
1147 if (rgd->rd_rgl->rl_unlinked == 0)
1148 rgd->rd_flags &= ~GFS2_RDF_CHECK;
1154 bi = rgd->rd_bits + x;
1157 gfs2_assert_warn(sdp, !bi->bi_clone);
1163 int update_rgrp_lvb(struct gfs2_rgrpd *rgd)
1167 if (rgd->rd_flags & GFS2_RDF_UPTODATE)
1170 if (cpu_to_be32(GFS2_MAGIC) != rgd->rd_rgl->rl_magic)
1171 return gfs2_rgrp_bh_get(rgd);
1173 rl_flags = be32_to_cpu(rgd->rd_rgl->rl_flags);
1174 rl_flags &= ~GFS2_RDF_MASK;
1175 rgd->rd_flags &= GFS2_RDF_MASK;
1176 rgd->rd_flags |= (rl_flags | GFS2_RDF_UPTODATE | GFS2_RDF_CHECK);
1177 if (rgd->rd_rgl->rl_unlinked == 0)
1178 rgd->rd_flags &= ~GFS2_RDF_CHECK;
1179 rgd->rd_free = be32_to_cpu(rgd->rd_rgl->rl_free);
1180 rgd->rd_free_clone = rgd->rd_free;
1181 rgd->rd_dinodes = be32_to_cpu(rgd->rd_rgl->rl_dinodes);
1182 rgd->rd_igeneration = be64_to_cpu(rgd->rd_rgl->rl_igeneration);
1186 int gfs2_rgrp_go_lock(struct gfs2_holder *gh)
1188 struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object;
1189 struct gfs2_sbd *sdp = rgd->rd_sbd;
1191 if (gh->gh_flags & GL_SKIP && sdp->sd_args.ar_rgrplvb)
1193 return gfs2_rgrp_bh_get((struct gfs2_rgrpd *)gh->gh_gl->gl_object);
1197 * gfs2_rgrp_go_unlock - Release RG bitmaps read in with gfs2_rgrp_bh_get()
1198 * @gh: The glock holder for the resource group
1202 void gfs2_rgrp_go_unlock(struct gfs2_holder *gh)
1204 struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object;
1205 int x, length = rgd->rd_length;
1207 for (x = 0; x < length; x++) {
1208 struct gfs2_bitmap *bi = rgd->rd_bits + x;
1217 int gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset,
1218 struct buffer_head *bh,
1219 const struct gfs2_bitmap *bi, unsigned minlen, u64 *ptrimmed)
1221 struct super_block *sb = sdp->sd_vfs;
1224 sector_t nr_blks = 0;
1230 for (x = 0; x < bi->bi_len; x++) {
1231 const u8 *clone = bi->bi_clone ? bi->bi_clone : bi->bi_bh->b_data;
1232 clone += bi->bi_offset;
1235 const u8 *orig = bh->b_data + bi->bi_offset + x;
1236 diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1));
1238 diff = ~(*clone | (*clone >> 1));
1243 blk = offset + ((bi->bi_start + x) * GFS2_NBBY);
1247 goto start_new_extent;
1248 if ((start + nr_blks) != blk) {
1249 if (nr_blks >= minlen) {
1250 rv = sb_issue_discard(sb,
1267 if (nr_blks >= minlen) {
1268 rv = sb_issue_discard(sb, start, nr_blks, GFP_NOFS, 0);
1274 *ptrimmed = trimmed;
1278 if (sdp->sd_args.ar_discard)
1279 fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv);
1280 sdp->sd_args.ar_discard = 0;
1285 * gfs2_fitrim - Generate discard requests for unused bits of the filesystem
1286 * @filp: Any file on the filesystem
1287 * @argp: Pointer to the arguments (also used to pass result)
1289 * Returns: 0 on success, otherwise error code
1292 int gfs2_fitrim(struct file *filp, void __user *argp)
1294 struct inode *inode = file_inode(filp);
1295 struct gfs2_sbd *sdp = GFS2_SB(inode);
1296 struct request_queue *q = bdev_get_queue(sdp->sd_vfs->s_bdev);
1297 struct buffer_head *bh;
1298 struct gfs2_rgrpd *rgd;
1299 struct gfs2_rgrpd *rgd_end;
1300 struct gfs2_holder gh;
1301 struct fstrim_range r;
1305 u64 start, end, minlen;
1307 unsigned bs_shift = sdp->sd_sb.sb_bsize_shift;
1309 if (!capable(CAP_SYS_ADMIN))
1312 if (!blk_queue_discard(q))
1315 if (copy_from_user(&r, argp, sizeof(r)))
1318 ret = gfs2_rindex_update(sdp);
1322 start = r.start >> bs_shift;
1323 end = start + (r.len >> bs_shift);
1324 minlen = max_t(u64, r.minlen,
1325 q->limits.discard_granularity) >> bs_shift;
1327 if (end <= start || minlen > sdp->sd_max_rg_data)
1330 rgd = gfs2_blk2rgrpd(sdp, start, 0);
1331 rgd_end = gfs2_blk2rgrpd(sdp, end, 0);
1333 if ((gfs2_rgrpd_get_first(sdp) == gfs2_rgrpd_get_next(rgd_end))
1334 && (start > rgd_end->rd_data0 + rgd_end->rd_data))
1335 return -EINVAL; /* start is beyond the end of the fs */
1339 ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1343 if (!(rgd->rd_flags & GFS2_RGF_TRIMMED)) {
1344 /* Trim each bitmap in the rgrp */
1345 for (x = 0; x < rgd->rd_length; x++) {
1346 struct gfs2_bitmap *bi = rgd->rd_bits + x;
1347 ret = gfs2_rgrp_send_discards(sdp,
1348 rgd->rd_data0, NULL, bi, minlen,
1351 gfs2_glock_dq_uninit(&gh);
1357 /* Mark rgrp as having been trimmed */
1358 ret = gfs2_trans_begin(sdp, RES_RG_HDR, 0);
1360 bh = rgd->rd_bits[0].bi_bh;
1361 rgd->rd_flags |= GFS2_RGF_TRIMMED;
1362 gfs2_trans_add_meta(rgd->rd_gl, bh);
1363 gfs2_rgrp_out(rgd, bh->b_data);
1364 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, bh->b_data);
1365 gfs2_trans_end(sdp);
1368 gfs2_glock_dq_uninit(&gh);
1373 rgd = gfs2_rgrpd_get_next(rgd);
1377 r.len = trimmed << bs_shift;
1378 if (copy_to_user(argp, &r, sizeof(r)))
1385 * rs_insert - insert a new multi-block reservation into the rgrp's rb_tree
1386 * @ip: the inode structure
1389 static void rs_insert(struct gfs2_inode *ip)
1391 struct rb_node **newn, *parent = NULL;
1393 struct gfs2_blkreserv *rs = ip->i_res;
1394 struct gfs2_rgrpd *rgd = rs->rs_rbm.rgd;
1395 u64 fsblock = gfs2_rbm_to_block(&rs->rs_rbm);
1397 BUG_ON(gfs2_rs_active(rs));
1399 spin_lock(&rgd->rd_rsspin);
1400 newn = &rgd->rd_rstree.rb_node;
1402 struct gfs2_blkreserv *cur =
1403 rb_entry(*newn, struct gfs2_blkreserv, rs_node);
1406 rc = rs_cmp(fsblock, rs->rs_free, cur);
1408 newn = &((*newn)->rb_right);
1410 newn = &((*newn)->rb_left);
1412 spin_unlock(&rgd->rd_rsspin);
1418 rb_link_node(&rs->rs_node, parent, newn);
1419 rb_insert_color(&rs->rs_node, &rgd->rd_rstree);
1421 /* Do our rgrp accounting for the reservation */
1422 rgd->rd_reserved += rs->rs_free; /* blocks reserved */
1423 spin_unlock(&rgd->rd_rsspin);
1424 trace_gfs2_rs(rs, TRACE_RS_INSERT);
1428 * rg_mblk_search - find a group of multiple free blocks to form a reservation
1429 * @rgd: the resource group descriptor
1430 * @ip: pointer to the inode for which we're reserving blocks
1431 * @ap: the allocation parameters
1435 static void rg_mblk_search(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip,
1436 const struct gfs2_alloc_parms *ap)
1438 struct gfs2_rbm rbm = { .rgd = rgd, };
1440 struct gfs2_blkreserv *rs = ip->i_res;
1442 u32 free_blocks = rgd->rd_free_clone - rgd->rd_reserved;
1444 struct inode *inode = &ip->i_inode;
1446 if (S_ISDIR(inode->i_mode))
1449 extlen = max_t(u32, atomic_read(&rs->rs_sizehint), ap->target);
1450 extlen = clamp(extlen, RGRP_RSRV_MINBLKS, free_blocks);
1452 if ((rgd->rd_free_clone < rgd->rd_reserved) || (free_blocks < extlen))
1455 /* Find bitmap block that contains bits for goal block */
1456 if (rgrp_contains_block(rgd, ip->i_goal))
1459 goal = rgd->rd_last_alloc + rgd->rd_data0;
1461 if (WARN_ON(gfs2_rbm_from_block(&rbm, goal)))
1464 ret = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, &extlen, ip, true, ap);
1467 rs->rs_free = extlen;
1468 rs->rs_inum = ip->i_no_addr;
1471 if (goal == rgd->rd_last_alloc + rgd->rd_data0)
1472 rgd->rd_last_alloc = 0;
1477 * gfs2_next_unreserved_block - Return next block that is not reserved
1478 * @rgd: The resource group
1479 * @block: The starting block
1480 * @length: The required length
1481 * @ip: Ignore any reservations for this inode
1483 * If the block does not appear in any reservation, then return the
1484 * block number unchanged. If it does appear in the reservation, then
1485 * keep looking through the tree of reservations in order to find the
1486 * first block number which is not reserved.
1489 static u64 gfs2_next_unreserved_block(struct gfs2_rgrpd *rgd, u64 block,
1491 const struct gfs2_inode *ip)
1493 struct gfs2_blkreserv *rs;
1497 spin_lock(&rgd->rd_rsspin);
1498 n = rgd->rd_rstree.rb_node;
1500 rs = rb_entry(n, struct gfs2_blkreserv, rs_node);
1501 rc = rs_cmp(block, length, rs);
1511 while ((rs_cmp(block, length, rs) == 0) && (ip->i_res != rs)) {
1512 block = gfs2_rbm_to_block(&rs->rs_rbm) + rs->rs_free;
1516 rs = rb_entry(n, struct gfs2_blkreserv, rs_node);
1520 spin_unlock(&rgd->rd_rsspin);
1525 * gfs2_reservation_check_and_update - Check for reservations during block alloc
1526 * @rbm: The current position in the resource group
1527 * @ip: The inode for which we are searching for blocks
1528 * @minext: The minimum extent length
1529 * @maxext: A pointer to the maximum extent structure
1531 * This checks the current position in the rgrp to see whether there is
1532 * a reservation covering this block. If not then this function is a
1533 * no-op. If there is, then the position is moved to the end of the
1534 * contiguous reservation(s) so that we are pointing at the first
1535 * non-reserved block.
1537 * Returns: 0 if no reservation, 1 if @rbm has changed, otherwise an error
1540 static int gfs2_reservation_check_and_update(struct gfs2_rbm *rbm,
1541 const struct gfs2_inode *ip,
1543 struct gfs2_extent *maxext)
1545 u64 block = gfs2_rbm_to_block(rbm);
1551 * If we have a minimum extent length, then skip over any extent
1552 * which is less than the min extent length in size.
1555 extlen = gfs2_free_extlen(rbm, minext);
1556 if (extlen <= maxext->len)
1561 * Check the extent which has been found against the reservations
1562 * and skip if parts of it are already reserved
1564 nblock = gfs2_next_unreserved_block(rbm->rgd, block, extlen, ip);
1565 if (nblock == block) {
1566 if (!minext || extlen >= minext)
1569 if (extlen > maxext->len) {
1570 maxext->len = extlen;
1574 nblock = block + extlen;
1576 ret = gfs2_rbm_from_block(rbm, nblock);
1583 * gfs2_rbm_find - Look for blocks of a particular state
1584 * @rbm: Value/result starting position and final position
1585 * @state: The state which we want to find
1586 * @minext: Pointer to the requested extent length (NULL for a single block)
1587 * This is updated to be the actual reservation size.
1588 * @ip: If set, check for reservations
1589 * @nowrap: Stop looking at the end of the rgrp, rather than wrapping
1590 * around until we've reached the starting point.
1591 * @ap: the allocation parameters
1594 * - If looking for free blocks, we set GBF_FULL on each bitmap which
1595 * has no free blocks in it.
1597 * Returns: 0 on success, -ENOSPC if there is no block of the requested state
1600 static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 *minext,
1601 const struct gfs2_inode *ip, bool nowrap,
1602 const struct gfs2_alloc_parms *ap)
1604 struct buffer_head *bh;
1610 int iters = rbm->rgd->rd_length;
1612 struct gfs2_bitmap *bi;
1613 struct gfs2_extent maxext = { .rbm.rgd = rbm->rgd, };
1615 /* If we are not starting at the beginning of a bitmap, then we
1616 * need to add one to the bitmap count to ensure that we search
1617 * the starting bitmap twice.
1619 if (rbm->offset != 0)
1624 if (test_bit(GBF_FULL, &bi->bi_flags) &&
1625 (state == GFS2_BLKST_FREE))
1629 buffer = bh->b_data + bi->bi_offset;
1630 WARN_ON(!buffer_uptodate(bh));
1631 if (state != GFS2_BLKST_UNLINKED && bi->bi_clone)
1632 buffer = bi->bi_clone + bi->bi_offset;
1633 initial_offset = rbm->offset;
1634 offset = gfs2_bitfit(buffer, bi->bi_len, rbm->offset, state);
1635 if (offset == BFITNOENT)
1637 rbm->offset = offset;
1641 initial_bii = rbm->bii;
1642 ret = gfs2_reservation_check_and_update(rbm, ip,
1643 minext ? *minext : 0,
1648 n += (rbm->bii - initial_bii);
1651 if (ret == -E2BIG) {
1654 n += (rbm->bii - initial_bii);
1655 goto res_covered_end_of_rgrp;
1659 bitmap_full: /* Mark bitmap as full and fall through */
1660 if ((state == GFS2_BLKST_FREE) && initial_offset == 0) {
1661 struct gfs2_bitmap *bi = rbm_bi(rbm);
1662 set_bit(GBF_FULL, &bi->bi_flags);
1665 next_bitmap: /* Find next bitmap in the rgrp */
1668 if (rbm->bii == rbm->rgd->rd_length)
1670 res_covered_end_of_rgrp:
1671 if ((rbm->bii == 0) && nowrap)
1679 if (minext == NULL || state != GFS2_BLKST_FREE)
1682 /* If the maximum extent we found is big enough to fulfill the
1683 minimum requirements, use it anyway. */
1686 *minext = maxext.len;
1694 * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes
1696 * @last_unlinked: block address of the last dinode we unlinked
1697 * @skip: block address we should explicitly not unlink
1699 * Returns: 0 if no error
1700 * The inode, if one has been found, in inode.
1703 static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip)
1706 struct gfs2_sbd *sdp = rgd->rd_sbd;
1707 struct gfs2_glock *gl;
1708 struct gfs2_inode *ip;
1711 struct gfs2_rbm rbm = { .rgd = rgd, .bii = 0, .offset = 0 };
1714 down_write(&sdp->sd_log_flush_lock);
1715 error = gfs2_rbm_find(&rbm, GFS2_BLKST_UNLINKED, NULL, NULL,
1717 up_write(&sdp->sd_log_flush_lock);
1718 if (error == -ENOSPC)
1720 if (WARN_ON_ONCE(error))
1723 block = gfs2_rbm_to_block(&rbm);
1724 if (gfs2_rbm_from_block(&rbm, block + 1))
1726 if (*last_unlinked != NO_BLOCK && block <= *last_unlinked)
1730 *last_unlinked = block;
1732 error = gfs2_glock_get(sdp, block, &gfs2_inode_glops, CREATE, &gl);
1736 /* If the inode is already in cache, we can ignore it here
1737 * because the existing inode disposal code will deal with
1738 * it when all refs have gone away. Accessing gl_object like
1739 * this is not safe in general. Here it is ok because we do
1740 * not dereference the pointer, and we only need an approx
1741 * answer to whether it is NULL or not.
1745 if (ip || queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
1750 /* Limit reclaim to sensible number of tasks */
1751 if (found > NR_CPUS)
1755 rgd->rd_flags &= ~GFS2_RDF_CHECK;
1760 * gfs2_rgrp_congested - Use stats to figure out whether an rgrp is congested
1761 * @rgd: The rgrp in question
1762 * @loops: An indication of how picky we can be (0=very, 1=less so)
1764 * This function uses the recently added glock statistics in order to
1765 * figure out whether a parciular resource group is suffering from
1766 * contention from multiple nodes. This is done purely on the basis
1767 * of timings, since this is the only data we have to work with and
1768 * our aim here is to reject a resource group which is highly contended
1769 * but (very important) not to do this too often in order to ensure that
1770 * we do not land up introducing fragmentation by changing resource
1771 * groups when not actually required.
1773 * The calculation is fairly simple, we want to know whether the SRTTB
1774 * (i.e. smoothed round trip time for blocking operations) to acquire
1775 * the lock for this rgrp's glock is significantly greater than the
1776 * time taken for resource groups on average. We introduce a margin in
1777 * the form of the variable @var which is computed as the sum of the two
1778 * respective variences, and multiplied by a factor depending on @loops
1779 * and whether we have a lot of data to base the decision on. This is
1780 * then tested against the square difference of the means in order to
1781 * decide whether the result is statistically significant or not.
1783 * Returns: A boolean verdict on the congestion status
1786 static bool gfs2_rgrp_congested(const struct gfs2_rgrpd *rgd, int loops)
1788 const struct gfs2_glock *gl = rgd->rd_gl;
1789 const struct gfs2_sbd *sdp = gl->gl_sbd;
1790 struct gfs2_lkstats *st;
1791 s64 r_dcount, l_dcount;
1792 s64 r_srttb, l_srttb;
1798 st = &this_cpu_ptr(sdp->sd_lkstats)->lkstats[LM_TYPE_RGRP];
1799 r_srttb = st->stats[GFS2_LKS_SRTTB];
1800 r_dcount = st->stats[GFS2_LKS_DCOUNT];
1801 var = st->stats[GFS2_LKS_SRTTVARB] +
1802 gl->gl_stats.stats[GFS2_LKS_SRTTVARB];
1805 l_srttb = gl->gl_stats.stats[GFS2_LKS_SRTTB];
1806 l_dcount = gl->gl_stats.stats[GFS2_LKS_DCOUNT];
1808 if ((l_dcount < 1) || (r_dcount < 1) || (r_srttb == 0))
1811 srttb_diff = r_srttb - l_srttb;
1812 sqr_diff = srttb_diff * srttb_diff;
1815 if (l_dcount < 8 || r_dcount < 8)
1820 return ((srttb_diff < 0) && (sqr_diff > var));
1824 * gfs2_rgrp_used_recently
1825 * @rs: The block reservation with the rgrp to test
1826 * @msecs: The time limit in milliseconds
1828 * Returns: True if the rgrp glock has been used within the time limit
1830 static bool gfs2_rgrp_used_recently(const struct gfs2_blkreserv *rs,
1835 tdiff = ktime_to_ns(ktime_sub(ktime_get_real(),
1836 rs->rs_rbm.rgd->rd_gl->gl_dstamp));
1838 return tdiff > (msecs * 1000 * 1000);
1841 static u32 gfs2_orlov_skip(const struct gfs2_inode *ip)
1843 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1846 get_random_bytes(&skip, sizeof(skip));
1847 return skip % sdp->sd_rgrps;
1850 static bool gfs2_select_rgrp(struct gfs2_rgrpd **pos, const struct gfs2_rgrpd *begin)
1852 struct gfs2_rgrpd *rgd = *pos;
1853 struct gfs2_sbd *sdp = rgd->rd_sbd;
1855 rgd = gfs2_rgrpd_get_next(rgd);
1857 rgd = gfs2_rgrpd_get_first(sdp);
1859 if (rgd != begin) /* If we didn't wrap */
1865 * gfs2_inplace_reserve - Reserve space in the filesystem
1866 * @ip: the inode to reserve space for
1867 * @ap: the allocation parameters
1872 int gfs2_inplace_reserve(struct gfs2_inode *ip, const struct gfs2_alloc_parms *ap)
1874 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1875 struct gfs2_rgrpd *begin = NULL;
1876 struct gfs2_blkreserv *rs = ip->i_res;
1877 int error = 0, rg_locked, flags = 0;
1878 u64 last_unlinked = NO_BLOCK;
1882 if (sdp->sd_args.ar_rgrplvb)
1884 if (gfs2_assert_warn(sdp, ap->target))
1886 if (gfs2_rs_active(rs)) {
1887 begin = rs->rs_rbm.rgd;
1888 } else if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, ip->i_goal)) {
1889 rs->rs_rbm.rgd = begin = ip->i_rgd;
1891 rs->rs_rbm.rgd = begin = gfs2_blk2rgrpd(sdp, ip->i_goal, 1);
1893 if (S_ISDIR(ip->i_inode.i_mode) && (ap->aflags & GFS2_AF_ORLOV))
1894 skip = gfs2_orlov_skip(ip);
1895 if (rs->rs_rbm.rgd == NULL)
1901 if (!gfs2_glock_is_locked_by_me(rs->rs_rbm.rgd->rd_gl)) {
1905 if (!gfs2_rs_active(rs) && (loops < 2) &&
1906 gfs2_rgrp_used_recently(rs, 1000) &&
1907 gfs2_rgrp_congested(rs->rs_rbm.rgd, loops))
1909 error = gfs2_glock_nq_init(rs->rs_rbm.rgd->rd_gl,
1910 LM_ST_EXCLUSIVE, flags,
1912 if (unlikely(error))
1914 if (!gfs2_rs_active(rs) && (loops < 2) &&
1915 gfs2_rgrp_congested(rs->rs_rbm.rgd, loops))
1917 if (sdp->sd_args.ar_rgrplvb) {
1918 error = update_rgrp_lvb(rs->rs_rbm.rgd);
1919 if (unlikely(error)) {
1920 gfs2_glock_dq_uninit(&rs->rs_rgd_gh);
1926 /* Skip unuseable resource groups */
1927 if (rs->rs_rbm.rgd->rd_flags & (GFS2_RGF_NOALLOC | GFS2_RDF_ERROR))
1930 if (sdp->sd_args.ar_rgrplvb)
1931 gfs2_rgrp_bh_get(rs->rs_rbm.rgd);
1933 /* Get a reservation if we don't already have one */
1934 if (!gfs2_rs_active(rs))
1935 rg_mblk_search(rs->rs_rbm.rgd, ip, ap);
1937 /* Skip rgrps when we can't get a reservation on first pass */
1938 if (!gfs2_rs_active(rs) && (loops < 1))
1941 /* If rgrp has enough free space, use it */
1942 if (rs->rs_rbm.rgd->rd_free_clone >= ap->target) {
1943 ip->i_rgd = rs->rs_rbm.rgd;
1947 /* Drop reservation, if we couldn't use reserved rgrp */
1948 if (gfs2_rs_active(rs))
1949 gfs2_rs_deltree(rs);
1951 /* Check for unlinked inodes which can be reclaimed */
1952 if (rs->rs_rbm.rgd->rd_flags & GFS2_RDF_CHECK)
1953 try_rgrp_unlink(rs->rs_rbm.rgd, &last_unlinked,
1956 /* Unlock rgrp if required */
1958 gfs2_glock_dq_uninit(&rs->rs_rgd_gh);
1960 /* Find the next rgrp, and continue looking */
1961 if (gfs2_select_rgrp(&rs->rs_rbm.rgd, begin))
1966 /* If we've scanned all the rgrps, but found no free blocks
1967 * then this checks for some less likely conditions before
1971 /* Check that fs hasn't grown if writing to rindex */
1972 if (ip == GFS2_I(sdp->sd_rindex) && !sdp->sd_rindex_uptodate) {
1973 error = gfs2_ri_update(ip);
1977 /* Flushing the log may release space */
1979 gfs2_log_flush(sdp, NULL);
1986 * gfs2_inplace_release - release an inplace reservation
1987 * @ip: the inode the reservation was taken out on
1989 * Release a reservation made by gfs2_inplace_reserve().
1992 void gfs2_inplace_release(struct gfs2_inode *ip)
1994 struct gfs2_blkreserv *rs = ip->i_res;
1996 if (rs->rs_rgd_gh.gh_gl)
1997 gfs2_glock_dq_uninit(&rs->rs_rgd_gh);
2001 * gfs2_get_block_type - Check a block in a RG is of given type
2002 * @rgd: the resource group holding the block
2003 * @block: the block number
2005 * Returns: The block type (GFS2_BLKST_*)
2008 static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block)
2010 struct gfs2_rbm rbm = { .rgd = rgd, };
2013 ret = gfs2_rbm_from_block(&rbm, block);
2014 WARN_ON_ONCE(ret != 0);
2016 return gfs2_testbit(&rbm);
2021 * gfs2_alloc_extent - allocate an extent from a given bitmap
2022 * @rbm: the resource group information
2023 * @dinode: TRUE if the first block we allocate is for a dinode
2024 * @n: The extent length (value/result)
2026 * Add the bitmap buffer to the transaction.
2027 * Set the found bits to @new_state to change block's allocation state.
2029 static void gfs2_alloc_extent(const struct gfs2_rbm *rbm, bool dinode,
2032 struct gfs2_rbm pos = { .rgd = rbm->rgd, };
2033 const unsigned int elen = *n;
2038 block = gfs2_rbm_to_block(rbm);
2039 gfs2_trans_add_meta(rbm->rgd->rd_gl, rbm_bi(rbm)->bi_bh);
2040 gfs2_setbit(rbm, true, dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED);
2043 ret = gfs2_rbm_from_block(&pos, block);
2044 if (ret || gfs2_testbit(&pos) != GFS2_BLKST_FREE)
2046 gfs2_trans_add_meta(pos.rgd->rd_gl, rbm_bi(&pos)->bi_bh);
2047 gfs2_setbit(&pos, true, GFS2_BLKST_USED);
2054 * rgblk_free - Change alloc state of given block(s)
2055 * @sdp: the filesystem
2056 * @bstart: the start of a run of blocks to free
2057 * @blen: the length of the block run (all must lie within ONE RG!)
2058 * @new_state: GFS2_BLKST_XXX the after-allocation block state
2060 * Returns: Resource group containing the block(s)
2063 static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart,
2064 u32 blen, unsigned char new_state)
2066 struct gfs2_rbm rbm;
2067 struct gfs2_bitmap *bi;
2069 rbm.rgd = gfs2_blk2rgrpd(sdp, bstart, 1);
2071 if (gfs2_consist(sdp))
2072 fs_err(sdp, "block = %llu\n", (unsigned long long)bstart);
2077 gfs2_rbm_from_block(&rbm, bstart);
2080 if (!bi->bi_clone) {
2081 bi->bi_clone = kmalloc(bi->bi_bh->b_size,
2082 GFP_NOFS | __GFP_NOFAIL);
2083 memcpy(bi->bi_clone + bi->bi_offset,
2084 bi->bi_bh->b_data + bi->bi_offset, bi->bi_len);
2086 gfs2_trans_add_meta(rbm.rgd->rd_gl, bi->bi_bh);
2087 gfs2_setbit(&rbm, false, new_state);
2094 * gfs2_rgrp_dump - print out an rgrp
2095 * @seq: The iterator
2096 * @gl: The glock in question
2100 int gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl)
2102 struct gfs2_rgrpd *rgd = gl->gl_object;
2103 struct gfs2_blkreserv *trs;
2104 const struct rb_node *n;
2108 gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u r:%u\n",
2109 (unsigned long long)rgd->rd_addr, rgd->rd_flags,
2110 rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes,
2112 spin_lock(&rgd->rd_rsspin);
2113 for (n = rb_first(&rgd->rd_rstree); n; n = rb_next(&trs->rs_node)) {
2114 trs = rb_entry(n, struct gfs2_blkreserv, rs_node);
2117 spin_unlock(&rgd->rd_rsspin);
2121 static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd)
2123 struct gfs2_sbd *sdp = rgd->rd_sbd;
2124 fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n",
2125 (unsigned long long)rgd->rd_addr);
2126 fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n");
2127 gfs2_rgrp_dump(NULL, rgd->rd_gl);
2128 rgd->rd_flags |= GFS2_RDF_ERROR;
2132 * gfs2_adjust_reservation - Adjust (or remove) a reservation after allocation
2133 * @ip: The inode we have just allocated blocks for
2134 * @rbm: The start of the allocated blocks
2135 * @len: The extent length
2137 * Adjusts a reservation after an allocation has taken place. If the
2138 * reservation does not match the allocation, or if it is now empty
2139 * then it is removed.
2142 static void gfs2_adjust_reservation(struct gfs2_inode *ip,
2143 const struct gfs2_rbm *rbm, unsigned len)
2145 struct gfs2_blkreserv *rs = ip->i_res;
2146 struct gfs2_rgrpd *rgd = rbm->rgd;
2151 spin_lock(&rgd->rd_rsspin);
2152 if (gfs2_rs_active(rs)) {
2153 if (gfs2_rbm_eq(&rs->rs_rbm, rbm)) {
2154 block = gfs2_rbm_to_block(rbm);
2155 ret = gfs2_rbm_from_block(&rs->rs_rbm, block + len);
2156 rlen = min(rs->rs_free, len);
2157 rs->rs_free -= rlen;
2158 rgd->rd_reserved -= rlen;
2159 trace_gfs2_rs(rs, TRACE_RS_CLAIM);
2160 if (rs->rs_free && !ret)
2166 spin_unlock(&rgd->rd_rsspin);
2170 * gfs2_set_alloc_start - Set starting point for block allocation
2171 * @rbm: The rbm which will be set to the required location
2172 * @ip: The gfs2 inode
2173 * @dinode: Flag to say if allocation includes a new inode
2175 * This sets the starting point from the reservation if one is active
2176 * otherwise it falls back to guessing a start point based on the
2177 * inode's goal block or the last allocation point in the rgrp.
2180 static void gfs2_set_alloc_start(struct gfs2_rbm *rbm,
2181 const struct gfs2_inode *ip, bool dinode)
2185 if (gfs2_rs_active(ip->i_res)) {
2186 *rbm = ip->i_res->rs_rbm;
2190 if (!dinode && rgrp_contains_block(rbm->rgd, ip->i_goal))
2193 goal = rbm->rgd->rd_last_alloc + rbm->rgd->rd_data0;
2195 gfs2_rbm_from_block(rbm, goal);
2199 * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode
2200 * @ip: the inode to allocate the block for
2201 * @bn: Used to return the starting block number
2202 * @nblocks: requested number of blocks/extent length (value/result)
2203 * @dinode: 1 if we're allocating a dinode block, else 0
2204 * @generation: the generation number of the inode
2206 * Returns: 0 or error
2209 int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *nblocks,
2210 bool dinode, u64 *generation)
2212 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2213 struct buffer_head *dibh;
2214 struct gfs2_rbm rbm = { .rgd = ip->i_rgd, };
2216 u64 block; /* block, within the file system scope */
2219 gfs2_set_alloc_start(&rbm, ip, dinode);
2220 error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, NULL, ip, false, NULL);
2222 if (error == -ENOSPC) {
2223 gfs2_set_alloc_start(&rbm, ip, dinode);
2224 error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, NULL, NULL, false,
2228 /* Since all blocks are reserved in advance, this shouldn't happen */
2230 fs_warn(sdp, "inum=%llu error=%d, nblocks=%u, full=%d\n",
2231 (unsigned long long)ip->i_no_addr, error, *nblocks,
2232 test_bit(GBF_FULL, &rbm.rgd->rd_bits->bi_flags));
2236 gfs2_alloc_extent(&rbm, dinode, nblocks);
2237 block = gfs2_rbm_to_block(&rbm);
2238 rbm.rgd->rd_last_alloc = block - rbm.rgd->rd_data0;
2239 if (gfs2_rs_active(ip->i_res))
2240 gfs2_adjust_reservation(ip, &rbm, *nblocks);
2246 ip->i_goal = block + ndata - 1;
2247 error = gfs2_meta_inode_buffer(ip, &dibh);
2249 struct gfs2_dinode *di =
2250 (struct gfs2_dinode *)dibh->b_data;
2251 gfs2_trans_add_meta(ip->i_gl, dibh);
2252 di->di_goal_meta = di->di_goal_data =
2253 cpu_to_be64(ip->i_goal);
2257 if (rbm.rgd->rd_free < *nblocks) {
2258 printk(KERN_WARNING "nblocks=%u\n", *nblocks);
2262 rbm.rgd->rd_free -= *nblocks;
2264 rbm.rgd->rd_dinodes++;
2265 *generation = rbm.rgd->rd_igeneration++;
2266 if (*generation == 0)
2267 *generation = rbm.rgd->rd_igeneration++;
2270 gfs2_trans_add_meta(rbm.rgd->rd_gl, rbm.rgd->rd_bits[0].bi_bh);
2271 gfs2_rgrp_out(rbm.rgd, rbm.rgd->rd_bits[0].bi_bh->b_data);
2272 gfs2_rgrp_ondisk2lvb(rbm.rgd->rd_rgl, rbm.rgd->rd_bits[0].bi_bh->b_data);
2274 gfs2_statfs_change(sdp, 0, -(s64)*nblocks, dinode ? 1 : 0);
2276 gfs2_trans_add_unrevoke(sdp, block, 1);
2278 gfs2_quota_change(ip, *nblocks, ip->i_inode.i_uid, ip->i_inode.i_gid);
2280 rbm.rgd->rd_free_clone -= *nblocks;
2281 trace_gfs2_block_alloc(ip, rbm.rgd, block, *nblocks,
2282 dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED);
2287 gfs2_rgrp_error(rbm.rgd);
2292 * __gfs2_free_blocks - free a contiguous run of block(s)
2293 * @ip: the inode these blocks are being freed from
2294 * @bstart: first block of a run of contiguous blocks
2295 * @blen: the length of the block run
2296 * @meta: 1 if the blocks represent metadata
2300 void __gfs2_free_blocks(struct gfs2_inode *ip, u64 bstart, u32 blen, int meta)
2302 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2303 struct gfs2_rgrpd *rgd;
2305 rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE);
2308 trace_gfs2_block_alloc(ip, rgd, bstart, blen, GFS2_BLKST_FREE);
2309 rgd->rd_free += blen;
2310 rgd->rd_flags &= ~GFS2_RGF_TRIMMED;
2311 gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh);
2312 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
2313 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data);
2315 /* Directories keep their data in the metadata address space */
2316 if (meta || ip->i_depth)
2317 gfs2_meta_wipe(ip, bstart, blen);
2321 * gfs2_free_meta - free a contiguous run of data block(s)
2322 * @ip: the inode these blocks are being freed from
2323 * @bstart: first block of a run of contiguous blocks
2324 * @blen: the length of the block run
2328 void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen)
2330 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2332 __gfs2_free_blocks(ip, bstart, blen, 1);
2333 gfs2_statfs_change(sdp, 0, +blen, 0);
2334 gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid);
2337 void gfs2_unlink_di(struct inode *inode)
2339 struct gfs2_inode *ip = GFS2_I(inode);
2340 struct gfs2_sbd *sdp = GFS2_SB(inode);
2341 struct gfs2_rgrpd *rgd;
2342 u64 blkno = ip->i_no_addr;
2344 rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED);
2347 trace_gfs2_block_alloc(ip, rgd, blkno, 1, GFS2_BLKST_UNLINKED);
2348 gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh);
2349 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
2350 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data);
2351 update_rgrp_lvb_unlinked(rgd, 1);
2354 static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno)
2356 struct gfs2_sbd *sdp = rgd->rd_sbd;
2357 struct gfs2_rgrpd *tmp_rgd;
2359 tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE);
2362 gfs2_assert_withdraw(sdp, rgd == tmp_rgd);
2364 if (!rgd->rd_dinodes)
2365 gfs2_consist_rgrpd(rgd);
2369 gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh);
2370 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
2371 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data);
2372 update_rgrp_lvb_unlinked(rgd, -1);
2374 gfs2_statfs_change(sdp, 0, +1, -1);
2378 void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip)
2380 gfs2_free_uninit_di(rgd, ip->i_no_addr);
2381 trace_gfs2_block_alloc(ip, rgd, ip->i_no_addr, 1, GFS2_BLKST_FREE);
2382 gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid);
2383 gfs2_meta_wipe(ip, ip->i_no_addr, 1);
2387 * gfs2_check_blk_type - Check the type of a block
2388 * @sdp: The superblock
2389 * @no_addr: The block number to check
2390 * @type: The block type we are looking for
2392 * Returns: 0 if the block type matches the expected type
2393 * -ESTALE if it doesn't match
2394 * or -ve errno if something went wrong while checking
2397 int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type)
2399 struct gfs2_rgrpd *rgd;
2400 struct gfs2_holder rgd_gh;
2401 int error = -EINVAL;
2403 rgd = gfs2_blk2rgrpd(sdp, no_addr, 1);
2407 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh);
2411 if (gfs2_get_block_type(rgd, no_addr) != type)
2414 gfs2_glock_dq_uninit(&rgd_gh);
2420 * gfs2_rlist_add - add a RG to a list of RGs
2422 * @rlist: the list of resource groups
2425 * Figure out what RG a block belongs to and add that RG to the list
2427 * FIXME: Don't use NOFAIL
2431 void gfs2_rlist_add(struct gfs2_inode *ip, struct gfs2_rgrp_list *rlist,
2434 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2435 struct gfs2_rgrpd *rgd;
2436 struct gfs2_rgrpd **tmp;
2437 unsigned int new_space;
2440 if (gfs2_assert_warn(sdp, !rlist->rl_ghs))
2443 if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, block))
2446 rgd = gfs2_blk2rgrpd(sdp, block, 1);
2448 fs_err(sdp, "rlist_add: no rgrp for block %llu\n", (unsigned long long)block);
2453 for (x = 0; x < rlist->rl_rgrps; x++)
2454 if (rlist->rl_rgd[x] == rgd)
2457 if (rlist->rl_rgrps == rlist->rl_space) {
2458 new_space = rlist->rl_space + 10;
2460 tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *),
2461 GFP_NOFS | __GFP_NOFAIL);
2463 if (rlist->rl_rgd) {
2464 memcpy(tmp, rlist->rl_rgd,
2465 rlist->rl_space * sizeof(struct gfs2_rgrpd *));
2466 kfree(rlist->rl_rgd);
2469 rlist->rl_space = new_space;
2470 rlist->rl_rgd = tmp;
2473 rlist->rl_rgd[rlist->rl_rgrps++] = rgd;
2477 * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate
2478 * and initialize an array of glock holders for them
2479 * @rlist: the list of resource groups
2480 * @state: the lock state to acquire the RG lock in
2482 * FIXME: Don't use NOFAIL
2486 void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state)
2490 rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder),
2491 GFP_NOFS | __GFP_NOFAIL);
2492 for (x = 0; x < rlist->rl_rgrps; x++)
2493 gfs2_holder_init(rlist->rl_rgd[x]->rd_gl,
2499 * gfs2_rlist_free - free a resource group list
2500 * @list: the list of resource groups
2504 void gfs2_rlist_free(struct gfs2_rgrp_list *rlist)
2508 kfree(rlist->rl_rgd);
2510 if (rlist->rl_ghs) {
2511 for (x = 0; x < rlist->rl_rgrps; x++)
2512 gfs2_holder_uninit(&rlist->rl_ghs[x]);
2513 kfree(rlist->rl_ghs);
2514 rlist->rl_ghs = NULL;