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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/slab.h>
13 #include <linux/spinlock.h>
14 #include <linux/completion.h>
15 #include <linux/buffer_head.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>
36 #include "trace_gfs2.h"
38 #define BFITNOENT ((u32)~0)
39 #define NO_BLOCK ((u64)~0)
41 #if BITS_PER_LONG == 32
42 #define LBITMASK (0x55555555UL)
43 #define LBITSKIP55 (0x55555555UL)
44 #define LBITSKIP00 (0x00000000UL)
46 #define LBITMASK (0x5555555555555555UL)
47 #define LBITSKIP55 (0x5555555555555555UL)
48 #define LBITSKIP00 (0x0000000000000000UL)
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.
57 * 1 = Used (not metadata)
58 * 2 = Unlinked (still in use) inode
67 static const char valid_change[16] = {
75 static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 *minext,
76 const struct gfs2_inode *ip, bool nowrap);
80 * gfs2_setbit - Set a bit in the bitmaps
81 * @rbm: The position of the bit to set
82 * @do_clone: Also set the clone bitmap, if it exists
83 * @new_state: the new state of the block
87 static inline void gfs2_setbit(const struct gfs2_rbm *rbm, bool do_clone,
88 unsigned char new_state)
90 unsigned char *byte1, *byte2, *end, cur_state;
91 struct gfs2_bitmap *bi = rbm_bi(rbm);
92 unsigned int buflen = bi->bi_len;
93 const unsigned int bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE;
95 byte1 = bi->bi_bh->b_data + bi->bi_offset + (rbm->offset / GFS2_NBBY);
96 end = bi->bi_bh->b_data + bi->bi_offset + buflen;
100 cur_state = (*byte1 >> bit) & GFS2_BIT_MASK;
102 if (unlikely(!valid_change[new_state * 4 + cur_state])) {
103 pr_warn("buf_blk = 0x%x old_state=%d, new_state=%d\n",
104 rbm->offset, cur_state, new_state);
105 pr_warn("rgrp=0x%llx bi_start=0x%x\n",
106 (unsigned long long)rbm->rgd->rd_addr, bi->bi_start);
107 pr_warn("bi_offset=0x%x bi_len=0x%x\n",
108 bi->bi_offset, bi->bi_len);
110 gfs2_consist_rgrpd(rbm->rgd);
113 *byte1 ^= (cur_state ^ new_state) << bit;
115 if (do_clone && bi->bi_clone) {
116 byte2 = bi->bi_clone + bi->bi_offset + (rbm->offset / GFS2_NBBY);
117 cur_state = (*byte2 >> bit) & GFS2_BIT_MASK;
118 *byte2 ^= (cur_state ^ new_state) << bit;
123 * gfs2_testbit - test a bit in the bitmaps
124 * @rbm: The bit to test
126 * Returns: The two bit block state of the requested bit
129 static inline u8 gfs2_testbit(const struct gfs2_rbm *rbm)
131 struct gfs2_bitmap *bi = rbm_bi(rbm);
132 const u8 *buffer = bi->bi_bh->b_data + bi->bi_offset;
136 byte = buffer + (rbm->offset / GFS2_NBBY);
137 bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE;
139 return (*byte >> bit) & GFS2_BIT_MASK;
144 * @ptr: Pointer to bitmap data
145 * @mask: Mask to use (normally 0x55555.... but adjusted for search start)
146 * @state: The state we are searching for
148 * We xor the bitmap data with a patter which is the bitwise opposite
149 * of what we are looking for, this gives rise to a pattern of ones
150 * wherever there is a match. Since we have two bits per entry, we
151 * take this pattern, shift it down by one place and then and it with
152 * the original. All the even bit positions (0,2,4, etc) then represent
153 * successful matches, so we mask with 0x55555..... to remove the unwanted
156 * This allows searching of a whole u64 at once (32 blocks) with a
157 * single test (on 64 bit arches).
160 static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state)
163 static const u64 search[] = {
164 [0] = 0xffffffffffffffffULL,
165 [1] = 0xaaaaaaaaaaaaaaaaULL,
166 [2] = 0x5555555555555555ULL,
167 [3] = 0x0000000000000000ULL,
169 tmp = le64_to_cpu(*ptr) ^ search[state];
176 * rs_cmp - multi-block reservation range compare
177 * @blk: absolute file system block number of the new reservation
178 * @len: number of blocks in the new reservation
179 * @rs: existing reservation to compare against
181 * returns: 1 if the block range is beyond the reach of the reservation
182 * -1 if the block range is before the start of the reservation
183 * 0 if the block range overlaps with the reservation
185 static inline int rs_cmp(u64 blk, u32 len, struct gfs2_blkreserv *rs)
187 u64 startblk = gfs2_rbm_to_block(&rs->rs_rbm);
189 if (blk >= startblk + rs->rs_free)
191 if (blk + len - 1 < startblk)
197 * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing
198 * a block in a given allocation state.
199 * @buf: the buffer that holds the bitmaps
200 * @len: the length (in bytes) of the buffer
201 * @goal: start search at this block's bit-pair (within @buffer)
202 * @state: GFS2_BLKST_XXX the state of the block we're looking for.
204 * Scope of @goal and returned block number is only within this bitmap buffer,
205 * not entire rgrp or filesystem. @buffer will be offset from the actual
206 * beginning of a bitmap block buffer, skipping any header structures, but
207 * headers are always a multiple of 64 bits long so that the buffer is
208 * always aligned to a 64 bit boundary.
210 * The size of the buffer is in bytes, but is it assumed that it is
211 * always ok to read a complete multiple of 64 bits at the end
212 * of the block in case the end is no aligned to a natural boundary.
214 * Return: the block number (bitmap buffer scope) that was found
217 static u32 gfs2_bitfit(const u8 *buf, const unsigned int len,
220 u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1);
221 const __le64 *ptr = ((__le64 *)buf) + (goal >> 5);
222 const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64)));
224 u64 mask = 0x5555555555555555ULL;
227 /* Mask off bits we don't care about at the start of the search */
229 tmp = gfs2_bit_search(ptr, mask, state);
231 while(tmp == 0 && ptr < end) {
232 tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state);
235 /* Mask off any bits which are more than len bytes from the start */
236 if (ptr == end && (len & (sizeof(u64) - 1)))
237 tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1))));
238 /* Didn't find anything, so return */
243 bit /= 2; /* two bits per entry in the bitmap */
244 return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit;
248 * gfs2_rbm_from_block - Set the rbm based upon rgd and block number
249 * @rbm: The rbm with rgd already set correctly
250 * @block: The block number (filesystem relative)
252 * This sets the bi and offset members of an rbm based on a
253 * resource group and a filesystem relative block number. The
254 * resource group must be set in the rbm on entry, the bi and
255 * offset members will be set by this function.
257 * Returns: 0 on success, or an error code
260 static int gfs2_rbm_from_block(struct gfs2_rbm *rbm, u64 block)
262 u64 rblock = block - rbm->rgd->rd_data0;
264 if (WARN_ON_ONCE(rblock > UINT_MAX))
266 if (block >= rbm->rgd->rd_data0 + rbm->rgd->rd_data)
270 rbm->offset = (u32)(rblock);
271 /* Check if the block is within the first block */
272 if (rbm->offset < rbm_bi(rbm)->bi_blocks)
275 /* Adjust for the size diff between gfs2_meta_header and gfs2_rgrp */
276 rbm->offset += (sizeof(struct gfs2_rgrp) -
277 sizeof(struct gfs2_meta_header)) * GFS2_NBBY;
278 rbm->bii = rbm->offset / rbm->rgd->rd_sbd->sd_blocks_per_bitmap;
279 rbm->offset -= rbm->bii * rbm->rgd->rd_sbd->sd_blocks_per_bitmap;
284 * gfs2_rbm_incr - increment an rbm structure
285 * @rbm: The rbm with rgd already set correctly
287 * This function takes an existing rbm structure and increments it to the next
288 * viable block offset.
290 * Returns: If incrementing the offset would cause the rbm to go past the
291 * end of the rgrp, true is returned, otherwise false.
295 static bool gfs2_rbm_incr(struct gfs2_rbm *rbm)
297 if (rbm->offset + 1 < rbm_bi(rbm)->bi_blocks) { /* in the same bitmap */
301 if (rbm->bii == rbm->rgd->rd_length - 1) /* at the last bitmap */
310 * gfs2_unaligned_extlen - Look for free blocks which are not byte aligned
311 * @rbm: Position to search (value/result)
312 * @n_unaligned: Number of unaligned blocks to check
313 * @len: Decremented for each block found (terminate on zero)
315 * Returns: true if a non-free block is encountered
318 static bool gfs2_unaligned_extlen(struct gfs2_rbm *rbm, u32 n_unaligned, u32 *len)
323 for (n = 0; n < n_unaligned; n++) {
324 res = gfs2_testbit(rbm);
325 if (res != GFS2_BLKST_FREE)
330 if (gfs2_rbm_incr(rbm))
338 * gfs2_free_extlen - Return extent length of free blocks
339 * @rrbm: Starting position
340 * @len: Max length to check
342 * Starting at the block specified by the rbm, see how many free blocks
343 * there are, not reading more than len blocks ahead. This can be done
344 * using memchr_inv when the blocks are byte aligned, but has to be done
345 * on a block by block basis in case of unaligned blocks. Also this
346 * function can cope with bitmap boundaries (although it must stop on
347 * a resource group boundary)
349 * Returns: Number of free blocks in the extent
352 static u32 gfs2_free_extlen(const struct gfs2_rbm *rrbm, u32 len)
354 struct gfs2_rbm rbm = *rrbm;
355 u32 n_unaligned = rbm.offset & 3;
359 u8 *ptr, *start, *end;
361 struct gfs2_bitmap *bi;
364 gfs2_unaligned_extlen(&rbm, 4 - n_unaligned, &len))
367 n_unaligned = len & 3;
368 /* Start is now byte aligned */
371 start = bi->bi_bh->b_data;
373 start = bi->bi_clone;
374 end = start + bi->bi_bh->b_size;
375 start += bi->bi_offset;
376 BUG_ON(rbm.offset & 3);
377 start += (rbm.offset / GFS2_NBBY);
378 bytes = min_t(u32, len / GFS2_NBBY, (end - start));
379 ptr = memchr_inv(start, 0, bytes);
380 chunk_size = ((ptr == NULL) ? bytes : (ptr - start));
381 chunk_size *= GFS2_NBBY;
382 BUG_ON(len < chunk_size);
384 block = gfs2_rbm_to_block(&rbm);
385 if (gfs2_rbm_from_block(&rbm, block + chunk_size)) {
393 n_unaligned = len & 3;
396 /* Deal with any bits left over at the end */
398 gfs2_unaligned_extlen(&rbm, n_unaligned, &len);
404 * gfs2_bitcount - count the number of bits in a certain state
405 * @rgd: the resource group descriptor
406 * @buffer: the buffer that holds the bitmaps
407 * @buflen: the length (in bytes) of the buffer
408 * @state: the state of the block we're looking for
410 * Returns: The number of bits
413 static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer,
414 unsigned int buflen, u8 state)
416 const u8 *byte = buffer;
417 const u8 *end = buffer + buflen;
418 const u8 state1 = state << 2;
419 const u8 state2 = state << 4;
420 const u8 state3 = state << 6;
423 for (; byte < end; byte++) {
424 if (((*byte) & 0x03) == state)
426 if (((*byte) & 0x0C) == state1)
428 if (((*byte) & 0x30) == state2)
430 if (((*byte) & 0xC0) == state3)
438 * gfs2_rgrp_verify - Verify that a resource group is consistent
443 void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd)
445 struct gfs2_sbd *sdp = rgd->rd_sbd;
446 struct gfs2_bitmap *bi = NULL;
447 u32 length = rgd->rd_length;
451 memset(count, 0, 4 * sizeof(u32));
453 /* Count # blocks in each of 4 possible allocation states */
454 for (buf = 0; buf < length; buf++) {
455 bi = rgd->rd_bits + buf;
456 for (x = 0; x < 4; x++)
457 count[x] += gfs2_bitcount(rgd,
463 if (count[0] != rgd->rd_free) {
464 if (gfs2_consist_rgrpd(rgd))
465 fs_err(sdp, "free data mismatch: %u != %u\n",
466 count[0], rgd->rd_free);
470 tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes;
471 if (count[1] != tmp) {
472 if (gfs2_consist_rgrpd(rgd))
473 fs_err(sdp, "used data mismatch: %u != %u\n",
478 if (count[2] + count[3] != rgd->rd_dinodes) {
479 if (gfs2_consist_rgrpd(rgd))
480 fs_err(sdp, "used metadata mismatch: %u != %u\n",
481 count[2] + count[3], rgd->rd_dinodes);
486 static inline int rgrp_contains_block(struct gfs2_rgrpd *rgd, u64 block)
488 u64 first = rgd->rd_data0;
489 u64 last = first + rgd->rd_data;
490 return first <= block && block < last;
494 * gfs2_blk2rgrpd - Find resource group for a given data/meta block number
495 * @sdp: The GFS2 superblock
496 * @blk: The data block number
497 * @exact: True if this needs to be an exact match
499 * Returns: The resource group, or NULL if not found
502 struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk, bool exact)
504 struct rb_node *n, *next;
505 struct gfs2_rgrpd *cur;
507 spin_lock(&sdp->sd_rindex_spin);
508 n = sdp->sd_rindex_tree.rb_node;
510 cur = rb_entry(n, struct gfs2_rgrpd, rd_node);
512 if (blk < cur->rd_addr)
514 else if (blk >= cur->rd_data0 + cur->rd_data)
517 spin_unlock(&sdp->sd_rindex_spin);
519 if (blk < cur->rd_addr)
521 if (blk >= cur->rd_data0 + cur->rd_data)
528 spin_unlock(&sdp->sd_rindex_spin);
534 * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem
535 * @sdp: The GFS2 superblock
537 * Returns: The first rgrp in the filesystem
540 struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp)
542 const struct rb_node *n;
543 struct gfs2_rgrpd *rgd;
545 spin_lock(&sdp->sd_rindex_spin);
546 n = rb_first(&sdp->sd_rindex_tree);
547 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node);
548 spin_unlock(&sdp->sd_rindex_spin);
554 * gfs2_rgrpd_get_next - get the next RG
555 * @rgd: the resource group descriptor
557 * Returns: The next rgrp
560 struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd)
562 struct gfs2_sbd *sdp = rgd->rd_sbd;
563 const struct rb_node *n;
565 spin_lock(&sdp->sd_rindex_spin);
566 n = rb_next(&rgd->rd_node);
568 n = rb_first(&sdp->sd_rindex_tree);
570 if (unlikely(&rgd->rd_node == n)) {
571 spin_unlock(&sdp->sd_rindex_spin);
574 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node);
575 spin_unlock(&sdp->sd_rindex_spin);
579 void check_and_update_goal(struct gfs2_inode *ip)
581 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
582 if (!ip->i_goal || gfs2_blk2rgrpd(sdp, ip->i_goal, 1) == NULL)
583 ip->i_goal = ip->i_no_addr;
586 void gfs2_free_clones(struct gfs2_rgrpd *rgd)
590 for (x = 0; x < rgd->rd_length; x++) {
591 struct gfs2_bitmap *bi = rgd->rd_bits + x;
598 * gfs2_rsqa_alloc - make sure we have a reservation assigned to the inode
599 * plus a quota allocations data structure, if necessary
600 * @ip: the inode for this reservation
602 int gfs2_rsqa_alloc(struct gfs2_inode *ip)
604 return gfs2_qa_alloc(ip);
607 static void dump_rs(struct seq_file *seq, const struct gfs2_blkreserv *rs)
609 gfs2_print_dbg(seq, " B: n:%llu s:%llu b:%u f:%u\n",
610 (unsigned long long)rs->rs_inum,
611 (unsigned long long)gfs2_rbm_to_block(&rs->rs_rbm),
612 rs->rs_rbm.offset, rs->rs_free);
616 * __rs_deltree - remove a multi-block reservation from the rgd tree
617 * @rs: The reservation to remove
620 static void __rs_deltree(struct gfs2_blkreserv *rs)
622 struct gfs2_rgrpd *rgd;
624 if (!gfs2_rs_active(rs))
627 rgd = rs->rs_rbm.rgd;
628 trace_gfs2_rs(rs, TRACE_RS_TREEDEL);
629 rb_erase(&rs->rs_node, &rgd->rd_rstree);
630 RB_CLEAR_NODE(&rs->rs_node);
633 u64 last_block = gfs2_rbm_to_block(&rs->rs_rbm) +
635 struct gfs2_rbm last_rbm = { .rgd = rs->rs_rbm.rgd, };
636 struct gfs2_bitmap *start, *last;
638 /* return reserved blocks to the rgrp */
639 BUG_ON(rs->rs_rbm.rgd->rd_reserved < rs->rs_free);
640 rs->rs_rbm.rgd->rd_reserved -= rs->rs_free;
641 /* The rgrp extent failure point is likely not to increase;
642 it will only do so if the freed blocks are somehow
643 contiguous with a span of free blocks that follows. Still,
644 it will force the number to be recalculated later. */
645 rgd->rd_extfail_pt += rs->rs_free;
647 if (gfs2_rbm_from_block(&last_rbm, last_block))
649 start = rbm_bi(&rs->rs_rbm);
650 last = rbm_bi(&last_rbm);
652 clear_bit(GBF_FULL, &start->bi_flags);
653 while (start++ != last);
658 * gfs2_rs_deltree - remove a multi-block reservation from the rgd tree
659 * @rs: The reservation to remove
662 void gfs2_rs_deltree(struct gfs2_blkreserv *rs)
664 struct gfs2_rgrpd *rgd;
666 rgd = rs->rs_rbm.rgd;
668 spin_lock(&rgd->rd_rsspin);
671 spin_unlock(&rgd->rd_rsspin);
676 * gfs2_rsqa_delete - delete a multi-block reservation and quota allocation
677 * @ip: The inode for this reservation
678 * @wcount: The inode's write count, or NULL
681 void gfs2_rsqa_delete(struct gfs2_inode *ip, atomic_t *wcount)
683 down_write(&ip->i_rw_mutex);
684 if ((wcount == NULL) || (atomic_read(wcount) <= 1))
685 gfs2_rs_deltree(&ip->i_res);
686 up_write(&ip->i_rw_mutex);
687 gfs2_qa_delete(ip, wcount);
691 * return_all_reservations - return all reserved blocks back to the rgrp.
692 * @rgd: the rgrp that needs its space back
694 * We previously reserved a bunch of blocks for allocation. Now we need to
695 * give them back. This leave the reservation structures in tact, but removes
696 * all of their corresponding "no-fly zones".
698 static void return_all_reservations(struct gfs2_rgrpd *rgd)
701 struct gfs2_blkreserv *rs;
703 spin_lock(&rgd->rd_rsspin);
704 while ((n = rb_first(&rgd->rd_rstree))) {
705 rs = rb_entry(n, struct gfs2_blkreserv, rs_node);
708 spin_unlock(&rgd->rd_rsspin);
711 void gfs2_clear_rgrpd(struct gfs2_sbd *sdp)
714 struct gfs2_rgrpd *rgd;
715 struct gfs2_glock *gl;
717 while ((n = rb_first(&sdp->sd_rindex_tree))) {
718 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node);
721 rb_erase(n, &sdp->sd_rindex_tree);
724 spin_lock(&gl->gl_lockref.lock);
725 gl->gl_object = NULL;
726 spin_unlock(&gl->gl_lockref.lock);
727 gfs2_rgrp_brelse(rgd);
728 gfs2_glock_add_to_lru(gl);
732 gfs2_free_clones(rgd);
733 return_all_reservations(rgd);
736 kmem_cache_free(gfs2_rgrpd_cachep, rgd);
740 static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd)
742 pr_info("ri_addr = %llu\n", (unsigned long long)rgd->rd_addr);
743 pr_info("ri_length = %u\n", rgd->rd_length);
744 pr_info("ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0);
745 pr_info("ri_data = %u\n", rgd->rd_data);
746 pr_info("ri_bitbytes = %u\n", rgd->rd_bitbytes);
750 * gfs2_compute_bitstructs - Compute the bitmap sizes
751 * @rgd: The resource group descriptor
753 * Calculates bitmap descriptors, one for each block that contains bitmap data
758 static int compute_bitstructs(struct gfs2_rgrpd *rgd)
760 struct gfs2_sbd *sdp = rgd->rd_sbd;
761 struct gfs2_bitmap *bi;
762 u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */
763 u32 bytes_left, bytes;
769 rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS);
773 bytes_left = rgd->rd_bitbytes;
775 for (x = 0; x < length; x++) {
776 bi = rgd->rd_bits + x;
779 /* small rgrp; bitmap stored completely in header block */
782 bi->bi_offset = sizeof(struct gfs2_rgrp);
785 bi->bi_blocks = bytes * GFS2_NBBY;
788 bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp);
789 bi->bi_offset = sizeof(struct gfs2_rgrp);
792 bi->bi_blocks = bytes * GFS2_NBBY;
794 } else if (x + 1 == length) {
796 bi->bi_offset = sizeof(struct gfs2_meta_header);
797 bi->bi_start = rgd->rd_bitbytes - bytes_left;
799 bi->bi_blocks = bytes * GFS2_NBBY;
802 bytes = sdp->sd_sb.sb_bsize -
803 sizeof(struct gfs2_meta_header);
804 bi->bi_offset = sizeof(struct gfs2_meta_header);
805 bi->bi_start = rgd->rd_bitbytes - bytes_left;
807 bi->bi_blocks = bytes * GFS2_NBBY;
814 gfs2_consist_rgrpd(rgd);
817 bi = rgd->rd_bits + (length - 1);
818 if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) {
819 if (gfs2_consist_rgrpd(rgd)) {
820 gfs2_rindex_print(rgd);
821 fs_err(sdp, "start=%u len=%u offset=%u\n",
822 bi->bi_start, bi->bi_len, bi->bi_offset);
831 * gfs2_ri_total - Total up the file system space, according to the rindex.
832 * @sdp: the filesystem
835 u64 gfs2_ri_total(struct gfs2_sbd *sdp)
838 struct inode *inode = sdp->sd_rindex;
839 struct gfs2_inode *ip = GFS2_I(inode);
840 char buf[sizeof(struct gfs2_rindex)];
843 for (rgrps = 0;; rgrps++) {
844 loff_t pos = rgrps * sizeof(struct gfs2_rindex);
846 if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode))
848 error = gfs2_internal_read(ip, buf, &pos,
849 sizeof(struct gfs2_rindex));
850 if (error != sizeof(struct gfs2_rindex))
852 total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data);
857 static int rgd_insert(struct gfs2_rgrpd *rgd)
859 struct gfs2_sbd *sdp = rgd->rd_sbd;
860 struct rb_node **newn = &sdp->sd_rindex_tree.rb_node, *parent = NULL;
862 /* Figure out where to put new node */
864 struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd,
868 if (rgd->rd_addr < cur->rd_addr)
869 newn = &((*newn)->rb_left);
870 else if (rgd->rd_addr > cur->rd_addr)
871 newn = &((*newn)->rb_right);
876 rb_link_node(&rgd->rd_node, parent, newn);
877 rb_insert_color(&rgd->rd_node, &sdp->sd_rindex_tree);
883 * read_rindex_entry - Pull in a new resource index entry from the disk
884 * @ip: Pointer to the rindex inode
886 * Returns: 0 on success, > 0 on EOF, error code otherwise
889 static int read_rindex_entry(struct gfs2_inode *ip)
891 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
892 const unsigned bsize = sdp->sd_sb.sb_bsize;
893 loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex);
894 struct gfs2_rindex buf;
896 struct gfs2_rgrpd *rgd;
898 if (pos >= i_size_read(&ip->i_inode))
901 error = gfs2_internal_read(ip, (char *)&buf, &pos,
902 sizeof(struct gfs2_rindex));
904 if (error != sizeof(struct gfs2_rindex))
905 return (error == 0) ? 1 : error;
907 rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS);
913 rgd->rd_addr = be64_to_cpu(buf.ri_addr);
914 rgd->rd_length = be32_to_cpu(buf.ri_length);
915 rgd->rd_data0 = be64_to_cpu(buf.ri_data0);
916 rgd->rd_data = be32_to_cpu(buf.ri_data);
917 rgd->rd_bitbytes = be32_to_cpu(buf.ri_bitbytes);
918 spin_lock_init(&rgd->rd_rsspin);
920 error = compute_bitstructs(rgd);
924 error = gfs2_glock_get(sdp, rgd->rd_addr,
925 &gfs2_rgrp_glops, CREATE, &rgd->rd_gl);
929 rgd->rd_rgl = (struct gfs2_rgrp_lvb *)rgd->rd_gl->gl_lksb.sb_lvbptr;
930 rgd->rd_flags &= ~(GFS2_RDF_UPTODATE | GFS2_RDF_PREFERRED);
931 if (rgd->rd_data > sdp->sd_max_rg_data)
932 sdp->sd_max_rg_data = rgd->rd_data;
933 spin_lock(&sdp->sd_rindex_spin);
934 error = rgd_insert(rgd);
935 spin_unlock(&sdp->sd_rindex_spin);
937 rgd->rd_gl->gl_object = rgd;
938 rgd->rd_gl->gl_vm.start = (rgd->rd_addr * bsize) & PAGE_MASK;
939 rgd->rd_gl->gl_vm.end = PAGE_ALIGN((rgd->rd_addr +
940 rgd->rd_length) * bsize) - 1;
944 error = 0; /* someone else read in the rgrp; free it and ignore it */
945 gfs2_glock_put(rgd->rd_gl);
950 kmem_cache_free(gfs2_rgrpd_cachep, rgd);
955 * set_rgrp_preferences - Run all the rgrps, selecting some we prefer to use
956 * @sdp: the GFS2 superblock
958 * The purpose of this function is to select a subset of the resource groups
959 * and mark them as PREFERRED. We do it in such a way that each node prefers
960 * to use a unique set of rgrps to minimize glock contention.
962 static void set_rgrp_preferences(struct gfs2_sbd *sdp)
964 struct gfs2_rgrpd *rgd, *first;
967 /* Skip an initial number of rgrps, based on this node's journal ID.
968 That should start each node out on its own set. */
969 rgd = gfs2_rgrpd_get_first(sdp);
970 for (i = 0; i < sdp->sd_lockstruct.ls_jid; i++)
971 rgd = gfs2_rgrpd_get_next(rgd);
975 rgd->rd_flags |= GFS2_RDF_PREFERRED;
976 for (i = 0; i < sdp->sd_journals; i++) {
977 rgd = gfs2_rgrpd_get_next(rgd);
978 if (!rgd || rgd == first)
981 } while (rgd && rgd != first);
985 * gfs2_ri_update - Pull in a new resource index from the disk
986 * @ip: pointer to the rindex inode
988 * Returns: 0 on successful update, error code otherwise
991 static int gfs2_ri_update(struct gfs2_inode *ip)
993 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
997 error = read_rindex_entry(ip);
998 } while (error == 0);
1003 if (RB_EMPTY_ROOT(&sdp->sd_rindex_tree)) {
1004 fs_err(sdp, "no resource groups found in the file system.\n");
1007 set_rgrp_preferences(sdp);
1009 sdp->sd_rindex_uptodate = 1;
1014 * gfs2_rindex_update - Update the rindex if required
1015 * @sdp: The GFS2 superblock
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).
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.
1027 * Returns: 0 on succeess, error code otherwise
1030 int gfs2_rindex_update(struct gfs2_sbd *sdp)
1032 struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex);
1033 struct gfs2_glock *gl = ip->i_gl;
1034 struct gfs2_holder ri_gh;
1036 int unlock_required = 0;
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);
1044 unlock_required = 1;
1046 if (!sdp->sd_rindex_uptodate)
1047 error = gfs2_ri_update(ip);
1048 if (unlock_required)
1049 gfs2_glock_dq_uninit(&ri_gh);
1055 static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf)
1057 const struct gfs2_rgrp *str = buf;
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);
1069 static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf)
1071 struct gfs2_rgrp *str = buf;
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));
1081 static int gfs2_rgrp_lvb_valid(struct gfs2_rgrpd *rgd)
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;
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)
1093 static void gfs2_rgrp_ondisk2lvb(struct gfs2_rgrp_lvb *rgl, const void *buf)
1095 const struct gfs2_rgrp *str = buf;
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;
1105 static void update_rgrp_lvb_unlinked(struct gfs2_rgrpd *rgd, u32 change)
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);
1112 static u32 count_unlinked(struct gfs2_rgrpd *rgd)
1114 struct gfs2_bitmap *bi;
1115 const u32 length = rgd->rd_length;
1116 const u8 *buffer = NULL;
1117 u32 i, goal, count = 0;
1119 for (i = 0, bi = rgd->rd_bits; i < length; i++, bi++) {
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)
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
1141 * Read in all of a Resource Group's header and bitmap blocks.
1142 * Caller must eventually call gfs2_rgrp_brelse() to free the bitmaps.
1147 static int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd)
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;
1156 if (rgd->rd_bits[0].bi_bh != NULL)
1159 for (x = 0; x < length; x++) {
1160 bi = rgd->rd_bits + x;
1161 error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, 0, &bi->bi_bh);
1166 for (y = length; y--;) {
1167 bi = rgd->rd_bits + y;
1168 error = gfs2_meta_wait(sdp, bi->bi_bh);
1171 if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB :
1172 GFS2_METATYPE_RG)) {
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;
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);
1192 else if (sdp->sd_args.ar_rgrplvb) {
1193 if (!gfs2_rgrp_lvb_valid(rgd)){
1194 gfs2_consist_rgrpd(rgd);
1198 if (rgd->rd_rgl->rl_unlinked == 0)
1199 rgd->rd_flags &= ~GFS2_RDF_CHECK;
1205 bi = rgd->rd_bits + x;
1208 gfs2_assert_warn(sdp, !bi->bi_clone);
1214 static int update_rgrp_lvb(struct gfs2_rgrpd *rgd)
1218 if (rgd->rd_flags & GFS2_RDF_UPTODATE)
1221 if (cpu_to_be32(GFS2_MAGIC) != rgd->rd_rgl->rl_magic)
1222 return gfs2_rgrp_bh_get(rgd);
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_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);
1237 int gfs2_rgrp_go_lock(struct gfs2_holder *gh)
1239 struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object;
1240 struct gfs2_sbd *sdp = rgd->rd_sbd;
1242 if (gh->gh_flags & GL_SKIP && sdp->sd_args.ar_rgrplvb)
1244 return gfs2_rgrp_bh_get(rgd);
1248 * gfs2_rgrp_brelse - Release RG bitmaps read in with gfs2_rgrp_bh_get()
1249 * @rgd: The resource group
1253 void gfs2_rgrp_brelse(struct gfs2_rgrpd *rgd)
1255 int x, length = rgd->rd_length;
1257 for (x = 0; x < length; x++) {
1258 struct gfs2_bitmap *bi = rgd->rd_bits + x;
1268 * gfs2_rgrp_go_unlock - Unlock a rgrp glock
1269 * @gh: The glock holder for the resource group
1273 void gfs2_rgrp_go_unlock(struct gfs2_holder *gh)
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);
1279 if (rgd && demote_requested)
1280 gfs2_rgrp_brelse(rgd);
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)
1287 struct super_block *sb = sdp->sd_vfs;
1290 sector_t nr_blks = 0;
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;
1301 const u8 *orig = bh->b_data + bi->bi_offset + x;
1302 diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1));
1304 diff = ~(*clone | (*clone >> 1));
1309 blk = offset + ((bi->bi_start + x) * GFS2_NBBY);
1313 goto start_new_extent;
1314 if ((start + nr_blks) != blk) {
1315 if (nr_blks >= minlen) {
1316 rv = sb_issue_discard(sb,
1333 if (nr_blks >= minlen) {
1334 rv = sb_issue_discard(sb, start, nr_blks, GFP_NOFS, 0);
1340 *ptrimmed = trimmed;
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;
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)
1355 * Returns: 0 on success, otherwise error code
1358 int gfs2_fitrim(struct file *filp, void __user *argp)
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;
1371 u64 start, end, minlen;
1373 unsigned bs_shift = sdp->sd_sb.sb_bsize_shift;
1375 if (!capable(CAP_SYS_ADMIN))
1378 if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
1381 if (!blk_queue_discard(q))
1384 if (copy_from_user(&r, argp, sizeof(r)))
1387 ret = gfs2_rindex_update(sdp);
1391 start = r.start >> bs_shift;
1392 end = start + (r.len >> bs_shift);
1393 minlen = max_t(u64, r.minlen,
1394 q->limits.discard_granularity) >> bs_shift;
1396 if (end <= start || minlen > sdp->sd_max_rg_data)
1399 rgd = gfs2_blk2rgrpd(sdp, start, 0);
1400 rgd_end = gfs2_blk2rgrpd(sdp, end, 0);
1402 if ((gfs2_rgrpd_get_first(sdp) == gfs2_rgrpd_get_next(rgd_end))
1403 && (start > rgd_end->rd_data0 + rgd_end->rd_data))
1404 return -EINVAL; /* start is beyond the end of the fs */
1408 ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1412 if (!(rgd->rd_flags & GFS2_RGF_TRIMMED)) {
1413 /* Trim each bitmap in the rgrp */
1414 for (x = 0; x < rgd->rd_length; x++) {
1415 struct gfs2_bitmap *bi = rgd->rd_bits + x;
1416 ret = gfs2_rgrp_send_discards(sdp,
1417 rgd->rd_data0, NULL, bi, minlen,
1420 gfs2_glock_dq_uninit(&gh);
1426 /* Mark rgrp as having been trimmed */
1427 ret = gfs2_trans_begin(sdp, RES_RG_HDR, 0);
1429 bh = rgd->rd_bits[0].bi_bh;
1430 rgd->rd_flags |= GFS2_RGF_TRIMMED;
1431 gfs2_trans_add_meta(rgd->rd_gl, bh);
1432 gfs2_rgrp_out(rgd, bh->b_data);
1433 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, bh->b_data);
1434 gfs2_trans_end(sdp);
1437 gfs2_glock_dq_uninit(&gh);
1442 rgd = gfs2_rgrpd_get_next(rgd);
1446 r.len = trimmed << bs_shift;
1447 if (copy_to_user(argp, &r, sizeof(r)))
1454 * rs_insert - insert a new multi-block reservation into the rgrp's rb_tree
1455 * @ip: the inode structure
1458 static void rs_insert(struct gfs2_inode *ip)
1460 struct rb_node **newn, *parent = NULL;
1462 struct gfs2_blkreserv *rs = &ip->i_res;
1463 struct gfs2_rgrpd *rgd = rs->rs_rbm.rgd;
1464 u64 fsblock = gfs2_rbm_to_block(&rs->rs_rbm);
1466 BUG_ON(gfs2_rs_active(rs));
1468 spin_lock(&rgd->rd_rsspin);
1469 newn = &rgd->rd_rstree.rb_node;
1471 struct gfs2_blkreserv *cur =
1472 rb_entry(*newn, struct gfs2_blkreserv, rs_node);
1475 rc = rs_cmp(fsblock, rs->rs_free, cur);
1477 newn = &((*newn)->rb_right);
1479 newn = &((*newn)->rb_left);
1481 spin_unlock(&rgd->rd_rsspin);
1487 rb_link_node(&rs->rs_node, parent, newn);
1488 rb_insert_color(&rs->rs_node, &rgd->rd_rstree);
1490 /* Do our rgrp accounting for the reservation */
1491 rgd->rd_reserved += rs->rs_free; /* blocks reserved */
1492 spin_unlock(&rgd->rd_rsspin);
1493 trace_gfs2_rs(rs, TRACE_RS_INSERT);
1497 * rg_mblk_search - find a group of multiple free blocks to form a reservation
1498 * @rgd: the resource group descriptor
1499 * @ip: pointer to the inode for which we're reserving blocks
1500 * @ap: the allocation parameters
1504 static void rg_mblk_search(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip,
1505 const struct gfs2_alloc_parms *ap)
1507 struct gfs2_rbm rbm = { .rgd = rgd, };
1509 struct gfs2_blkreserv *rs = &ip->i_res;
1511 u32 free_blocks = rgd->rd_free_clone - rgd->rd_reserved;
1513 struct inode *inode = &ip->i_inode;
1515 if (S_ISDIR(inode->i_mode))
1518 extlen = max_t(u32, atomic_read(&rs->rs_sizehint), ap->target);
1519 extlen = clamp(extlen, RGRP_RSRV_MINBLKS, free_blocks);
1521 if ((rgd->rd_free_clone < rgd->rd_reserved) || (free_blocks < extlen))
1524 /* Find bitmap block that contains bits for goal block */
1525 if (rgrp_contains_block(rgd, ip->i_goal))
1528 goal = rgd->rd_last_alloc + rgd->rd_data0;
1530 if (WARN_ON(gfs2_rbm_from_block(&rbm, goal)))
1533 ret = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, &extlen, ip, true);
1536 rs->rs_free = extlen;
1537 rs->rs_inum = ip->i_no_addr;
1540 if (goal == rgd->rd_last_alloc + rgd->rd_data0)
1541 rgd->rd_last_alloc = 0;
1546 * gfs2_next_unreserved_block - Return next block that is not reserved
1547 * @rgd: The resource group
1548 * @block: The starting block
1549 * @length: The required length
1550 * @ip: Ignore any reservations for this inode
1552 * If the block does not appear in any reservation, then return the
1553 * block number unchanged. If it does appear in the reservation, then
1554 * keep looking through the tree of reservations in order to find the
1555 * first block number which is not reserved.
1558 static u64 gfs2_next_unreserved_block(struct gfs2_rgrpd *rgd, u64 block,
1560 const struct gfs2_inode *ip)
1562 struct gfs2_blkreserv *rs;
1566 spin_lock(&rgd->rd_rsspin);
1567 n = rgd->rd_rstree.rb_node;
1569 rs = rb_entry(n, struct gfs2_blkreserv, rs_node);
1570 rc = rs_cmp(block, length, rs);
1580 while ((rs_cmp(block, length, rs) == 0) && (&ip->i_res != rs)) {
1581 block = gfs2_rbm_to_block(&rs->rs_rbm) + rs->rs_free;
1585 rs = rb_entry(n, struct gfs2_blkreserv, rs_node);
1589 spin_unlock(&rgd->rd_rsspin);
1594 * gfs2_reservation_check_and_update - Check for reservations during block alloc
1595 * @rbm: The current position in the resource group
1596 * @ip: The inode for which we are searching for blocks
1597 * @minext: The minimum extent length
1598 * @maxext: A pointer to the maximum extent structure
1600 * This checks the current position in the rgrp to see whether there is
1601 * a reservation covering this block. If not then this function is a
1602 * no-op. If there is, then the position is moved to the end of the
1603 * contiguous reservation(s) so that we are pointing at the first
1604 * non-reserved block.
1606 * Returns: 0 if no reservation, 1 if @rbm has changed, otherwise an error
1609 static int gfs2_reservation_check_and_update(struct gfs2_rbm *rbm,
1610 const struct gfs2_inode *ip,
1612 struct gfs2_extent *maxext)
1614 u64 block = gfs2_rbm_to_block(rbm);
1620 * If we have a minimum extent length, then skip over any extent
1621 * which is less than the min extent length in size.
1624 extlen = gfs2_free_extlen(rbm, minext);
1625 if (extlen <= maxext->len)
1630 * Check the extent which has been found against the reservations
1631 * and skip if parts of it are already reserved
1633 nblock = gfs2_next_unreserved_block(rbm->rgd, block, extlen, ip);
1634 if (nblock == block) {
1635 if (!minext || extlen >= minext)
1638 if (extlen > maxext->len) {
1639 maxext->len = extlen;
1643 nblock = block + extlen;
1645 ret = gfs2_rbm_from_block(rbm, nblock);
1652 * gfs2_rbm_find - Look for blocks of a particular state
1653 * @rbm: Value/result starting position and final position
1654 * @state: The state which we want to find
1655 * @minext: Pointer to the requested extent length (NULL for a single block)
1656 * This is updated to be the actual reservation size.
1657 * @ip: If set, check for reservations
1658 * @nowrap: Stop looking at the end of the rgrp, rather than wrapping
1659 * around until we've reached the starting point.
1662 * - If looking for free blocks, we set GBF_FULL on each bitmap which
1663 * has no free blocks in it.
1664 * - If looking for free blocks, we set rd_extfail_pt on each rgrp which
1665 * has come up short on a free block search.
1667 * Returns: 0 on success, -ENOSPC if there is no block of the requested state
1670 static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 *minext,
1671 const struct gfs2_inode *ip, bool nowrap)
1673 struct buffer_head *bh;
1676 int first_bii = rbm->bii;
1677 u32 first_offset = rbm->offset;
1681 int iters = rbm->rgd->rd_length;
1683 struct gfs2_bitmap *bi;
1684 struct gfs2_extent maxext = { .rbm.rgd = rbm->rgd, };
1686 /* If we are not starting at the beginning of a bitmap, then we
1687 * need to add one to the bitmap count to ensure that we search
1688 * the starting bitmap twice.
1690 if (rbm->offset != 0)
1695 if ((ip == NULL || !gfs2_rs_active(&ip->i_res)) &&
1696 test_bit(GBF_FULL, &bi->bi_flags) &&
1697 (state == GFS2_BLKST_FREE))
1701 buffer = bh->b_data + bi->bi_offset;
1702 WARN_ON(!buffer_uptodate(bh));
1703 if (state != GFS2_BLKST_UNLINKED && bi->bi_clone)
1704 buffer = bi->bi_clone + bi->bi_offset;
1705 initial_offset = rbm->offset;
1706 offset = gfs2_bitfit(buffer, bi->bi_len, rbm->offset, state);
1707 if (offset == BFITNOENT)
1709 rbm->offset = offset;
1713 initial_bii = rbm->bii;
1714 ret = gfs2_reservation_check_and_update(rbm, ip,
1715 minext ? *minext : 0,
1720 n += (rbm->bii - initial_bii);
1723 if (ret == -E2BIG) {
1726 n += (rbm->bii - initial_bii);
1727 goto res_covered_end_of_rgrp;
1731 bitmap_full: /* Mark bitmap as full and fall through */
1732 if ((state == GFS2_BLKST_FREE) && initial_offset == 0)
1733 set_bit(GBF_FULL, &bi->bi_flags);
1735 next_bitmap: /* Find next bitmap in the rgrp */
1738 if (rbm->bii == rbm->rgd->rd_length)
1740 res_covered_end_of_rgrp:
1741 if ((rbm->bii == 0) && nowrap)
1749 if (minext == NULL || state != GFS2_BLKST_FREE)
1752 /* If the extent was too small, and it's smaller than the smallest
1753 to have failed before, remember for future reference that it's
1754 useless to search this rgrp again for this amount or more. */
1755 if ((first_offset == 0) && (first_bii == 0) &&
1756 (*minext < rbm->rgd->rd_extfail_pt))
1757 rbm->rgd->rd_extfail_pt = *minext;
1759 /* If the maximum extent we found is big enough to fulfill the
1760 minimum requirements, use it anyway. */
1763 *minext = maxext.len;
1771 * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes
1773 * @last_unlinked: block address of the last dinode we unlinked
1774 * @skip: block address we should explicitly not unlink
1776 * Returns: 0 if no error
1777 * The inode, if one has been found, in inode.
1780 static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip)
1783 struct gfs2_sbd *sdp = rgd->rd_sbd;
1784 struct gfs2_glock *gl;
1785 struct gfs2_inode *ip;
1788 struct gfs2_rbm rbm = { .rgd = rgd, .bii = 0, .offset = 0 };
1791 down_write(&sdp->sd_log_flush_lock);
1792 error = gfs2_rbm_find(&rbm, GFS2_BLKST_UNLINKED, NULL, NULL,
1794 up_write(&sdp->sd_log_flush_lock);
1795 if (error == -ENOSPC)
1797 if (WARN_ON_ONCE(error))
1800 block = gfs2_rbm_to_block(&rbm);
1801 if (gfs2_rbm_from_block(&rbm, block + 1))
1803 if (*last_unlinked != NO_BLOCK && block <= *last_unlinked)
1807 *last_unlinked = block;
1809 error = gfs2_glock_get(sdp, block, &gfs2_iopen_glops, CREATE, &gl);
1813 /* If the inode is already in cache, we can ignore it here
1814 * because the existing inode disposal code will deal with
1815 * it when all refs have gone away. Accessing gl_object like
1816 * this is not safe in general. Here it is ok because we do
1817 * not dereference the pointer, and we only need an approx
1818 * answer to whether it is NULL or not.
1822 if (ip || queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
1827 /* Limit reclaim to sensible number of tasks */
1828 if (found > NR_CPUS)
1832 rgd->rd_flags &= ~GFS2_RDF_CHECK;
1837 * gfs2_rgrp_congested - Use stats to figure out whether an rgrp is congested
1838 * @rgd: The rgrp in question
1839 * @loops: An indication of how picky we can be (0=very, 1=less so)
1841 * This function uses the recently added glock statistics in order to
1842 * figure out whether a parciular resource group is suffering from
1843 * contention from multiple nodes. This is done purely on the basis
1844 * of timings, since this is the only data we have to work with and
1845 * our aim here is to reject a resource group which is highly contended
1846 * but (very important) not to do this too often in order to ensure that
1847 * we do not land up introducing fragmentation by changing resource
1848 * groups when not actually required.
1850 * The calculation is fairly simple, we want to know whether the SRTTB
1851 * (i.e. smoothed round trip time for blocking operations) to acquire
1852 * the lock for this rgrp's glock is significantly greater than the
1853 * time taken for resource groups on average. We introduce a margin in
1854 * the form of the variable @var which is computed as the sum of the two
1855 * respective variences, and multiplied by a factor depending on @loops
1856 * and whether we have a lot of data to base the decision on. This is
1857 * then tested against the square difference of the means in order to
1858 * decide whether the result is statistically significant or not.
1860 * Returns: A boolean verdict on the congestion status
1863 static bool gfs2_rgrp_congested(const struct gfs2_rgrpd *rgd, int loops)
1865 const struct gfs2_glock *gl = rgd->rd_gl;
1866 const struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1867 struct gfs2_lkstats *st;
1868 u64 r_dcount, l_dcount;
1869 u64 l_srttb, a_srttb = 0;
1873 int cpu, nonzero = 0;
1876 for_each_present_cpu(cpu) {
1877 st = &per_cpu_ptr(sdp->sd_lkstats, cpu)->lkstats[LM_TYPE_RGRP];
1878 if (st->stats[GFS2_LKS_SRTTB]) {
1879 a_srttb += st->stats[GFS2_LKS_SRTTB];
1883 st = &this_cpu_ptr(sdp->sd_lkstats)->lkstats[LM_TYPE_RGRP];
1885 do_div(a_srttb, nonzero);
1886 r_dcount = st->stats[GFS2_LKS_DCOUNT];
1887 var = st->stats[GFS2_LKS_SRTTVARB] +
1888 gl->gl_stats.stats[GFS2_LKS_SRTTVARB];
1891 l_srttb = gl->gl_stats.stats[GFS2_LKS_SRTTB];
1892 l_dcount = gl->gl_stats.stats[GFS2_LKS_DCOUNT];
1894 if ((l_dcount < 1) || (r_dcount < 1) || (a_srttb == 0))
1897 srttb_diff = a_srttb - l_srttb;
1898 sqr_diff = srttb_diff * srttb_diff;
1901 if (l_dcount < 8 || r_dcount < 8)
1906 return ((srttb_diff < 0) && (sqr_diff > var));
1910 * gfs2_rgrp_used_recently
1911 * @rs: The block reservation with the rgrp to test
1912 * @msecs: The time limit in milliseconds
1914 * Returns: True if the rgrp glock has been used within the time limit
1916 static bool gfs2_rgrp_used_recently(const struct gfs2_blkreserv *rs,
1921 tdiff = ktime_to_ns(ktime_sub(ktime_get_real(),
1922 rs->rs_rbm.rgd->rd_gl->gl_dstamp));
1924 return tdiff > (msecs * 1000 * 1000);
1927 static u32 gfs2_orlov_skip(const struct gfs2_inode *ip)
1929 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1932 get_random_bytes(&skip, sizeof(skip));
1933 return skip % sdp->sd_rgrps;
1936 static bool gfs2_select_rgrp(struct gfs2_rgrpd **pos, const struct gfs2_rgrpd *begin)
1938 struct gfs2_rgrpd *rgd = *pos;
1939 struct gfs2_sbd *sdp = rgd->rd_sbd;
1941 rgd = gfs2_rgrpd_get_next(rgd);
1943 rgd = gfs2_rgrpd_get_first(sdp);
1945 if (rgd != begin) /* If we didn't wrap */
1951 * fast_to_acquire - determine if a resource group will be fast to acquire
1953 * If this is one of our preferred rgrps, it should be quicker to acquire,
1954 * because we tried to set ourselves up as dlm lock master.
1956 static inline int fast_to_acquire(struct gfs2_rgrpd *rgd)
1958 struct gfs2_glock *gl = rgd->rd_gl;
1960 if (gl->gl_state != LM_ST_UNLOCKED && list_empty(&gl->gl_holders) &&
1961 !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
1962 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1964 if (rgd->rd_flags & GFS2_RDF_PREFERRED)
1970 * gfs2_inplace_reserve - Reserve space in the filesystem
1971 * @ip: the inode to reserve space for
1972 * @ap: the allocation parameters
1974 * We try our best to find an rgrp that has at least ap->target blocks
1975 * available. After a couple of passes (loops == 2), the prospects of finding
1976 * such an rgrp diminish. At this stage, we return the first rgrp that has
1977 * atleast ap->min_target blocks available. Either way, we set ap->allowed to
1978 * the number of blocks available in the chosen rgrp.
1980 * Returns: 0 on success,
1981 * -ENOMEM if a suitable rgrp can't be found
1985 int gfs2_inplace_reserve(struct gfs2_inode *ip, struct gfs2_alloc_parms *ap)
1987 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1988 struct gfs2_rgrpd *begin = NULL;
1989 struct gfs2_blkreserv *rs = &ip->i_res;
1990 int error = 0, rg_locked, flags = 0;
1991 u64 last_unlinked = NO_BLOCK;
1995 if (sdp->sd_args.ar_rgrplvb)
1997 if (gfs2_assert_warn(sdp, ap->target))
1999 if (gfs2_rs_active(rs)) {
2000 begin = rs->rs_rbm.rgd;
2001 } else if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, ip->i_goal)) {
2002 rs->rs_rbm.rgd = begin = ip->i_rgd;
2004 check_and_update_goal(ip);
2005 rs->rs_rbm.rgd = begin = gfs2_blk2rgrpd(sdp, ip->i_goal, 1);
2007 if (S_ISDIR(ip->i_inode.i_mode) && (ap->aflags & GFS2_AF_ORLOV))
2008 skip = gfs2_orlov_skip(ip);
2009 if (rs->rs_rbm.rgd == NULL)
2015 if (!gfs2_glock_is_locked_by_me(rs->rs_rbm.rgd->rd_gl)) {
2019 if (!gfs2_rs_active(rs)) {
2021 !fast_to_acquire(rs->rs_rbm.rgd))
2024 gfs2_rgrp_used_recently(rs, 1000) &&
2025 gfs2_rgrp_congested(rs->rs_rbm.rgd, loops))
2028 error = gfs2_glock_nq_init(rs->rs_rbm.rgd->rd_gl,
2029 LM_ST_EXCLUSIVE, flags,
2031 if (unlikely(error))
2033 if (!gfs2_rs_active(rs) && (loops < 2) &&
2034 gfs2_rgrp_congested(rs->rs_rbm.rgd, loops))
2036 if (sdp->sd_args.ar_rgrplvb) {
2037 error = update_rgrp_lvb(rs->rs_rbm.rgd);
2038 if (unlikely(error)) {
2039 gfs2_glock_dq_uninit(&rs->rs_rgd_gh);
2045 /* Skip unuseable resource groups */
2046 if ((rs->rs_rbm.rgd->rd_flags & (GFS2_RGF_NOALLOC |
2048 (loops == 0 && ap->target > rs->rs_rbm.rgd->rd_extfail_pt))
2051 if (sdp->sd_args.ar_rgrplvb)
2052 gfs2_rgrp_bh_get(rs->rs_rbm.rgd);
2054 /* Get a reservation if we don't already have one */
2055 if (!gfs2_rs_active(rs))
2056 rg_mblk_search(rs->rs_rbm.rgd, ip, ap);
2058 /* Skip rgrps when we can't get a reservation on first pass */
2059 if (!gfs2_rs_active(rs) && (loops < 1))
2062 /* If rgrp has enough free space, use it */
2063 if (rs->rs_rbm.rgd->rd_free_clone >= ap->target ||
2064 (loops == 2 && ap->min_target &&
2065 rs->rs_rbm.rgd->rd_free_clone >= ap->min_target)) {
2066 ip->i_rgd = rs->rs_rbm.rgd;
2067 ap->allowed = ip->i_rgd->rd_free_clone;
2071 /* Check for unlinked inodes which can be reclaimed */
2072 if (rs->rs_rbm.rgd->rd_flags & GFS2_RDF_CHECK)
2073 try_rgrp_unlink(rs->rs_rbm.rgd, &last_unlinked,
2076 /* Drop reservation, if we couldn't use reserved rgrp */
2077 if (gfs2_rs_active(rs))
2078 gfs2_rs_deltree(rs);
2080 /* Unlock rgrp if required */
2082 gfs2_glock_dq_uninit(&rs->rs_rgd_gh);
2084 /* Find the next rgrp, and continue looking */
2085 if (gfs2_select_rgrp(&rs->rs_rbm.rgd, begin))
2090 /* If we've scanned all the rgrps, but found no free blocks
2091 * then this checks for some less likely conditions before
2095 /* Check that fs hasn't grown if writing to rindex */
2096 if (ip == GFS2_I(sdp->sd_rindex) && !sdp->sd_rindex_uptodate) {
2097 error = gfs2_ri_update(ip);
2101 /* Flushing the log may release space */
2103 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
2110 * gfs2_inplace_release - release an inplace reservation
2111 * @ip: the inode the reservation was taken out on
2113 * Release a reservation made by gfs2_inplace_reserve().
2116 void gfs2_inplace_release(struct gfs2_inode *ip)
2118 struct gfs2_blkreserv *rs = &ip->i_res;
2120 if (gfs2_holder_initialized(&rs->rs_rgd_gh))
2121 gfs2_glock_dq_uninit(&rs->rs_rgd_gh);
2125 * gfs2_get_block_type - Check a block in a RG is of given type
2126 * @rgd: the resource group holding the block
2127 * @block: the block number
2129 * Returns: The block type (GFS2_BLKST_*)
2132 static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block)
2134 struct gfs2_rbm rbm = { .rgd = rgd, };
2137 ret = gfs2_rbm_from_block(&rbm, block);
2138 WARN_ON_ONCE(ret != 0);
2140 return gfs2_testbit(&rbm);
2145 * gfs2_alloc_extent - allocate an extent from a given bitmap
2146 * @rbm: the resource group information
2147 * @dinode: TRUE if the first block we allocate is for a dinode
2148 * @n: The extent length (value/result)
2150 * Add the bitmap buffer to the transaction.
2151 * Set the found bits to @new_state to change block's allocation state.
2153 static void gfs2_alloc_extent(const struct gfs2_rbm *rbm, bool dinode,
2156 struct gfs2_rbm pos = { .rgd = rbm->rgd, };
2157 const unsigned int elen = *n;
2162 block = gfs2_rbm_to_block(rbm);
2163 gfs2_trans_add_meta(rbm->rgd->rd_gl, rbm_bi(rbm)->bi_bh);
2164 gfs2_setbit(rbm, true, dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED);
2167 ret = gfs2_rbm_from_block(&pos, block);
2168 if (ret || gfs2_testbit(&pos) != GFS2_BLKST_FREE)
2170 gfs2_trans_add_meta(pos.rgd->rd_gl, rbm_bi(&pos)->bi_bh);
2171 gfs2_setbit(&pos, true, GFS2_BLKST_USED);
2178 * rgblk_free - Change alloc state of given block(s)
2179 * @sdp: the filesystem
2180 * @bstart: the start of a run of blocks to free
2181 * @blen: the length of the block run (all must lie within ONE RG!)
2182 * @new_state: GFS2_BLKST_XXX the after-allocation block state
2184 * Returns: Resource group containing the block(s)
2187 static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart,
2188 u32 blen, unsigned char new_state)
2190 struct gfs2_rbm rbm;
2191 struct gfs2_bitmap *bi, *bi_prev = NULL;
2193 rbm.rgd = gfs2_blk2rgrpd(sdp, bstart, 1);
2195 if (gfs2_consist(sdp))
2196 fs_err(sdp, "block = %llu\n", (unsigned long long)bstart);
2200 gfs2_rbm_from_block(&rbm, bstart);
2203 if (bi != bi_prev) {
2204 if (!bi->bi_clone) {
2205 bi->bi_clone = kmalloc(bi->bi_bh->b_size,
2206 GFP_NOFS | __GFP_NOFAIL);
2207 memcpy(bi->bi_clone + bi->bi_offset,
2208 bi->bi_bh->b_data + bi->bi_offset,
2211 gfs2_trans_add_meta(rbm.rgd->rd_gl, bi->bi_bh);
2214 gfs2_setbit(&rbm, false, new_state);
2215 gfs2_rbm_incr(&rbm);
2222 * gfs2_rgrp_dump - print out an rgrp
2223 * @seq: The iterator
2224 * @gl: The glock in question
2228 void gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl)
2230 struct gfs2_rgrpd *rgd = gl->gl_object;
2231 struct gfs2_blkreserv *trs;
2232 const struct rb_node *n;
2236 gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u r:%u e:%u\n",
2237 (unsigned long long)rgd->rd_addr, rgd->rd_flags,
2238 rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes,
2239 rgd->rd_reserved, rgd->rd_extfail_pt);
2240 spin_lock(&rgd->rd_rsspin);
2241 for (n = rb_first(&rgd->rd_rstree); n; n = rb_next(&trs->rs_node)) {
2242 trs = rb_entry(n, struct gfs2_blkreserv, rs_node);
2245 spin_unlock(&rgd->rd_rsspin);
2248 static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd)
2250 struct gfs2_sbd *sdp = rgd->rd_sbd;
2251 fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n",
2252 (unsigned long long)rgd->rd_addr);
2253 fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n");
2254 gfs2_rgrp_dump(NULL, rgd->rd_gl);
2255 rgd->rd_flags |= GFS2_RDF_ERROR;
2259 * gfs2_adjust_reservation - Adjust (or remove) a reservation after allocation
2260 * @ip: The inode we have just allocated blocks for
2261 * @rbm: The start of the allocated blocks
2262 * @len: The extent length
2264 * Adjusts a reservation after an allocation has taken place. If the
2265 * reservation does not match the allocation, or if it is now empty
2266 * then it is removed.
2269 static void gfs2_adjust_reservation(struct gfs2_inode *ip,
2270 const struct gfs2_rbm *rbm, unsigned len)
2272 struct gfs2_blkreserv *rs = &ip->i_res;
2273 struct gfs2_rgrpd *rgd = rbm->rgd;
2278 spin_lock(&rgd->rd_rsspin);
2279 if (gfs2_rs_active(rs)) {
2280 if (gfs2_rbm_eq(&rs->rs_rbm, rbm)) {
2281 block = gfs2_rbm_to_block(rbm);
2282 ret = gfs2_rbm_from_block(&rs->rs_rbm, block + len);
2283 rlen = min(rs->rs_free, len);
2284 rs->rs_free -= rlen;
2285 rgd->rd_reserved -= rlen;
2286 trace_gfs2_rs(rs, TRACE_RS_CLAIM);
2287 if (rs->rs_free && !ret)
2289 /* We used up our block reservation, so we should
2290 reserve more blocks next time. */
2291 atomic_add(RGRP_RSRV_ADDBLKS, &rs->rs_sizehint);
2296 spin_unlock(&rgd->rd_rsspin);
2300 * gfs2_set_alloc_start - Set starting point for block allocation
2301 * @rbm: The rbm which will be set to the required location
2302 * @ip: The gfs2 inode
2303 * @dinode: Flag to say if allocation includes a new inode
2305 * This sets the starting point from the reservation if one is active
2306 * otherwise it falls back to guessing a start point based on the
2307 * inode's goal block or the last allocation point in the rgrp.
2310 static void gfs2_set_alloc_start(struct gfs2_rbm *rbm,
2311 const struct gfs2_inode *ip, bool dinode)
2315 if (gfs2_rs_active(&ip->i_res)) {
2316 *rbm = ip->i_res.rs_rbm;
2320 if (!dinode && rgrp_contains_block(rbm->rgd, ip->i_goal))
2323 goal = rbm->rgd->rd_last_alloc + rbm->rgd->rd_data0;
2325 gfs2_rbm_from_block(rbm, goal);
2329 * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode
2330 * @ip: the inode to allocate the block for
2331 * @bn: Used to return the starting block number
2332 * @nblocks: requested number of blocks/extent length (value/result)
2333 * @dinode: 1 if we're allocating a dinode block, else 0
2334 * @generation: the generation number of the inode
2336 * Returns: 0 or error
2339 int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *nblocks,
2340 bool dinode, u64 *generation)
2342 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2343 struct buffer_head *dibh;
2344 struct gfs2_rbm rbm = { .rgd = ip->i_rgd, };
2346 u64 block; /* block, within the file system scope */
2349 gfs2_set_alloc_start(&rbm, ip, dinode);
2350 error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, NULL, ip, false);
2352 if (error == -ENOSPC) {
2353 gfs2_set_alloc_start(&rbm, ip, dinode);
2354 error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, NULL, NULL, false);
2357 /* Since all blocks are reserved in advance, this shouldn't happen */
2359 fs_warn(sdp, "inum=%llu error=%d, nblocks=%u, full=%d fail_pt=%d\n",
2360 (unsigned long long)ip->i_no_addr, error, *nblocks,
2361 test_bit(GBF_FULL, &rbm.rgd->rd_bits->bi_flags),
2362 rbm.rgd->rd_extfail_pt);
2366 gfs2_alloc_extent(&rbm, dinode, nblocks);
2367 block = gfs2_rbm_to_block(&rbm);
2368 rbm.rgd->rd_last_alloc = block - rbm.rgd->rd_data0;
2369 if (gfs2_rs_active(&ip->i_res))
2370 gfs2_adjust_reservation(ip, &rbm, *nblocks);
2376 ip->i_goal = block + ndata - 1;
2377 error = gfs2_meta_inode_buffer(ip, &dibh);
2379 struct gfs2_dinode *di =
2380 (struct gfs2_dinode *)dibh->b_data;
2381 gfs2_trans_add_meta(ip->i_gl, dibh);
2382 di->di_goal_meta = di->di_goal_data =
2383 cpu_to_be64(ip->i_goal);
2387 if (rbm.rgd->rd_free < *nblocks) {
2388 pr_warn("nblocks=%u\n", *nblocks);
2392 rbm.rgd->rd_free -= *nblocks;
2394 rbm.rgd->rd_dinodes++;
2395 *generation = rbm.rgd->rd_igeneration++;
2396 if (*generation == 0)
2397 *generation = rbm.rgd->rd_igeneration++;
2400 gfs2_trans_add_meta(rbm.rgd->rd_gl, rbm.rgd->rd_bits[0].bi_bh);
2401 gfs2_rgrp_out(rbm.rgd, rbm.rgd->rd_bits[0].bi_bh->b_data);
2402 gfs2_rgrp_ondisk2lvb(rbm.rgd->rd_rgl, rbm.rgd->rd_bits[0].bi_bh->b_data);
2404 gfs2_statfs_change(sdp, 0, -(s64)*nblocks, dinode ? 1 : 0);
2406 gfs2_trans_add_unrevoke(sdp, block, *nblocks);
2408 gfs2_quota_change(ip, *nblocks, ip->i_inode.i_uid, ip->i_inode.i_gid);
2410 rbm.rgd->rd_free_clone -= *nblocks;
2411 trace_gfs2_block_alloc(ip, rbm.rgd, block, *nblocks,
2412 dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED);
2417 gfs2_rgrp_error(rbm.rgd);
2422 * __gfs2_free_blocks - free a contiguous run of block(s)
2423 * @ip: the inode these blocks are being freed from
2424 * @bstart: first block of a run of contiguous blocks
2425 * @blen: the length of the block run
2426 * @meta: 1 if the blocks represent metadata
2430 void __gfs2_free_blocks(struct gfs2_inode *ip, u64 bstart, u32 blen, int meta)
2432 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2433 struct gfs2_rgrpd *rgd;
2435 rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE);
2438 trace_gfs2_block_alloc(ip, rgd, bstart, blen, GFS2_BLKST_FREE);
2439 rgd->rd_free += blen;
2440 rgd->rd_flags &= ~GFS2_RGF_TRIMMED;
2441 gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh);
2442 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
2443 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data);
2445 /* Directories keep their data in the metadata address space */
2446 if (meta || ip->i_depth)
2447 gfs2_meta_wipe(ip, bstart, blen);
2451 * gfs2_free_meta - free a contiguous run of data block(s)
2452 * @ip: the inode these blocks are being freed from
2453 * @bstart: first block of a run of contiguous blocks
2454 * @blen: the length of the block run
2458 void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen)
2460 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2462 __gfs2_free_blocks(ip, bstart, blen, 1);
2463 gfs2_statfs_change(sdp, 0, +blen, 0);
2464 gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid);
2467 void gfs2_unlink_di(struct inode *inode)
2469 struct gfs2_inode *ip = GFS2_I(inode);
2470 struct gfs2_sbd *sdp = GFS2_SB(inode);
2471 struct gfs2_rgrpd *rgd;
2472 u64 blkno = ip->i_no_addr;
2474 rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED);
2477 trace_gfs2_block_alloc(ip, rgd, blkno, 1, GFS2_BLKST_UNLINKED);
2478 gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh);
2479 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
2480 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data);
2481 update_rgrp_lvb_unlinked(rgd, 1);
2484 static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno)
2486 struct gfs2_sbd *sdp = rgd->rd_sbd;
2487 struct gfs2_rgrpd *tmp_rgd;
2489 tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE);
2492 gfs2_assert_withdraw(sdp, rgd == tmp_rgd);
2494 if (!rgd->rd_dinodes)
2495 gfs2_consist_rgrpd(rgd);
2499 gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh);
2500 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
2501 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data);
2502 update_rgrp_lvb_unlinked(rgd, -1);
2504 gfs2_statfs_change(sdp, 0, +1, -1);
2508 void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip)
2510 gfs2_free_uninit_di(rgd, ip->i_no_addr);
2511 trace_gfs2_block_alloc(ip, rgd, ip->i_no_addr, 1, GFS2_BLKST_FREE);
2512 gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid);
2513 gfs2_meta_wipe(ip, ip->i_no_addr, 1);
2517 * gfs2_check_blk_type - Check the type of a block
2518 * @sdp: The superblock
2519 * @no_addr: The block number to check
2520 * @type: The block type we are looking for
2522 * Returns: 0 if the block type matches the expected type
2523 * -ESTALE if it doesn't match
2524 * or -ve errno if something went wrong while checking
2527 int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type)
2529 struct gfs2_rgrpd *rgd;
2530 struct gfs2_holder rgd_gh;
2531 int error = -EINVAL;
2533 rgd = gfs2_blk2rgrpd(sdp, no_addr, 1);
2537 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh);
2541 if (gfs2_get_block_type(rgd, no_addr) != type)
2544 gfs2_glock_dq_uninit(&rgd_gh);
2550 * gfs2_rlist_add - add a RG to a list of RGs
2552 * @rlist: the list of resource groups
2555 * Figure out what RG a block belongs to and add that RG to the list
2557 * FIXME: Don't use NOFAIL
2561 void gfs2_rlist_add(struct gfs2_inode *ip, struct gfs2_rgrp_list *rlist,
2564 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2565 struct gfs2_rgrpd *rgd;
2566 struct gfs2_rgrpd **tmp;
2567 unsigned int new_space;
2570 if (gfs2_assert_warn(sdp, !rlist->rl_ghs))
2573 if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, block))
2576 rgd = gfs2_blk2rgrpd(sdp, block, 1);
2578 fs_err(sdp, "rlist_add: no rgrp for block %llu\n", (unsigned long long)block);
2583 for (x = 0; x < rlist->rl_rgrps; x++)
2584 if (rlist->rl_rgd[x] == rgd)
2587 if (rlist->rl_rgrps == rlist->rl_space) {
2588 new_space = rlist->rl_space + 10;
2590 tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *),
2591 GFP_NOFS | __GFP_NOFAIL);
2593 if (rlist->rl_rgd) {
2594 memcpy(tmp, rlist->rl_rgd,
2595 rlist->rl_space * sizeof(struct gfs2_rgrpd *));
2596 kfree(rlist->rl_rgd);
2599 rlist->rl_space = new_space;
2600 rlist->rl_rgd = tmp;
2603 rlist->rl_rgd[rlist->rl_rgrps++] = rgd;
2607 * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate
2608 * and initialize an array of glock holders for them
2609 * @rlist: the list of resource groups
2610 * @state: the lock state to acquire the RG lock in
2612 * FIXME: Don't use NOFAIL
2616 void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state)
2620 rlist->rl_ghs = kmalloc(rlist->rl_rgrps * sizeof(struct gfs2_holder),
2621 GFP_NOFS | __GFP_NOFAIL);
2622 for (x = 0; x < rlist->rl_rgrps; x++)
2623 gfs2_holder_init(rlist->rl_rgd[x]->rd_gl,
2629 * gfs2_rlist_free - free a resource group list
2630 * @rlist: the list of resource groups
2634 void gfs2_rlist_free(struct gfs2_rgrp_list *rlist)
2638 kfree(rlist->rl_rgd);
2640 if (rlist->rl_ghs) {
2641 for (x = 0; x < rlist->rl_rgrps; x++)
2642 gfs2_holder_uninit(&rlist->rl_ghs[x]);
2643 kfree(rlist->rl_ghs);
2644 rlist->rl_ghs = NULL;