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>
32 #include "trace_gfs2.h"
34 #define BFITNOENT ((u32)~0)
35 #define NO_BLOCK ((u64)~0)
37 #if BITS_PER_LONG == 32
38 #define LBITMASK (0x55555555UL)
39 #define LBITSKIP55 (0x55555555UL)
40 #define LBITSKIP00 (0x00000000UL)
42 #define LBITMASK (0x5555555555555555UL)
43 #define LBITSKIP55 (0x5555555555555555UL)
44 #define LBITSKIP00 (0x0000000000000000UL)
48 * These routines are used by the resource group routines (rgrp.c)
49 * to keep track of block allocation. Each block is represented by two
50 * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks.
53 * 1 = Used (not metadata)
54 * 2 = Unlinked (still in use) inode
58 static const char valid_change[16] = {
66 static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal,
67 unsigned char old_state, unsigned char new_state,
71 * gfs2_setbit - Set a bit in the bitmaps
72 * @buffer: the buffer that holds the bitmaps
73 * @buflen: the length (in bytes) of the buffer
74 * @block: the block to set
75 * @new_state: the new state of the block
79 static inline void gfs2_setbit(struct gfs2_rgrpd *rgd, unsigned char *buf1,
80 unsigned char *buf2, unsigned int offset,
81 unsigned int buflen, u32 block,
82 unsigned char new_state)
84 unsigned char *byte1, *byte2, *end, cur_state;
85 const unsigned int bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE;
87 byte1 = buf1 + offset + (block / GFS2_NBBY);
88 end = buf1 + offset + buflen;
92 cur_state = (*byte1 >> bit) & GFS2_BIT_MASK;
94 if (unlikely(!valid_change[new_state * 4 + cur_state])) {
95 gfs2_consist_rgrpd(rgd);
98 *byte1 ^= (cur_state ^ new_state) << bit;
101 byte2 = buf2 + offset + (block / GFS2_NBBY);
102 cur_state = (*byte2 >> bit) & GFS2_BIT_MASK;
103 *byte2 ^= (cur_state ^ new_state) << bit;
108 * gfs2_testbit - test a bit in the bitmaps
109 * @buffer: the buffer that holds the bitmaps
110 * @buflen: the length (in bytes) of the buffer
111 * @block: the block to read
115 static inline unsigned char gfs2_testbit(struct gfs2_rgrpd *rgd,
116 const unsigned char *buffer,
117 unsigned int buflen, u32 block)
119 const unsigned char *byte, *end;
120 unsigned char cur_state;
123 byte = buffer + (block / GFS2_NBBY);
124 bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE;
125 end = buffer + buflen;
127 gfs2_assert(rgd->rd_sbd, byte < end);
129 cur_state = (*byte >> bit) & GFS2_BIT_MASK;
136 * @ptr: Pointer to bitmap data
137 * @mask: Mask to use (normally 0x55555.... but adjusted for search start)
138 * @state: The state we are searching for
140 * We xor the bitmap data with a patter which is the bitwise opposite
141 * of what we are looking for, this gives rise to a pattern of ones
142 * wherever there is a match. Since we have two bits per entry, we
143 * take this pattern, shift it down by one place and then and it with
144 * the original. All the even bit positions (0,2,4, etc) then represent
145 * successful matches, so we mask with 0x55555..... to remove the unwanted
148 * This allows searching of a whole u64 at once (32 blocks) with a
149 * single test (on 64 bit arches).
152 static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state)
155 static const u64 search[] = {
156 [0] = 0xffffffffffffffffULL,
157 [1] = 0xaaaaaaaaaaaaaaaaULL,
158 [2] = 0x5555555555555555ULL,
159 [3] = 0x0000000000000000ULL,
161 tmp = le64_to_cpu(*ptr) ^ search[state];
168 * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing
169 * a block in a given allocation state.
170 * @buffer: the buffer that holds the bitmaps
171 * @len: the length (in bytes) of the buffer
172 * @goal: start search at this block's bit-pair (within @buffer)
173 * @state: GFS2_BLKST_XXX the state of the block we're looking for.
175 * Scope of @goal and returned block number is only within this bitmap buffer,
176 * not entire rgrp or filesystem. @buffer will be offset from the actual
177 * beginning of a bitmap block buffer, skipping any header structures, but
178 * headers are always a multiple of 64 bits long so that the buffer is
179 * always aligned to a 64 bit boundary.
181 * The size of the buffer is in bytes, but is it assumed that it is
182 * always ok to read a complete multiple of 64 bits at the end
183 * of the block in case the end is no aligned to a natural boundary.
185 * Return: the block number (bitmap buffer scope) that was found
188 static u32 gfs2_bitfit(const u8 *buf, const unsigned int len,
191 u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1);
192 const __le64 *ptr = ((__le64 *)buf) + (goal >> 5);
193 const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64)));
195 u64 mask = 0x5555555555555555ULL;
200 /* Mask off bits we don't care about at the start of the search */
202 tmp = gfs2_bit_search(ptr, mask, state);
204 while(tmp == 0 && ptr < end) {
205 tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state);
208 /* Mask off any bits which are more than len bytes from the start */
209 if (ptr == end && (len & (sizeof(u64) - 1)))
210 tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1))));
211 /* Didn't find anything, so return */
216 bit /= 2; /* two bits per entry in the bitmap */
217 return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit;
221 * gfs2_bitcount - count the number of bits in a certain state
222 * @buffer: the buffer that holds the bitmaps
223 * @buflen: the length (in bytes) of the buffer
224 * @state: the state of the block we're looking for
226 * Returns: The number of bits
229 static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer,
230 unsigned int buflen, u8 state)
232 const u8 *byte = buffer;
233 const u8 *end = buffer + buflen;
234 const u8 state1 = state << 2;
235 const u8 state2 = state << 4;
236 const u8 state3 = state << 6;
239 for (; byte < end; byte++) {
240 if (((*byte) & 0x03) == state)
242 if (((*byte) & 0x0C) == state1)
244 if (((*byte) & 0x30) == state2)
246 if (((*byte) & 0xC0) == state3)
254 * gfs2_rgrp_verify - Verify that a resource group is consistent
255 * @sdp: the filesystem
260 void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd)
262 struct gfs2_sbd *sdp = rgd->rd_sbd;
263 struct gfs2_bitmap *bi = NULL;
264 u32 length = rgd->rd_length;
268 memset(count, 0, 4 * sizeof(u32));
270 /* Count # blocks in each of 4 possible allocation states */
271 for (buf = 0; buf < length; buf++) {
272 bi = rgd->rd_bits + buf;
273 for (x = 0; x < 4; x++)
274 count[x] += gfs2_bitcount(rgd,
280 if (count[0] != rgd->rd_free) {
281 if (gfs2_consist_rgrpd(rgd))
282 fs_err(sdp, "free data mismatch: %u != %u\n",
283 count[0], rgd->rd_free);
287 tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes;
288 if (count[1] != tmp) {
289 if (gfs2_consist_rgrpd(rgd))
290 fs_err(sdp, "used data mismatch: %u != %u\n",
295 if (count[2] + count[3] != rgd->rd_dinodes) {
296 if (gfs2_consist_rgrpd(rgd))
297 fs_err(sdp, "used metadata mismatch: %u != %u\n",
298 count[2] + count[3], rgd->rd_dinodes);
303 static inline int rgrp_contains_block(struct gfs2_rgrpd *rgd, u64 block)
305 u64 first = rgd->rd_data0;
306 u64 last = first + rgd->rd_data;
307 return first <= block && block < last;
311 * gfs2_blk2rgrpd - Find resource group for a given data/meta block number
312 * @sdp: The GFS2 superblock
313 * @n: The data block number
315 * Returns: The resource group, or NULL if not found
318 struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk)
320 struct gfs2_rgrpd *rgd;
322 spin_lock(&sdp->sd_rindex_spin);
324 list_for_each_entry(rgd, &sdp->sd_rindex_mru_list, rd_list_mru) {
325 if (rgrp_contains_block(rgd, blk)) {
326 list_move(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list);
327 spin_unlock(&sdp->sd_rindex_spin);
332 spin_unlock(&sdp->sd_rindex_spin);
338 * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem
339 * @sdp: The GFS2 superblock
341 * Returns: The first rgrp in the filesystem
344 struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp)
346 gfs2_assert(sdp, !list_empty(&sdp->sd_rindex_list));
347 return list_entry(sdp->sd_rindex_list.next, struct gfs2_rgrpd, rd_list);
351 * gfs2_rgrpd_get_next - get the next RG
354 * Returns: The next rgrp
357 struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd)
359 if (rgd->rd_list.next == &rgd->rd_sbd->sd_rindex_list)
361 return list_entry(rgd->rd_list.next, struct gfs2_rgrpd, rd_list);
364 static void clear_rgrpdi(struct gfs2_sbd *sdp)
366 struct list_head *head;
367 struct gfs2_rgrpd *rgd;
368 struct gfs2_glock *gl;
370 spin_lock(&sdp->sd_rindex_spin);
371 sdp->sd_rindex_forward = NULL;
372 spin_unlock(&sdp->sd_rindex_spin);
374 head = &sdp->sd_rindex_list;
375 while (!list_empty(head)) {
376 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_list);
379 list_del(&rgd->rd_list);
380 list_del(&rgd->rd_list_mru);
383 gl->gl_object = NULL;
388 kmem_cache_free(gfs2_rgrpd_cachep, rgd);
392 void gfs2_clear_rgrpd(struct gfs2_sbd *sdp)
394 mutex_lock(&sdp->sd_rindex_mutex);
396 mutex_unlock(&sdp->sd_rindex_mutex);
399 static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd)
401 printk(KERN_INFO " ri_addr = %llu\n", (unsigned long long)rgd->rd_addr);
402 printk(KERN_INFO " ri_length = %u\n", rgd->rd_length);
403 printk(KERN_INFO " ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0);
404 printk(KERN_INFO " ri_data = %u\n", rgd->rd_data);
405 printk(KERN_INFO " ri_bitbytes = %u\n", rgd->rd_bitbytes);
409 * gfs2_compute_bitstructs - Compute the bitmap sizes
410 * @rgd: The resource group descriptor
412 * Calculates bitmap descriptors, one for each block that contains bitmap data
417 static int compute_bitstructs(struct gfs2_rgrpd *rgd)
419 struct gfs2_sbd *sdp = rgd->rd_sbd;
420 struct gfs2_bitmap *bi;
421 u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */
422 u32 bytes_left, bytes;
428 rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS);
432 bytes_left = rgd->rd_bitbytes;
434 for (x = 0; x < length; x++) {
435 bi = rgd->rd_bits + x;
438 /* small rgrp; bitmap stored completely in header block */
441 bi->bi_offset = sizeof(struct gfs2_rgrp);
446 bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp);
447 bi->bi_offset = sizeof(struct gfs2_rgrp);
451 } else if (x + 1 == length) {
453 bi->bi_offset = sizeof(struct gfs2_meta_header);
454 bi->bi_start = rgd->rd_bitbytes - bytes_left;
458 bytes = sdp->sd_sb.sb_bsize -
459 sizeof(struct gfs2_meta_header);
460 bi->bi_offset = sizeof(struct gfs2_meta_header);
461 bi->bi_start = rgd->rd_bitbytes - bytes_left;
469 gfs2_consist_rgrpd(rgd);
472 bi = rgd->rd_bits + (length - 1);
473 if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) {
474 if (gfs2_consist_rgrpd(rgd)) {
475 gfs2_rindex_print(rgd);
476 fs_err(sdp, "start=%u len=%u offset=%u\n",
477 bi->bi_start, bi->bi_len, bi->bi_offset);
486 * gfs2_ri_total - Total up the file system space, according to the rindex.
489 u64 gfs2_ri_total(struct gfs2_sbd *sdp)
492 struct inode *inode = sdp->sd_rindex;
493 struct gfs2_inode *ip = GFS2_I(inode);
494 char buf[sizeof(struct gfs2_rindex)];
495 struct file_ra_state ra_state;
498 mutex_lock(&sdp->sd_rindex_mutex);
499 file_ra_state_init(&ra_state, inode->i_mapping);
500 for (rgrps = 0;; rgrps++) {
501 loff_t pos = rgrps * sizeof(struct gfs2_rindex);
503 if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode))
505 error = gfs2_internal_read(ip, &ra_state, buf, &pos,
506 sizeof(struct gfs2_rindex));
507 if (error != sizeof(struct gfs2_rindex))
509 total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data);
511 mutex_unlock(&sdp->sd_rindex_mutex);
515 static void gfs2_rindex_in(struct gfs2_rgrpd *rgd, const void *buf)
517 const struct gfs2_rindex *str = buf;
519 rgd->rd_addr = be64_to_cpu(str->ri_addr);
520 rgd->rd_length = be32_to_cpu(str->ri_length);
521 rgd->rd_data0 = be64_to_cpu(str->ri_data0);
522 rgd->rd_data = be32_to_cpu(str->ri_data);
523 rgd->rd_bitbytes = be32_to_cpu(str->ri_bitbytes);
527 * read_rindex_entry - Pull in a new resource index entry from the disk
528 * @gl: The glock covering the rindex inode
530 * Returns: 0 on success, error code otherwise
533 static int read_rindex_entry(struct gfs2_inode *ip,
534 struct file_ra_state *ra_state)
536 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
537 loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex);
538 char buf[sizeof(struct gfs2_rindex)];
540 struct gfs2_rgrpd *rgd;
542 error = gfs2_internal_read(ip, ra_state, buf, &pos,
543 sizeof(struct gfs2_rindex));
546 if (error != sizeof(struct gfs2_rindex)) {
552 rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS);
557 mutex_init(&rgd->rd_mutex);
558 lops_init_le(&rgd->rd_le, &gfs2_rg_lops);
561 list_add_tail(&rgd->rd_list, &sdp->sd_rindex_list);
562 list_add_tail(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list);
564 gfs2_rindex_in(rgd, buf);
565 error = compute_bitstructs(rgd);
569 error = gfs2_glock_get(sdp, rgd->rd_addr,
570 &gfs2_rgrp_glops, CREATE, &rgd->rd_gl);
574 rgd->rd_gl->gl_object = rgd;
575 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
580 * gfs2_ri_update - Pull in a new resource index from the disk
581 * @ip: pointer to the rindex inode
583 * Returns: 0 on successful update, error code otherwise
586 int gfs2_ri_update(struct gfs2_inode *ip)
588 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
589 struct inode *inode = &ip->i_inode;
590 struct file_ra_state ra_state;
591 u64 rgrp_count = i_size_read(inode);
592 struct gfs2_rgrpd *rgd;
593 unsigned int max_data = 0;
596 do_div(rgrp_count, sizeof(struct gfs2_rindex));
599 file_ra_state_init(&ra_state, inode->i_mapping);
600 for (sdp->sd_rgrps = 0; sdp->sd_rgrps < rgrp_count; sdp->sd_rgrps++) {
601 error = read_rindex_entry(ip, &ra_state);
608 list_for_each_entry(rgd, &sdp->sd_rindex_list, rd_list)
609 if (rgd->rd_data > max_data)
610 max_data = rgd->rd_data;
611 sdp->sd_max_rg_data = max_data;
612 sdp->sd_rindex_uptodate = 1;
617 * gfs2_rindex_hold - Grab a lock on the rindex
618 * @sdp: The GFS2 superblock
619 * @ri_gh: the glock holder
621 * We grab a lock on the rindex inode to make sure that it doesn't
622 * change whilst we are performing an operation. We keep this lock
623 * for quite long periods of time compared to other locks. This
624 * doesn't matter, since it is shared and it is very, very rarely
625 * accessed in the exclusive mode (i.e. only when expanding the filesystem).
627 * This makes sure that we're using the latest copy of the resource index
628 * special file, which might have been updated if someone expanded the
629 * filesystem (via gfs2_grow utility), which adds new resource groups.
631 * Returns: 0 on success, error code otherwise
634 int gfs2_rindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ri_gh)
636 struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex);
637 struct gfs2_glock *gl = ip->i_gl;
640 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, ri_gh);
644 /* Read new copy from disk if we don't have the latest */
645 if (!sdp->sd_rindex_uptodate) {
646 mutex_lock(&sdp->sd_rindex_mutex);
647 if (!sdp->sd_rindex_uptodate) {
648 error = gfs2_ri_update(ip);
650 gfs2_glock_dq_uninit(ri_gh);
652 mutex_unlock(&sdp->sd_rindex_mutex);
658 static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf)
660 const struct gfs2_rgrp *str = buf;
663 rg_flags = be32_to_cpu(str->rg_flags);
664 rg_flags &= ~GFS2_RDF_MASK;
665 rgd->rd_flags &= GFS2_RDF_MASK;
666 rgd->rd_flags |= rg_flags;
667 rgd->rd_free = be32_to_cpu(str->rg_free);
668 rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes);
669 rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration);
672 static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf)
674 struct gfs2_rgrp *str = buf;
676 str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK);
677 str->rg_free = cpu_to_be32(rgd->rd_free);
678 str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes);
679 str->__pad = cpu_to_be32(0);
680 str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration);
681 memset(&str->rg_reserved, 0, sizeof(str->rg_reserved));
685 * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps
686 * @rgd: the struct gfs2_rgrpd describing the RG to read in
688 * Read in all of a Resource Group's header and bitmap blocks.
689 * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps.
694 int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd)
696 struct gfs2_sbd *sdp = rgd->rd_sbd;
697 struct gfs2_glock *gl = rgd->rd_gl;
698 unsigned int length = rgd->rd_length;
699 struct gfs2_bitmap *bi;
703 mutex_lock(&rgd->rd_mutex);
705 spin_lock(&sdp->sd_rindex_spin);
706 if (rgd->rd_bh_count) {
708 spin_unlock(&sdp->sd_rindex_spin);
709 mutex_unlock(&rgd->rd_mutex);
712 spin_unlock(&sdp->sd_rindex_spin);
714 for (x = 0; x < length; x++) {
715 bi = rgd->rd_bits + x;
716 error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, &bi->bi_bh);
721 for (y = length; y--;) {
722 bi = rgd->rd_bits + y;
723 error = gfs2_meta_wait(sdp, bi->bi_bh);
726 if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB :
733 if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) {
734 for (x = 0; x < length; x++)
735 clear_bit(GBF_FULL, &rgd->rd_bits[x].bi_flags);
736 gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data);
737 rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK);
740 spin_lock(&sdp->sd_rindex_spin);
741 rgd->rd_free_clone = rgd->rd_free;
743 spin_unlock(&sdp->sd_rindex_spin);
745 mutex_unlock(&rgd->rd_mutex);
751 bi = rgd->rd_bits + x;
754 gfs2_assert_warn(sdp, !bi->bi_clone);
756 mutex_unlock(&rgd->rd_mutex);
761 void gfs2_rgrp_bh_hold(struct gfs2_rgrpd *rgd)
763 struct gfs2_sbd *sdp = rgd->rd_sbd;
765 spin_lock(&sdp->sd_rindex_spin);
766 gfs2_assert_warn(rgd->rd_sbd, rgd->rd_bh_count);
768 spin_unlock(&sdp->sd_rindex_spin);
772 * gfs2_rgrp_bh_put - Release RG bitmaps read in with gfs2_rgrp_bh_get()
773 * @rgd: the struct gfs2_rgrpd describing the RG to read in
777 void gfs2_rgrp_bh_put(struct gfs2_rgrpd *rgd)
779 struct gfs2_sbd *sdp = rgd->rd_sbd;
780 int x, length = rgd->rd_length;
782 spin_lock(&sdp->sd_rindex_spin);
783 gfs2_assert_warn(rgd->rd_sbd, rgd->rd_bh_count);
784 if (--rgd->rd_bh_count) {
785 spin_unlock(&sdp->sd_rindex_spin);
789 for (x = 0; x < length; x++) {
790 struct gfs2_bitmap *bi = rgd->rd_bits + x;
797 spin_unlock(&sdp->sd_rindex_spin);
800 static void gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset,
801 const struct gfs2_bitmap *bi)
803 struct super_block *sb = sdp->sd_vfs;
804 struct block_device *bdev = sb->s_bdev;
805 const unsigned int sects_per_blk = sdp->sd_sb.sb_bsize /
806 bdev_logical_block_size(sb->s_bdev);
809 sector_t nr_sects = 0;
813 for (x = 0; x < bi->bi_len; x++) {
814 const u8 *orig = bi->bi_bh->b_data + bi->bi_offset + x;
815 const u8 *clone = bi->bi_clone + bi->bi_offset + x;
816 u8 diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1));
820 blk = offset + ((bi->bi_start + x) * GFS2_NBBY);
821 blk *= sects_per_blk; /* convert to sectors */
825 goto start_new_extent;
826 if ((start + nr_sects) != blk) {
827 rv = blkdev_issue_discard(bdev, start,
836 nr_sects += sects_per_blk;
839 blk += sects_per_blk;
843 rv = blkdev_issue_discard(bdev, start, nr_sects, GFP_NOFS, 0);
849 fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv);
850 sdp->sd_args.ar_discard = 0;
853 void gfs2_rgrp_repolish_clones(struct gfs2_rgrpd *rgd)
855 struct gfs2_sbd *sdp = rgd->rd_sbd;
856 unsigned int length = rgd->rd_length;
859 for (x = 0; x < length; x++) {
860 struct gfs2_bitmap *bi = rgd->rd_bits + x;
863 if (sdp->sd_args.ar_discard)
864 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bi);
865 clear_bit(GBF_FULL, &bi->bi_flags);
866 memcpy(bi->bi_clone + bi->bi_offset,
867 bi->bi_bh->b_data + bi->bi_offset, bi->bi_len);
870 spin_lock(&sdp->sd_rindex_spin);
871 rgd->rd_free_clone = rgd->rd_free;
872 spin_unlock(&sdp->sd_rindex_spin);
876 * gfs2_alloc_get - get the struct gfs2_alloc structure for an inode
877 * @ip: the incore GFS2 inode structure
879 * Returns: the struct gfs2_alloc
882 struct gfs2_alloc *gfs2_alloc_get(struct gfs2_inode *ip)
884 BUG_ON(ip->i_alloc != NULL);
885 ip->i_alloc = kzalloc(sizeof(struct gfs2_alloc), GFP_NOFS);
890 * try_rgrp_fit - See if a given reservation will fit in a given RG
892 * @al: the struct gfs2_alloc structure describing the reservation
894 * If there's room for the requested blocks to be allocated from the RG:
895 * Sets the $al_rgd field in @al.
897 * Returns: 1 on success (it fits), 0 on failure (it doesn't fit)
900 static int try_rgrp_fit(struct gfs2_rgrpd *rgd, struct gfs2_alloc *al)
902 struct gfs2_sbd *sdp = rgd->rd_sbd;
905 if (rgd->rd_flags & (GFS2_RGF_NOALLOC | GFS2_RDF_ERROR))
908 spin_lock(&sdp->sd_rindex_spin);
909 if (rgd->rd_free_clone >= al->al_requested) {
913 spin_unlock(&sdp->sd_rindex_spin);
919 * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes
922 * Returns: 0 if no error
923 * The inode, if one has been found, in inode.
926 static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip)
930 struct gfs2_sbd *sdp = rgd->rd_sbd;
932 struct gfs2_glock *gl;
933 struct gfs2_inode *ip;
937 while (goal < rgd->rd_data) {
938 down_write(&sdp->sd_log_flush_lock);
940 block = rgblk_search(rgd, goal, GFS2_BLKST_UNLINKED,
941 GFS2_BLKST_UNLINKED, &n);
942 up_write(&sdp->sd_log_flush_lock);
943 if (block == BFITNOENT)
945 /* rgblk_search can return a block < goal, so we need to
946 keep it marching forward. */
947 no_addr = block + rgd->rd_data0;
949 if (*last_unlinked != NO_BLOCK && no_addr <= *last_unlinked)
953 *last_unlinked = no_addr;
955 error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE, &gl);
959 /* If the inode is already in cache, we can ignore it here
960 * because the existing inode disposal code will deal with
961 * it when all refs have gone away. Accessing gl_object like
962 * this is not safe in general. Here it is ok because we do
963 * not dereference the pointer, and we only need an approx
964 * answer to whether it is NULL or not.
968 if (ip || queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
973 /* Limit reclaim to sensible number of tasks */
974 if (found > 2*NR_CPUS)
978 rgd->rd_flags &= ~GFS2_RDF_CHECK;
983 * recent_rgrp_next - get next RG from "recent" list
984 * @cur_rgd: current rgrp
986 * Returns: The next rgrp in the recent list
989 static struct gfs2_rgrpd *recent_rgrp_next(struct gfs2_rgrpd *cur_rgd)
991 struct gfs2_sbd *sdp = cur_rgd->rd_sbd;
992 struct list_head *head;
993 struct gfs2_rgrpd *rgd;
995 spin_lock(&sdp->sd_rindex_spin);
996 head = &sdp->sd_rindex_mru_list;
997 if (unlikely(cur_rgd->rd_list_mru.next == head)) {
998 spin_unlock(&sdp->sd_rindex_spin);
1001 rgd = list_entry(cur_rgd->rd_list_mru.next, struct gfs2_rgrpd, rd_list_mru);
1002 spin_unlock(&sdp->sd_rindex_spin);
1007 * forward_rgrp_get - get an rgrp to try next from full list
1008 * @sdp: The GFS2 superblock
1010 * Returns: The rgrp to try next
1013 static struct gfs2_rgrpd *forward_rgrp_get(struct gfs2_sbd *sdp)
1015 struct gfs2_rgrpd *rgd;
1016 unsigned int journals = gfs2_jindex_size(sdp);
1017 unsigned int rg = 0, x;
1019 spin_lock(&sdp->sd_rindex_spin);
1021 rgd = sdp->sd_rindex_forward;
1023 if (sdp->sd_rgrps >= journals)
1024 rg = sdp->sd_rgrps * sdp->sd_jdesc->jd_jid / journals;
1026 for (x = 0, rgd = gfs2_rgrpd_get_first(sdp); x < rg;
1027 x++, rgd = gfs2_rgrpd_get_next(rgd))
1030 sdp->sd_rindex_forward = rgd;
1033 spin_unlock(&sdp->sd_rindex_spin);
1039 * forward_rgrp_set - set the forward rgrp pointer
1040 * @sdp: the filesystem
1041 * @rgd: The new forward rgrp
1045 static void forward_rgrp_set(struct gfs2_sbd *sdp, struct gfs2_rgrpd *rgd)
1047 spin_lock(&sdp->sd_rindex_spin);
1048 sdp->sd_rindex_forward = rgd;
1049 spin_unlock(&sdp->sd_rindex_spin);
1053 * get_local_rgrp - Choose and lock a rgrp for allocation
1054 * @ip: the inode to reserve space for
1055 * @rgp: the chosen and locked rgrp
1057 * Try to acquire rgrp in way which avoids contending with others.
1062 static int get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked)
1064 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1065 struct gfs2_rgrpd *rgd, *begin = NULL;
1066 struct gfs2_alloc *al = ip->i_alloc;
1067 int flags = LM_FLAG_TRY;
1070 int error, rg_locked;
1072 rgd = gfs2_blk2rgrpd(sdp, ip->i_goal);
1077 if (gfs2_glock_is_locked_by_me(rgd->rd_gl)) {
1081 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1082 LM_FLAG_TRY, &al->al_rgd_gh);
1086 if (try_rgrp_fit(rgd, al))
1088 if (rgd->rd_flags & GFS2_RDF_CHECK)
1089 try_rgrp_unlink(rgd, last_unlinked, ip->i_no_addr);
1091 gfs2_glock_dq_uninit(&al->al_rgd_gh);
1094 rgd = recent_rgrp_next(rgd);
1102 /* Go through full list of rgrps */
1104 begin = rgd = forward_rgrp_get(sdp);
1109 if (gfs2_glock_is_locked_by_me(rgd->rd_gl)) {
1113 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, flags,
1118 if (try_rgrp_fit(rgd, al))
1120 if (rgd->rd_flags & GFS2_RDF_CHECK)
1121 try_rgrp_unlink(rgd, last_unlinked, ip->i_no_addr);
1123 gfs2_glock_dq_uninit(&al->al_rgd_gh);
1134 rgd = gfs2_rgrpd_get_next(rgd);
1136 rgd = gfs2_rgrpd_get_first(sdp);
1145 gfs2_log_flush(sdp, NULL);
1151 spin_lock(&sdp->sd_rindex_spin);
1152 list_move(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list);
1153 spin_unlock(&sdp->sd_rindex_spin);
1154 rgd = gfs2_rgrpd_get_next(rgd);
1156 rgd = gfs2_rgrpd_get_first(sdp);
1157 forward_rgrp_set(sdp, rgd);
1164 * gfs2_inplace_reserve_i - Reserve space in the filesystem
1165 * @ip: the inode to reserve space for
1170 int gfs2_inplace_reserve_i(struct gfs2_inode *ip, int hold_rindex,
1171 char *file, unsigned int line)
1173 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1174 struct gfs2_alloc *al = ip->i_alloc;
1176 u64 last_unlinked = NO_BLOCK;
1179 if (gfs2_assert_warn(sdp, al->al_requested))
1183 /* We need to hold the rindex unless the inode we're using is
1184 the rindex itself, in which case it's already held. */
1185 if (ip != GFS2_I(sdp->sd_rindex))
1186 error = gfs2_rindex_hold(sdp, &al->al_ri_gh);
1187 else if (!sdp->sd_rgrps) /* We may not have the rindex read
1189 error = gfs2_ri_update(ip);
1196 error = get_local_rgrp(ip, &last_unlinked);
1197 /* If there is no space, flushing the log may release some */
1199 if (ip == GFS2_I(sdp->sd_rindex) &&
1200 !sdp->sd_rindex_uptodate) {
1201 error = gfs2_ri_update(ip);
1206 gfs2_log_flush(sdp, NULL);
1208 } while (error && tries++ < 3);
1211 if (hold_rindex && ip != GFS2_I(sdp->sd_rindex))
1212 gfs2_glock_dq_uninit(&al->al_ri_gh);
1216 /* no error, so we have the rgrp set in the inode's allocation. */
1224 * gfs2_inplace_release - release an inplace reservation
1225 * @ip: the inode the reservation was taken out on
1227 * Release a reservation made by gfs2_inplace_reserve().
1230 void gfs2_inplace_release(struct gfs2_inode *ip)
1232 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1233 struct gfs2_alloc *al = ip->i_alloc;
1235 if (gfs2_assert_warn(sdp, al->al_alloced <= al->al_requested) == -1)
1236 fs_warn(sdp, "al_alloced = %u, al_requested = %u "
1237 "al_file = %s, al_line = %u\n",
1238 al->al_alloced, al->al_requested, al->al_file,
1242 if (al->al_rgd_gh.gh_gl)
1243 gfs2_glock_dq_uninit(&al->al_rgd_gh);
1244 if (ip != GFS2_I(sdp->sd_rindex) && al->al_ri_gh.gh_gl)
1245 gfs2_glock_dq_uninit(&al->al_ri_gh);
1249 * gfs2_get_block_type - Check a block in a RG is of given type
1250 * @rgd: the resource group holding the block
1251 * @block: the block number
1253 * Returns: The block type (GFS2_BLKST_*)
1256 static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block)
1258 struct gfs2_bitmap *bi = NULL;
1259 u32 length, rgrp_block, buf_block;
1263 length = rgd->rd_length;
1264 rgrp_block = block - rgd->rd_data0;
1266 for (buf = 0; buf < length; buf++) {
1267 bi = rgd->rd_bits + buf;
1268 if (rgrp_block < (bi->bi_start + bi->bi_len) * GFS2_NBBY)
1272 gfs2_assert(rgd->rd_sbd, buf < length);
1273 buf_block = rgrp_block - bi->bi_start * GFS2_NBBY;
1275 type = gfs2_testbit(rgd, bi->bi_bh->b_data + bi->bi_offset,
1276 bi->bi_len, buf_block);
1282 * rgblk_search - find a block in @old_state, change allocation
1283 * state to @new_state
1284 * @rgd: the resource group descriptor
1285 * @goal: the goal block within the RG (start here to search for avail block)
1286 * @old_state: GFS2_BLKST_XXX the before-allocation state to find
1287 * @new_state: GFS2_BLKST_XXX the after-allocation block state
1288 * @n: The extent length
1290 * Walk rgrp's bitmap to find bits that represent a block in @old_state.
1291 * Add the found bitmap buffer to the transaction.
1292 * Set the found bits to @new_state to change block's allocation state.
1294 * This function never fails, because we wouldn't call it unless we
1295 * know (from reservation results, etc.) that a block is available.
1297 * Scope of @goal and returned block is just within rgrp, not the whole
1300 * Returns: the block number allocated
1303 static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal,
1304 unsigned char old_state, unsigned char new_state,
1307 struct gfs2_bitmap *bi = NULL;
1308 const u32 length = rgd->rd_length;
1309 u32 blk = BFITNOENT;
1310 unsigned int buf, x;
1311 const unsigned int elen = *n;
1312 const u8 *buffer = NULL;
1315 /* Find bitmap block that contains bits for goal block */
1316 for (buf = 0; buf < length; buf++) {
1317 bi = rgd->rd_bits + buf;
1318 /* Convert scope of "goal" from rgrp-wide to within found bit block */
1319 if (goal < (bi->bi_start + bi->bi_len) * GFS2_NBBY) {
1320 goal -= bi->bi_start * GFS2_NBBY;
1328 /* Search (up to entire) bitmap in this rgrp for allocatable block.
1329 "x <= length", instead of "x < length", because we typically start
1330 the search in the middle of a bit block, but if we can't find an
1331 allocatable block anywhere else, we want to be able wrap around and
1332 search in the first part of our first-searched bit block. */
1333 for (x = 0; x <= length; x++) {
1334 bi = rgd->rd_bits + buf;
1336 if (test_bit(GBF_FULL, &bi->bi_flags) &&
1337 (old_state == GFS2_BLKST_FREE))
1340 /* The GFS2_BLKST_UNLINKED state doesn't apply to the clone
1341 bitmaps, so we must search the originals for that. */
1342 buffer = bi->bi_bh->b_data + bi->bi_offset;
1343 if (old_state != GFS2_BLKST_UNLINKED && bi->bi_clone)
1344 buffer = bi->bi_clone + bi->bi_offset;
1346 blk = gfs2_bitfit(buffer, bi->bi_len, goal, old_state);
1347 if (blk != BFITNOENT)
1350 if ((goal == 0) && (old_state == GFS2_BLKST_FREE))
1351 set_bit(GBF_FULL, &bi->bi_flags);
1353 /* Try next bitmap block (wrap back to rgrp header if at end) */
1360 if (blk == BFITNOENT)
1363 if (old_state == new_state)
1366 gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1);
1367 gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, bi->bi_offset,
1368 bi->bi_len, blk, new_state);
1372 if (goal >= (bi->bi_len * GFS2_NBBY))
1374 if (gfs2_testbit(rgd, buffer, bi->bi_len, goal) !=
1377 gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, bi->bi_offset,
1378 bi->bi_len, goal, new_state);
1382 return (bi->bi_start * GFS2_NBBY) + blk;
1386 * rgblk_free - Change alloc state of given block(s)
1387 * @sdp: the filesystem
1388 * @bstart: the start of a run of blocks to free
1389 * @blen: the length of the block run (all must lie within ONE RG!)
1390 * @new_state: GFS2_BLKST_XXX the after-allocation block state
1392 * Returns: Resource group containing the block(s)
1395 static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart,
1396 u32 blen, unsigned char new_state)
1398 struct gfs2_rgrpd *rgd;
1399 struct gfs2_bitmap *bi = NULL;
1400 u32 length, rgrp_blk, buf_blk;
1403 rgd = gfs2_blk2rgrpd(sdp, bstart);
1405 if (gfs2_consist(sdp))
1406 fs_err(sdp, "block = %llu\n", (unsigned long long)bstart);
1410 length = rgd->rd_length;
1412 rgrp_blk = bstart - rgd->rd_data0;
1415 for (buf = 0; buf < length; buf++) {
1416 bi = rgd->rd_bits + buf;
1417 if (rgrp_blk < (bi->bi_start + bi->bi_len) * GFS2_NBBY)
1421 gfs2_assert(rgd->rd_sbd, buf < length);
1423 buf_blk = rgrp_blk - bi->bi_start * GFS2_NBBY;
1426 if (!bi->bi_clone) {
1427 bi->bi_clone = kmalloc(bi->bi_bh->b_size,
1428 GFP_NOFS | __GFP_NOFAIL);
1429 memcpy(bi->bi_clone + bi->bi_offset,
1430 bi->bi_bh->b_data + bi->bi_offset,
1433 gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1);
1434 gfs2_setbit(rgd, bi->bi_bh->b_data, NULL, bi->bi_offset,
1435 bi->bi_len, buf_blk, new_state);
1442 * gfs2_rgrp_dump - print out an rgrp
1443 * @seq: The iterator
1444 * @gl: The glock in question
1448 int gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl)
1450 const struct gfs2_rgrpd *rgd = gl->gl_object;
1453 gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u\n",
1454 (unsigned long long)rgd->rd_addr, rgd->rd_flags,
1455 rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes);
1459 static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd)
1461 struct gfs2_sbd *sdp = rgd->rd_sbd;
1462 fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n",
1463 (unsigned long long)rgd->rd_addr);
1464 fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n");
1465 gfs2_rgrp_dump(NULL, rgd->rd_gl);
1466 rgd->rd_flags |= GFS2_RDF_ERROR;
1470 * gfs2_alloc_block - Allocate one or more blocks
1471 * @ip: the inode to allocate the block for
1472 * @bn: Used to return the starting block number
1473 * @n: requested number of blocks/extent length (value/result)
1475 * Returns: 0 or error
1478 int gfs2_alloc_block(struct gfs2_inode *ip, u64 *bn, unsigned int *n)
1480 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1481 struct buffer_head *dibh;
1482 struct gfs2_alloc *al = ip->i_alloc;
1483 struct gfs2_rgrpd *rgd;
1488 /* Only happens if there is a bug in gfs2, return something distinctive
1489 * to ensure that it is noticed.
1496 if (rgrp_contains_block(rgd, ip->i_goal))
1497 goal = ip->i_goal - rgd->rd_data0;
1499 goal = rgd->rd_last_alloc;
1501 blk = rgblk_search(rgd, goal, GFS2_BLKST_FREE, GFS2_BLKST_USED, n);
1503 /* Since all blocks are reserved in advance, this shouldn't happen */
1504 if (blk == BFITNOENT)
1507 rgd->rd_last_alloc = blk;
1508 block = rgd->rd_data0 + blk;
1510 error = gfs2_meta_inode_buffer(ip, &dibh);
1512 struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
1513 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1514 di->di_goal_meta = di->di_goal_data = cpu_to_be64(ip->i_goal);
1517 if (rgd->rd_free < *n)
1522 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1523 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1525 al->al_alloced += *n;
1527 gfs2_statfs_change(sdp, 0, -(s64)*n, 0);
1528 gfs2_quota_change(ip, *n, ip->i_inode.i_uid, ip->i_inode.i_gid);
1530 spin_lock(&sdp->sd_rindex_spin);
1531 rgd->rd_free_clone -= *n;
1532 spin_unlock(&sdp->sd_rindex_spin);
1533 trace_gfs2_block_alloc(ip, block, *n, GFS2_BLKST_USED);
1538 gfs2_rgrp_error(rgd);
1543 * gfs2_alloc_di - Allocate a dinode
1544 * @dip: the directory that the inode is going in
1545 * @bn: the block number which is allocated
1546 * @generation: the generation number of the inode
1548 * Returns: 0 on success or error
1551 int gfs2_alloc_di(struct gfs2_inode *dip, u64 *bn, u64 *generation)
1553 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1554 struct gfs2_alloc *al = dip->i_alloc;
1555 struct gfs2_rgrpd *rgd = al->al_rgd;
1560 blk = rgblk_search(rgd, rgd->rd_last_alloc,
1561 GFS2_BLKST_FREE, GFS2_BLKST_DINODE, &n);
1563 /* Since all blocks are reserved in advance, this shouldn't happen */
1564 if (blk == BFITNOENT)
1567 rgd->rd_last_alloc = blk;
1568 block = rgd->rd_data0 + blk;
1569 if (rgd->rd_free == 0)
1574 *generation = rgd->rd_igeneration++;
1575 if (*generation == 0)
1576 *generation = rgd->rd_igeneration++;
1577 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1578 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1582 gfs2_statfs_change(sdp, 0, -1, +1);
1583 gfs2_trans_add_unrevoke(sdp, block, 1);
1585 spin_lock(&sdp->sd_rindex_spin);
1586 rgd->rd_free_clone--;
1587 spin_unlock(&sdp->sd_rindex_spin);
1588 trace_gfs2_block_alloc(dip, block, 1, GFS2_BLKST_DINODE);
1593 gfs2_rgrp_error(rgd);
1598 * gfs2_free_data - free a contiguous run of data block(s)
1599 * @ip: the inode these blocks are being freed from
1600 * @bstart: first block of a run of contiguous blocks
1601 * @blen: the length of the block run
1605 void gfs2_free_data(struct gfs2_inode *ip, u64 bstart, u32 blen)
1607 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1608 struct gfs2_rgrpd *rgd;
1610 rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE);
1613 trace_gfs2_block_alloc(ip, bstart, blen, GFS2_BLKST_FREE);
1614 rgd->rd_free += blen;
1616 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1617 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1619 gfs2_trans_add_rg(rgd);
1621 gfs2_statfs_change(sdp, 0, +blen, 0);
1622 gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid);
1626 * gfs2_free_meta - free a contiguous run of data block(s)
1627 * @ip: the inode these blocks are being freed from
1628 * @bstart: first block of a run of contiguous blocks
1629 * @blen: the length of the block run
1633 void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen)
1635 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1636 struct gfs2_rgrpd *rgd;
1638 rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE);
1641 trace_gfs2_block_alloc(ip, bstart, blen, GFS2_BLKST_FREE);
1642 rgd->rd_free += blen;
1644 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1645 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1647 gfs2_trans_add_rg(rgd);
1649 gfs2_statfs_change(sdp, 0, +blen, 0);
1650 gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid);
1651 gfs2_meta_wipe(ip, bstart, blen);
1654 void gfs2_unlink_di(struct inode *inode)
1656 struct gfs2_inode *ip = GFS2_I(inode);
1657 struct gfs2_sbd *sdp = GFS2_SB(inode);
1658 struct gfs2_rgrpd *rgd;
1659 u64 blkno = ip->i_no_addr;
1661 rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED);
1664 trace_gfs2_block_alloc(ip, blkno, 1, GFS2_BLKST_UNLINKED);
1665 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1666 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1667 gfs2_trans_add_rg(rgd);
1670 static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno)
1672 struct gfs2_sbd *sdp = rgd->rd_sbd;
1673 struct gfs2_rgrpd *tmp_rgd;
1675 tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE);
1678 gfs2_assert_withdraw(sdp, rgd == tmp_rgd);
1680 if (!rgd->rd_dinodes)
1681 gfs2_consist_rgrpd(rgd);
1685 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1686 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1688 gfs2_statfs_change(sdp, 0, +1, -1);
1689 gfs2_trans_add_rg(rgd);
1693 void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip)
1695 gfs2_free_uninit_di(rgd, ip->i_no_addr);
1696 trace_gfs2_block_alloc(ip, ip->i_no_addr, 1, GFS2_BLKST_FREE);
1697 gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid);
1698 gfs2_meta_wipe(ip, ip->i_no_addr, 1);
1702 * gfs2_check_blk_type - Check the type of a block
1703 * @sdp: The superblock
1704 * @no_addr: The block number to check
1705 * @type: The block type we are looking for
1707 * Returns: 0 if the block type matches the expected type
1708 * -ESTALE if it doesn't match
1709 * or -ve errno if something went wrong while checking
1712 int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type)
1714 struct gfs2_rgrpd *rgd;
1715 struct gfs2_holder ri_gh, rgd_gh;
1716 struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex);
1720 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
1721 error = gfs2_rindex_hold(sdp, &ri_gh);
1728 rgd = gfs2_blk2rgrpd(sdp, no_addr);
1732 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh);
1736 if (gfs2_get_block_type(rgd, no_addr) != type)
1739 gfs2_glock_dq_uninit(&rgd_gh);
1742 gfs2_glock_dq_uninit(&ri_gh);
1748 * gfs2_rlist_add - add a RG to a list of RGs
1749 * @sdp: the filesystem
1750 * @rlist: the list of resource groups
1753 * Figure out what RG a block belongs to and add that RG to the list
1755 * FIXME: Don't use NOFAIL
1759 void gfs2_rlist_add(struct gfs2_sbd *sdp, struct gfs2_rgrp_list *rlist,
1762 struct gfs2_rgrpd *rgd;
1763 struct gfs2_rgrpd **tmp;
1764 unsigned int new_space;
1767 if (gfs2_assert_warn(sdp, !rlist->rl_ghs))
1770 rgd = gfs2_blk2rgrpd(sdp, block);
1772 if (gfs2_consist(sdp))
1773 fs_err(sdp, "block = %llu\n", (unsigned long long)block);
1777 for (x = 0; x < rlist->rl_rgrps; x++)
1778 if (rlist->rl_rgd[x] == rgd)
1781 if (rlist->rl_rgrps == rlist->rl_space) {
1782 new_space = rlist->rl_space + 10;
1784 tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *),
1785 GFP_NOFS | __GFP_NOFAIL);
1787 if (rlist->rl_rgd) {
1788 memcpy(tmp, rlist->rl_rgd,
1789 rlist->rl_space * sizeof(struct gfs2_rgrpd *));
1790 kfree(rlist->rl_rgd);
1793 rlist->rl_space = new_space;
1794 rlist->rl_rgd = tmp;
1797 rlist->rl_rgd[rlist->rl_rgrps++] = rgd;
1801 * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate
1802 * and initialize an array of glock holders for them
1803 * @rlist: the list of resource groups
1804 * @state: the lock state to acquire the RG lock in
1805 * @flags: the modifier flags for the holder structures
1807 * FIXME: Don't use NOFAIL
1811 void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state)
1815 rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder),
1816 GFP_NOFS | __GFP_NOFAIL);
1817 for (x = 0; x < rlist->rl_rgrps; x++)
1818 gfs2_holder_init(rlist->rl_rgd[x]->rd_gl,
1824 * gfs2_rlist_free - free a resource group list
1825 * @list: the list of resource groups
1829 void gfs2_rlist_free(struct gfs2_rgrp_list *rlist)
1833 kfree(rlist->rl_rgd);
1835 if (rlist->rl_ghs) {
1836 for (x = 0; x < rlist->rl_rgrps; x++)
1837 gfs2_holder_uninit(&rlist->rl_ghs[x]);
1838 kfree(rlist->rl_ghs);