2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2007 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/lm_interface.h>
32 #define BFITNOENT ((u32)~0)
35 * These routines are used by the resource group routines (rgrp.c)
36 * to keep track of block allocation. Each block is represented by two
37 * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks.
40 * 1 = Used (not metadata)
41 * 2 = Unlinked (still in use) inode
45 static const char valid_change[16] = {
54 * gfs2_setbit - Set a bit in the bitmaps
55 * @buffer: the buffer that holds the bitmaps
56 * @buflen: the length (in bytes) of the buffer
57 * @block: the block to set
58 * @new_state: the new state of the block
62 static void gfs2_setbit(struct gfs2_rgrpd *rgd, unsigned char *buffer,
63 unsigned int buflen, u32 block,
64 unsigned char new_state)
66 unsigned char *byte, *end, cur_state;
69 byte = buffer + (block / GFS2_NBBY);
70 bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE;
71 end = buffer + buflen;
73 gfs2_assert(rgd->rd_sbd, byte < end);
75 cur_state = (*byte >> bit) & GFS2_BIT_MASK;
77 if (valid_change[new_state * 4 + cur_state]) {
78 *byte ^= cur_state << bit;
79 *byte |= new_state << bit;
81 gfs2_consist_rgrpd(rgd);
85 * gfs2_testbit - test a bit in the bitmaps
86 * @buffer: the buffer that holds the bitmaps
87 * @buflen: the length (in bytes) of the buffer
88 * @block: the block to read
92 static unsigned char gfs2_testbit(struct gfs2_rgrpd *rgd, unsigned char *buffer,
93 unsigned int buflen, u32 block)
95 unsigned char *byte, *end, cur_state;
98 byte = buffer + (block / GFS2_NBBY);
99 bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE;
100 end = buffer + buflen;
102 gfs2_assert(rgd->rd_sbd, byte < end);
104 cur_state = (*byte >> bit) & GFS2_BIT_MASK;
110 * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing
111 * a block in a given allocation state.
112 * @buffer: the buffer that holds the bitmaps
113 * @buflen: the length (in bytes) of the buffer
114 * @goal: start search at this block's bit-pair (within @buffer)
115 * @old_state: GFS2_BLKST_XXX the state of the block we're looking for;
116 * bit 0 = alloc(1)/free(0), bit 1 = meta(1)/data(0)
118 * Scope of @goal and returned block number is only within this bitmap buffer,
119 * not entire rgrp or filesystem. @buffer will be offset from the actual
120 * beginning of a bitmap block buffer, skipping any header structures.
122 * Return: the block number (bitmap buffer scope) that was found
125 static u32 gfs2_bitfit(struct gfs2_rgrpd *rgd, unsigned char *buffer,
126 unsigned int buflen, u32 goal,
127 unsigned char old_state)
129 unsigned char *byte, *end, alloc;
133 byte = buffer + (goal / GFS2_NBBY);
134 bit = (goal % GFS2_NBBY) * GFS2_BIT_SIZE;
135 end = buffer + buflen;
136 alloc = (old_state & 1) ? 0 : 0x55;
139 if ((*byte & 0x55) == alloc) {
140 blk += (8 - bit) >> 1;
148 if (((*byte >> bit) & GFS2_BIT_MASK) == old_state)
151 bit += GFS2_BIT_SIZE;
164 * gfs2_bitcount - count the number of bits in a certain state
165 * @buffer: the buffer that holds the bitmaps
166 * @buflen: the length (in bytes) of the buffer
167 * @state: the state of the block we're looking for
169 * Returns: The number of bits
172 static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, unsigned char *buffer,
173 unsigned int buflen, unsigned char state)
175 unsigned char *byte = buffer;
176 unsigned char *end = buffer + buflen;
177 unsigned char state1 = state << 2;
178 unsigned char state2 = state << 4;
179 unsigned char state3 = state << 6;
182 for (; byte < end; byte++) {
183 if (((*byte) & 0x03) == state)
185 if (((*byte) & 0x0C) == state1)
187 if (((*byte) & 0x30) == state2)
189 if (((*byte) & 0xC0) == state3)
197 * gfs2_rgrp_verify - Verify that a resource group is consistent
198 * @sdp: the filesystem
203 void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd)
205 struct gfs2_sbd *sdp = rgd->rd_sbd;
206 struct gfs2_bitmap *bi = NULL;
207 u32 length = rgd->rd_ri.ri_length;
211 memset(count, 0, 4 * sizeof(u32));
213 /* Count # blocks in each of 4 possible allocation states */
214 for (buf = 0; buf < length; buf++) {
215 bi = rgd->rd_bits + buf;
216 for (x = 0; x < 4; x++)
217 count[x] += gfs2_bitcount(rgd,
223 if (count[0] != rgd->rd_rg.rg_free) {
224 if (gfs2_consist_rgrpd(rgd))
225 fs_err(sdp, "free data mismatch: %u != %u\n",
226 count[0], rgd->rd_rg.rg_free);
230 tmp = rgd->rd_ri.ri_data -
232 rgd->rd_rg.rg_dinodes;
233 if (count[1] + count[2] != tmp) {
234 if (gfs2_consist_rgrpd(rgd))
235 fs_err(sdp, "used data mismatch: %u != %u\n",
240 if (count[3] != rgd->rd_rg.rg_dinodes) {
241 if (gfs2_consist_rgrpd(rgd))
242 fs_err(sdp, "used metadata mismatch: %u != %u\n",
243 count[3], rgd->rd_rg.rg_dinodes);
247 if (count[2] > count[3]) {
248 if (gfs2_consist_rgrpd(rgd))
249 fs_err(sdp, "unlinked inodes > inodes: %u\n",
256 static inline int rgrp_contains_block(struct gfs2_rindex_host *ri, u64 block)
258 u64 first = ri->ri_data0;
259 u64 last = first + ri->ri_data;
260 return first <= block && block < last;
264 * gfs2_blk2rgrpd - Find resource group for a given data/meta block number
265 * @sdp: The GFS2 superblock
266 * @n: The data block number
268 * Returns: The resource group, or NULL if not found
271 struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk)
273 struct gfs2_rgrpd *rgd;
275 spin_lock(&sdp->sd_rindex_spin);
277 list_for_each_entry(rgd, &sdp->sd_rindex_mru_list, rd_list_mru) {
278 if (rgrp_contains_block(&rgd->rd_ri, blk)) {
279 list_move(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list);
280 spin_unlock(&sdp->sd_rindex_spin);
285 spin_unlock(&sdp->sd_rindex_spin);
291 * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem
292 * @sdp: The GFS2 superblock
294 * Returns: The first rgrp in the filesystem
297 struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp)
299 gfs2_assert(sdp, !list_empty(&sdp->sd_rindex_list));
300 return list_entry(sdp->sd_rindex_list.next, struct gfs2_rgrpd, rd_list);
304 * gfs2_rgrpd_get_next - get the next RG
307 * Returns: The next rgrp
310 struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd)
312 if (rgd->rd_list.next == &rgd->rd_sbd->sd_rindex_list)
314 return list_entry(rgd->rd_list.next, struct gfs2_rgrpd, rd_list);
317 static void clear_rgrpdi(struct gfs2_sbd *sdp)
319 struct list_head *head;
320 struct gfs2_rgrpd *rgd;
321 struct gfs2_glock *gl;
323 spin_lock(&sdp->sd_rindex_spin);
324 sdp->sd_rindex_forward = NULL;
325 head = &sdp->sd_rindex_recent_list;
326 while (!list_empty(head)) {
327 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_recent);
328 list_del(&rgd->rd_recent);
330 spin_unlock(&sdp->sd_rindex_spin);
332 head = &sdp->sd_rindex_list;
333 while (!list_empty(head)) {
334 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_list);
337 list_del(&rgd->rd_list);
338 list_del(&rgd->rd_list_mru);
341 gl->gl_object = NULL;
350 void gfs2_clear_rgrpd(struct gfs2_sbd *sdp)
352 mutex_lock(&sdp->sd_rindex_mutex);
354 mutex_unlock(&sdp->sd_rindex_mutex);
358 * gfs2_compute_bitstructs - Compute the bitmap sizes
359 * @rgd: The resource group descriptor
361 * Calculates bitmap descriptors, one for each block that contains bitmap data
366 static int compute_bitstructs(struct gfs2_rgrpd *rgd)
368 struct gfs2_sbd *sdp = rgd->rd_sbd;
369 struct gfs2_bitmap *bi;
370 u32 length = rgd->rd_ri.ri_length; /* # blocks in hdr & bitmap */
371 u32 bytes_left, bytes;
377 rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS);
381 bytes_left = rgd->rd_ri.ri_bitbytes;
383 for (x = 0; x < length; x++) {
384 bi = rgd->rd_bits + x;
386 /* small rgrp; bitmap stored completely in header block */
389 bi->bi_offset = sizeof(struct gfs2_rgrp);
394 bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp);
395 bi->bi_offset = sizeof(struct gfs2_rgrp);
399 } else if (x + 1 == length) {
401 bi->bi_offset = sizeof(struct gfs2_meta_header);
402 bi->bi_start = rgd->rd_ri.ri_bitbytes - bytes_left;
406 bytes = sdp->sd_sb.sb_bsize -
407 sizeof(struct gfs2_meta_header);
408 bi->bi_offset = sizeof(struct gfs2_meta_header);
409 bi->bi_start = rgd->rd_ri.ri_bitbytes - bytes_left;
417 gfs2_consist_rgrpd(rgd);
420 bi = rgd->rd_bits + (length - 1);
421 if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_ri.ri_data) {
422 if (gfs2_consist_rgrpd(rgd)) {
423 gfs2_rindex_print(&rgd->rd_ri);
424 fs_err(sdp, "start=%u len=%u offset=%u\n",
425 bi->bi_start, bi->bi_len, bi->bi_offset);
434 * gfs2_ri_total - Total up the file system space, according to the rindex.
437 u64 gfs2_ri_total(struct gfs2_sbd *sdp)
440 struct inode *inode = sdp->sd_rindex;
441 struct gfs2_inode *ip = GFS2_I(inode);
442 struct gfs2_rindex_host ri;
443 char buf[sizeof(struct gfs2_rindex)];
444 struct file_ra_state ra_state;
447 mutex_lock(&sdp->sd_rindex_mutex);
448 file_ra_state_init(&ra_state, inode->i_mapping);
449 for (rgrps = 0;; rgrps++) {
450 loff_t pos = rgrps * sizeof(struct gfs2_rindex);
452 if (pos + sizeof(struct gfs2_rindex) >= ip->i_di.di_size)
454 error = gfs2_internal_read(ip, &ra_state, buf, &pos,
455 sizeof(struct gfs2_rindex));
456 if (error != sizeof(struct gfs2_rindex))
458 gfs2_rindex_in(&ri, buf);
459 total_data += ri.ri_data;
461 mutex_unlock(&sdp->sd_rindex_mutex);
466 * read_rindex_entry - Pull in a new resource index entry from the disk
467 * @gl: The glock covering the rindex inode
469 * Returns: 0 on success, error code otherwise
472 static int read_rindex_entry(struct gfs2_inode *ip,
473 struct file_ra_state *ra_state)
475 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
476 loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex);
477 char buf[sizeof(struct gfs2_rindex)];
479 struct gfs2_rgrpd *rgd;
481 error = gfs2_internal_read(ip, ra_state, buf, &pos,
482 sizeof(struct gfs2_rindex));
485 if (error != sizeof(struct gfs2_rindex)) {
491 rgd = kzalloc(sizeof(struct gfs2_rgrpd), GFP_NOFS);
496 mutex_init(&rgd->rd_mutex);
497 lops_init_le(&rgd->rd_le, &gfs2_rg_lops);
500 list_add_tail(&rgd->rd_list, &sdp->sd_rindex_list);
501 list_add_tail(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list);
503 gfs2_rindex_in(&rgd->rd_ri, buf);
504 error = compute_bitstructs(rgd);
508 error = gfs2_glock_get(sdp, rgd->rd_ri.ri_addr,
509 &gfs2_rgrp_glops, CREATE, &rgd->rd_gl);
513 rgd->rd_gl->gl_object = rgd;
514 rgd->rd_rg_vn = rgd->rd_gl->gl_vn - 1;
519 * gfs2_ri_update - Pull in a new resource index from the disk
520 * @ip: pointer to the rindex inode
522 * Returns: 0 on successful update, error code otherwise
525 static int gfs2_ri_update(struct gfs2_inode *ip)
527 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
528 struct inode *inode = &ip->i_inode;
529 struct file_ra_state ra_state;
530 u64 rgrp_count = ip->i_di.di_size;
533 if (do_div(rgrp_count, sizeof(struct gfs2_rindex))) {
534 gfs2_consist_inode(ip);
540 file_ra_state_init(&ra_state, inode->i_mapping);
541 for (sdp->sd_rgrps = 0; sdp->sd_rgrps < rgrp_count; sdp->sd_rgrps++) {
542 error = read_rindex_entry(ip, &ra_state);
549 sdp->sd_rindex_vn = ip->i_gl->gl_vn;
554 * gfs2_ri_update_special - Pull in a new resource index from the disk
556 * This is a special version that's safe to call from gfs2_inplace_reserve_i.
557 * In this case we know that we don't have any resource groups in memory yet.
559 * @ip: pointer to the rindex inode
561 * Returns: 0 on successful update, error code otherwise
563 static int gfs2_ri_update_special(struct gfs2_inode *ip)
565 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
566 struct inode *inode = &ip->i_inode;
567 struct file_ra_state ra_state;
570 file_ra_state_init(&ra_state, inode->i_mapping);
571 for (sdp->sd_rgrps = 0;; sdp->sd_rgrps++) {
572 /* Ignore partials */
573 if ((sdp->sd_rgrps + 1) * sizeof(struct gfs2_rindex) >
576 error = read_rindex_entry(ip, &ra_state);
583 sdp->sd_rindex_vn = ip->i_gl->gl_vn;
588 * gfs2_rindex_hold - Grab a lock on the rindex
589 * @sdp: The GFS2 superblock
590 * @ri_gh: the glock holder
592 * We grab a lock on the rindex inode to make sure that it doesn't
593 * change whilst we are performing an operation. We keep this lock
594 * for quite long periods of time compared to other locks. This
595 * doesn't matter, since it is shared and it is very, very rarely
596 * accessed in the exclusive mode (i.e. only when expanding the filesystem).
598 * This makes sure that we're using the latest copy of the resource index
599 * special file, which might have been updated if someone expanded the
600 * filesystem (via gfs2_grow utility), which adds new resource groups.
602 * Returns: 0 on success, error code otherwise
605 int gfs2_rindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ri_gh)
607 struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex);
608 struct gfs2_glock *gl = ip->i_gl;
611 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, ri_gh);
615 /* Read new copy from disk if we don't have the latest */
616 if (sdp->sd_rindex_vn != gl->gl_vn) {
617 mutex_lock(&sdp->sd_rindex_mutex);
618 if (sdp->sd_rindex_vn != gl->gl_vn) {
619 error = gfs2_ri_update(ip);
621 gfs2_glock_dq_uninit(ri_gh);
623 mutex_unlock(&sdp->sd_rindex_mutex);
630 * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps
631 * @rgd: the struct gfs2_rgrpd describing the RG to read in
633 * Read in all of a Resource Group's header and bitmap blocks.
634 * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps.
639 int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd)
641 struct gfs2_sbd *sdp = rgd->rd_sbd;
642 struct gfs2_glock *gl = rgd->rd_gl;
643 unsigned int length = rgd->rd_ri.ri_length;
644 struct gfs2_bitmap *bi;
648 mutex_lock(&rgd->rd_mutex);
650 spin_lock(&sdp->sd_rindex_spin);
651 if (rgd->rd_bh_count) {
653 spin_unlock(&sdp->sd_rindex_spin);
654 mutex_unlock(&rgd->rd_mutex);
657 spin_unlock(&sdp->sd_rindex_spin);
659 for (x = 0; x < length; x++) {
660 bi = rgd->rd_bits + x;
661 error = gfs2_meta_read(gl, rgd->rd_ri.ri_addr + x, 0, &bi->bi_bh);
666 for (y = length; y--;) {
667 bi = rgd->rd_bits + y;
668 error = gfs2_meta_wait(sdp, bi->bi_bh);
671 if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB :
678 if (rgd->rd_rg_vn != gl->gl_vn) {
679 gfs2_rgrp_in(&rgd->rd_rg, (rgd->rd_bits[0].bi_bh)->b_data);
680 rgd->rd_rg_vn = gl->gl_vn;
683 spin_lock(&sdp->sd_rindex_spin);
684 rgd->rd_free_clone = rgd->rd_rg.rg_free;
686 spin_unlock(&sdp->sd_rindex_spin);
688 mutex_unlock(&rgd->rd_mutex);
694 bi = rgd->rd_bits + x;
697 gfs2_assert_warn(sdp, !bi->bi_clone);
699 mutex_unlock(&rgd->rd_mutex);
704 void gfs2_rgrp_bh_hold(struct gfs2_rgrpd *rgd)
706 struct gfs2_sbd *sdp = rgd->rd_sbd;
708 spin_lock(&sdp->sd_rindex_spin);
709 gfs2_assert_warn(rgd->rd_sbd, rgd->rd_bh_count);
711 spin_unlock(&sdp->sd_rindex_spin);
715 * gfs2_rgrp_bh_put - Release RG bitmaps read in with gfs2_rgrp_bh_get()
716 * @rgd: the struct gfs2_rgrpd describing the RG to read in
720 void gfs2_rgrp_bh_put(struct gfs2_rgrpd *rgd)
722 struct gfs2_sbd *sdp = rgd->rd_sbd;
723 int x, length = rgd->rd_ri.ri_length;
725 spin_lock(&sdp->sd_rindex_spin);
726 gfs2_assert_warn(rgd->rd_sbd, rgd->rd_bh_count);
727 if (--rgd->rd_bh_count) {
728 spin_unlock(&sdp->sd_rindex_spin);
732 for (x = 0; x < length; x++) {
733 struct gfs2_bitmap *bi = rgd->rd_bits + x;
740 spin_unlock(&sdp->sd_rindex_spin);
743 void gfs2_rgrp_repolish_clones(struct gfs2_rgrpd *rgd)
745 struct gfs2_sbd *sdp = rgd->rd_sbd;
746 unsigned int length = rgd->rd_ri.ri_length;
749 for (x = 0; x < length; x++) {
750 struct gfs2_bitmap *bi = rgd->rd_bits + x;
753 memcpy(bi->bi_clone + bi->bi_offset,
754 bi->bi_bh->b_data + bi->bi_offset, bi->bi_len);
757 spin_lock(&sdp->sd_rindex_spin);
758 rgd->rd_free_clone = rgd->rd_rg.rg_free;
759 spin_unlock(&sdp->sd_rindex_spin);
763 * gfs2_alloc_get - get the struct gfs2_alloc structure for an inode
764 * @ip: the incore GFS2 inode structure
766 * Returns: the struct gfs2_alloc
769 struct gfs2_alloc *gfs2_alloc_get(struct gfs2_inode *ip)
771 struct gfs2_alloc *al = &ip->i_alloc;
773 /* FIXME: Should assert that the correct locks are held here... */
774 memset(al, 0, sizeof(*al));
779 * try_rgrp_fit - See if a given reservation will fit in a given RG
781 * @al: the struct gfs2_alloc structure describing the reservation
783 * If there's room for the requested blocks to be allocated from the RG:
784 * Sets the $al_rgd field in @al.
786 * Returns: 1 on success (it fits), 0 on failure (it doesn't fit)
789 static int try_rgrp_fit(struct gfs2_rgrpd *rgd, struct gfs2_alloc *al)
791 struct gfs2_sbd *sdp = rgd->rd_sbd;
794 if (rgd->rd_rg.rg_flags & GFS2_RGF_NOALLOC)
797 spin_lock(&sdp->sd_rindex_spin);
798 if (rgd->rd_free_clone >= al->al_requested) {
802 spin_unlock(&sdp->sd_rindex_spin);
808 * recent_rgrp_first - get first RG from "recent" list
809 * @sdp: The GFS2 superblock
810 * @rglast: address of the rgrp used last
812 * Returns: The first rgrp in the recent list
815 static struct gfs2_rgrpd *recent_rgrp_first(struct gfs2_sbd *sdp,
818 struct gfs2_rgrpd *rgd = NULL;
820 spin_lock(&sdp->sd_rindex_spin);
822 if (list_empty(&sdp->sd_rindex_recent_list))
828 list_for_each_entry(rgd, &sdp->sd_rindex_recent_list, rd_recent) {
829 if (rgd->rd_ri.ri_addr == rglast)
834 rgd = list_entry(sdp->sd_rindex_recent_list.next, struct gfs2_rgrpd,
837 spin_unlock(&sdp->sd_rindex_spin);
842 * recent_rgrp_next - get next RG from "recent" list
843 * @cur_rgd: current rgrp
846 * Returns: The next rgrp in the recent list
849 static struct gfs2_rgrpd *recent_rgrp_next(struct gfs2_rgrpd *cur_rgd,
852 struct gfs2_sbd *sdp = cur_rgd->rd_sbd;
853 struct list_head *head;
854 struct gfs2_rgrpd *rgd;
856 spin_lock(&sdp->sd_rindex_spin);
858 head = &sdp->sd_rindex_recent_list;
860 list_for_each_entry(rgd, head, rd_recent) {
861 if (rgd == cur_rgd) {
862 if (cur_rgd->rd_recent.next != head)
863 rgd = list_entry(cur_rgd->rd_recent.next,
864 struct gfs2_rgrpd, rd_recent);
869 list_del(&cur_rgd->rd_recent);
876 if (!list_empty(head))
877 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_recent);
880 spin_unlock(&sdp->sd_rindex_spin);
885 * recent_rgrp_add - add an RG to tail of "recent" list
886 * @new_rgd: The rgrp to add
890 static void recent_rgrp_add(struct gfs2_rgrpd *new_rgd)
892 struct gfs2_sbd *sdp = new_rgd->rd_sbd;
893 struct gfs2_rgrpd *rgd;
894 unsigned int count = 0;
895 unsigned int max = sdp->sd_rgrps / gfs2_jindex_size(sdp);
897 spin_lock(&sdp->sd_rindex_spin);
899 list_for_each_entry(rgd, &sdp->sd_rindex_recent_list, rd_recent) {
906 list_add_tail(&new_rgd->rd_recent, &sdp->sd_rindex_recent_list);
909 spin_unlock(&sdp->sd_rindex_spin);
913 * forward_rgrp_get - get an rgrp to try next from full list
914 * @sdp: The GFS2 superblock
916 * Returns: The rgrp to try next
919 static struct gfs2_rgrpd *forward_rgrp_get(struct gfs2_sbd *sdp)
921 struct gfs2_rgrpd *rgd;
922 unsigned int journals = gfs2_jindex_size(sdp);
923 unsigned int rg = 0, x;
925 spin_lock(&sdp->sd_rindex_spin);
927 rgd = sdp->sd_rindex_forward;
929 if (sdp->sd_rgrps >= journals)
930 rg = sdp->sd_rgrps * sdp->sd_jdesc->jd_jid / journals;
932 for (x = 0, rgd = gfs2_rgrpd_get_first(sdp); x < rg;
933 x++, rgd = gfs2_rgrpd_get_next(rgd))
936 sdp->sd_rindex_forward = rgd;
939 spin_unlock(&sdp->sd_rindex_spin);
945 * forward_rgrp_set - set the forward rgrp pointer
946 * @sdp: the filesystem
947 * @rgd: The new forward rgrp
951 static void forward_rgrp_set(struct gfs2_sbd *sdp, struct gfs2_rgrpd *rgd)
953 spin_lock(&sdp->sd_rindex_spin);
954 sdp->sd_rindex_forward = rgd;
955 spin_unlock(&sdp->sd_rindex_spin);
959 * get_local_rgrp - Choose and lock a rgrp for allocation
960 * @ip: the inode to reserve space for
961 * @rgp: the chosen and locked rgrp
963 * Try to acquire rgrp in way which avoids contending with others.
968 static int get_local_rgrp(struct gfs2_inode *ip)
970 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
971 struct gfs2_rgrpd *rgd, *begin = NULL;
972 struct gfs2_alloc *al = &ip->i_alloc;
973 int flags = LM_FLAG_TRY;
978 /* Try recently successful rgrps */
980 rgd = recent_rgrp_first(sdp, ip->i_last_rg_alloc);
983 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
984 LM_FLAG_TRY, &al->al_rgd_gh);
987 if (try_rgrp_fit(rgd, al))
989 gfs2_glock_dq_uninit(&al->al_rgd_gh);
990 rgd = recent_rgrp_next(rgd, 1);
994 rgd = recent_rgrp_next(rgd, 0);
1002 /* Go through full list of rgrps */
1004 begin = rgd = forward_rgrp_get(sdp);
1007 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, flags,
1011 if (try_rgrp_fit(rgd, al))
1013 gfs2_glock_dq_uninit(&al->al_rgd_gh);
1024 rgd = gfs2_rgrpd_get_next(rgd);
1026 rgd = gfs2_rgrpd_get_first(sdp);
1035 gfs2_log_flush(sdp, NULL);
1040 ip->i_last_rg_alloc = rgd->rd_ri.ri_addr;
1043 recent_rgrp_add(rgd);
1044 rgd = gfs2_rgrpd_get_next(rgd);
1046 rgd = gfs2_rgrpd_get_first(sdp);
1047 forward_rgrp_set(sdp, rgd);
1054 * gfs2_inplace_reserve_i - Reserve space in the filesystem
1055 * @ip: the inode to reserve space for
1060 int gfs2_inplace_reserve_i(struct gfs2_inode *ip, char *file, unsigned int line)
1062 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1063 struct gfs2_alloc *al = &ip->i_alloc;
1066 if (gfs2_assert_warn(sdp, al->al_requested))
1069 /* We need to hold the rindex unless the inode we're using is
1070 the rindex itself, in which case it's already held. */
1071 if (ip != GFS2_I(sdp->sd_rindex))
1072 error = gfs2_rindex_hold(sdp, &al->al_ri_gh);
1073 else if (!sdp->sd_rgrps) /* We may not have the rindex read in, so: */
1074 error = gfs2_ri_update_special(ip);
1079 error = get_local_rgrp(ip);
1081 if (ip != GFS2_I(sdp->sd_rindex))
1082 gfs2_glock_dq_uninit(&al->al_ri_gh);
1093 * gfs2_inplace_release - release an inplace reservation
1094 * @ip: the inode the reservation was taken out on
1096 * Release a reservation made by gfs2_inplace_reserve().
1099 void gfs2_inplace_release(struct gfs2_inode *ip)
1101 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1102 struct gfs2_alloc *al = &ip->i_alloc;
1104 if (gfs2_assert_warn(sdp, al->al_alloced <= al->al_requested) == -1)
1105 fs_warn(sdp, "al_alloced = %u, al_requested = %u "
1106 "al_file = %s, al_line = %u\n",
1107 al->al_alloced, al->al_requested, al->al_file,
1111 gfs2_glock_dq_uninit(&al->al_rgd_gh);
1112 if (ip != GFS2_I(sdp->sd_rindex))
1113 gfs2_glock_dq_uninit(&al->al_ri_gh);
1117 * gfs2_get_block_type - Check a block in a RG is of given type
1118 * @rgd: the resource group holding the block
1119 * @block: the block number
1121 * Returns: The block type (GFS2_BLKST_*)
1124 unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block)
1126 struct gfs2_bitmap *bi = NULL;
1127 u32 length, rgrp_block, buf_block;
1131 length = rgd->rd_ri.ri_length;
1132 rgrp_block = block - rgd->rd_ri.ri_data0;
1134 for (buf = 0; buf < length; buf++) {
1135 bi = rgd->rd_bits + buf;
1136 if (rgrp_block < (bi->bi_start + bi->bi_len) * GFS2_NBBY)
1140 gfs2_assert(rgd->rd_sbd, buf < length);
1141 buf_block = rgrp_block - bi->bi_start * GFS2_NBBY;
1143 type = gfs2_testbit(rgd, bi->bi_bh->b_data + bi->bi_offset,
1144 bi->bi_len, buf_block);
1150 * rgblk_search - find a block in @old_state, change allocation
1151 * state to @new_state
1152 * @rgd: the resource group descriptor
1153 * @goal: the goal block within the RG (start here to search for avail block)
1154 * @old_state: GFS2_BLKST_XXX the before-allocation state to find
1155 * @new_state: GFS2_BLKST_XXX the after-allocation block state
1157 * Walk rgrp's bitmap to find bits that represent a block in @old_state.
1158 * Add the found bitmap buffer to the transaction.
1159 * Set the found bits to @new_state to change block's allocation state.
1161 * This function never fails, because we wouldn't call it unless we
1162 * know (from reservation results, etc.) that a block is available.
1164 * Scope of @goal and returned block is just within rgrp, not the whole
1167 * Returns: the block number allocated
1170 static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal,
1171 unsigned char old_state, unsigned char new_state)
1173 struct gfs2_bitmap *bi = NULL;
1174 u32 length = rgd->rd_ri.ri_length;
1176 unsigned int buf, x;
1178 /* Find bitmap block that contains bits for goal block */
1179 for (buf = 0; buf < length; buf++) {
1180 bi = rgd->rd_bits + buf;
1181 if (goal < (bi->bi_start + bi->bi_len) * GFS2_NBBY)
1185 gfs2_assert(rgd->rd_sbd, buf < length);
1187 /* Convert scope of "goal" from rgrp-wide to within found bit block */
1188 goal -= bi->bi_start * GFS2_NBBY;
1190 /* Search (up to entire) bitmap in this rgrp for allocatable block.
1191 "x <= length", instead of "x < length", because we typically start
1192 the search in the middle of a bit block, but if we can't find an
1193 allocatable block anywhere else, we want to be able wrap around and
1194 search in the first part of our first-searched bit block. */
1195 for (x = 0; x <= length; x++) {
1197 blk = gfs2_bitfit(rgd, bi->bi_clone + bi->bi_offset,
1198 bi->bi_len, goal, old_state);
1200 blk = gfs2_bitfit(rgd,
1201 bi->bi_bh->b_data + bi->bi_offset,
1202 bi->bi_len, goal, old_state);
1203 if (blk != BFITNOENT)
1206 /* Try next bitmap block (wrap back to rgrp header if at end) */
1207 buf = (buf + 1) % length;
1208 bi = rgd->rd_bits + buf;
1212 if (gfs2_assert_withdraw(rgd->rd_sbd, x <= length))
1215 gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1);
1216 gfs2_setbit(rgd, bi->bi_bh->b_data + bi->bi_offset,
1217 bi->bi_len, blk, new_state);
1219 gfs2_setbit(rgd, bi->bi_clone + bi->bi_offset,
1220 bi->bi_len, blk, new_state);
1222 return bi->bi_start * GFS2_NBBY + blk;
1226 * rgblk_free - Change alloc state of given block(s)
1227 * @sdp: the filesystem
1228 * @bstart: the start of a run of blocks to free
1229 * @blen: the length of the block run (all must lie within ONE RG!)
1230 * @new_state: GFS2_BLKST_XXX the after-allocation block state
1232 * Returns: Resource group containing the block(s)
1235 static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart,
1236 u32 blen, unsigned char new_state)
1238 struct gfs2_rgrpd *rgd;
1239 struct gfs2_bitmap *bi = NULL;
1240 u32 length, rgrp_blk, buf_blk;
1243 rgd = gfs2_blk2rgrpd(sdp, bstart);
1245 if (gfs2_consist(sdp))
1246 fs_err(sdp, "block = %llu\n", (unsigned long long)bstart);
1250 length = rgd->rd_ri.ri_length;
1252 rgrp_blk = bstart - rgd->rd_ri.ri_data0;
1255 for (buf = 0; buf < length; buf++) {
1256 bi = rgd->rd_bits + buf;
1257 if (rgrp_blk < (bi->bi_start + bi->bi_len) * GFS2_NBBY)
1261 gfs2_assert(rgd->rd_sbd, buf < length);
1263 buf_blk = rgrp_blk - bi->bi_start * GFS2_NBBY;
1266 if (!bi->bi_clone) {
1267 bi->bi_clone = kmalloc(bi->bi_bh->b_size,
1268 GFP_NOFS | __GFP_NOFAIL);
1269 memcpy(bi->bi_clone + bi->bi_offset,
1270 bi->bi_bh->b_data + bi->bi_offset,
1273 gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1);
1274 gfs2_setbit(rgd, bi->bi_bh->b_data + bi->bi_offset,
1275 bi->bi_len, buf_blk, new_state);
1282 * gfs2_alloc_data - Allocate a data block
1283 * @ip: the inode to allocate the data block for
1285 * Returns: the allocated block
1288 u64 gfs2_alloc_data(struct gfs2_inode *ip)
1290 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1291 struct gfs2_alloc *al = &ip->i_alloc;
1292 struct gfs2_rgrpd *rgd = al->al_rgd;
1296 if (rgrp_contains_block(&rgd->rd_ri, ip->i_di.di_goal_data))
1297 goal = ip->i_di.di_goal_data - rgd->rd_ri.ri_data0;
1299 goal = rgd->rd_last_alloc_data;
1301 blk = rgblk_search(rgd, goal, GFS2_BLKST_FREE, GFS2_BLKST_USED);
1302 rgd->rd_last_alloc_data = blk;
1304 block = rgd->rd_ri.ri_data0 + blk;
1305 ip->i_di.di_goal_data = block;
1307 gfs2_assert_withdraw(sdp, rgd->rd_rg.rg_free);
1308 rgd->rd_rg.rg_free--;
1310 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1311 gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data);
1315 gfs2_statfs_change(sdp, 0, -1, 0);
1316 gfs2_quota_change(ip, +1, ip->i_inode.i_uid, ip->i_inode.i_gid);
1318 spin_lock(&sdp->sd_rindex_spin);
1319 rgd->rd_free_clone--;
1320 spin_unlock(&sdp->sd_rindex_spin);
1326 * gfs2_alloc_meta - Allocate a metadata block
1327 * @ip: the inode to allocate the metadata block for
1329 * Returns: the allocated block
1332 u64 gfs2_alloc_meta(struct gfs2_inode *ip)
1334 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1335 struct gfs2_alloc *al = &ip->i_alloc;
1336 struct gfs2_rgrpd *rgd = al->al_rgd;
1340 if (rgrp_contains_block(&rgd->rd_ri, ip->i_di.di_goal_meta))
1341 goal = ip->i_di.di_goal_meta - rgd->rd_ri.ri_data0;
1343 goal = rgd->rd_last_alloc_meta;
1345 blk = rgblk_search(rgd, goal, GFS2_BLKST_FREE, GFS2_BLKST_USED);
1346 rgd->rd_last_alloc_meta = blk;
1348 block = rgd->rd_ri.ri_data0 + blk;
1349 ip->i_di.di_goal_meta = block;
1351 gfs2_assert_withdraw(sdp, rgd->rd_rg.rg_free);
1352 rgd->rd_rg.rg_free--;
1354 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1355 gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data);
1359 gfs2_statfs_change(sdp, 0, -1, 0);
1360 gfs2_quota_change(ip, +1, ip->i_inode.i_uid, ip->i_inode.i_gid);
1361 gfs2_trans_add_unrevoke(sdp, block);
1363 spin_lock(&sdp->sd_rindex_spin);
1364 rgd->rd_free_clone--;
1365 spin_unlock(&sdp->sd_rindex_spin);
1371 * gfs2_alloc_di - Allocate a dinode
1372 * @dip: the directory that the inode is going in
1374 * Returns: the block allocated
1377 u64 gfs2_alloc_di(struct gfs2_inode *dip, u64 *generation)
1379 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1380 struct gfs2_alloc *al = &dip->i_alloc;
1381 struct gfs2_rgrpd *rgd = al->al_rgd;
1385 blk = rgblk_search(rgd, rgd->rd_last_alloc_meta,
1386 GFS2_BLKST_FREE, GFS2_BLKST_DINODE);
1388 rgd->rd_last_alloc_meta = blk;
1390 block = rgd->rd_ri.ri_data0 + blk;
1392 gfs2_assert_withdraw(sdp, rgd->rd_rg.rg_free);
1393 rgd->rd_rg.rg_free--;
1394 rgd->rd_rg.rg_dinodes++;
1395 *generation = rgd->rd_rg.rg_igeneration++;
1396 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1397 gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data);
1401 gfs2_statfs_change(sdp, 0, -1, +1);
1402 gfs2_trans_add_unrevoke(sdp, block);
1404 spin_lock(&sdp->sd_rindex_spin);
1405 rgd->rd_free_clone--;
1406 spin_unlock(&sdp->sd_rindex_spin);
1412 * gfs2_free_data - free a contiguous run of data block(s)
1413 * @ip: the inode these blocks are being freed from
1414 * @bstart: first block of a run of contiguous blocks
1415 * @blen: the length of the block run
1419 void gfs2_free_data(struct gfs2_inode *ip, u64 bstart, u32 blen)
1421 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1422 struct gfs2_rgrpd *rgd;
1424 rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE);
1428 rgd->rd_rg.rg_free += blen;
1430 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1431 gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data);
1433 gfs2_trans_add_rg(rgd);
1435 gfs2_statfs_change(sdp, 0, +blen, 0);
1436 gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid);
1440 * gfs2_free_meta - free a contiguous run of data block(s)
1441 * @ip: the inode these blocks are being freed from
1442 * @bstart: first block of a run of contiguous blocks
1443 * @blen: the length of the block run
1447 void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen)
1449 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1450 struct gfs2_rgrpd *rgd;
1452 rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE);
1456 rgd->rd_rg.rg_free += blen;
1458 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1459 gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data);
1461 gfs2_trans_add_rg(rgd);
1463 gfs2_statfs_change(sdp, 0, +blen, 0);
1464 gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid);
1465 gfs2_meta_wipe(ip, bstart, blen);
1468 void gfs2_unlink_di(struct inode *inode)
1470 struct gfs2_inode *ip = GFS2_I(inode);
1471 struct gfs2_sbd *sdp = GFS2_SB(inode);
1472 struct gfs2_rgrpd *rgd;
1473 u64 blkno = ip->i_no_addr;
1475 rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED);
1478 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1479 gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data);
1480 gfs2_trans_add_rg(rgd);
1483 static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno)
1485 struct gfs2_sbd *sdp = rgd->rd_sbd;
1486 struct gfs2_rgrpd *tmp_rgd;
1488 tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE);
1491 gfs2_assert_withdraw(sdp, rgd == tmp_rgd);
1493 if (!rgd->rd_rg.rg_dinodes)
1494 gfs2_consist_rgrpd(rgd);
1495 rgd->rd_rg.rg_dinodes--;
1496 rgd->rd_rg.rg_free++;
1498 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1499 gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data);
1501 gfs2_statfs_change(sdp, 0, +1, -1);
1502 gfs2_trans_add_rg(rgd);
1506 void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip)
1508 gfs2_free_uninit_di(rgd, ip->i_no_addr);
1509 gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid);
1510 gfs2_meta_wipe(ip, ip->i_no_addr, 1);
1514 * gfs2_rlist_add - add a RG to a list of RGs
1515 * @sdp: the filesystem
1516 * @rlist: the list of resource groups
1519 * Figure out what RG a block belongs to and add that RG to the list
1521 * FIXME: Don't use NOFAIL
1525 void gfs2_rlist_add(struct gfs2_sbd *sdp, struct gfs2_rgrp_list *rlist,
1528 struct gfs2_rgrpd *rgd;
1529 struct gfs2_rgrpd **tmp;
1530 unsigned int new_space;
1533 if (gfs2_assert_warn(sdp, !rlist->rl_ghs))
1536 rgd = gfs2_blk2rgrpd(sdp, block);
1538 if (gfs2_consist(sdp))
1539 fs_err(sdp, "block = %llu\n", (unsigned long long)block);
1543 for (x = 0; x < rlist->rl_rgrps; x++)
1544 if (rlist->rl_rgd[x] == rgd)
1547 if (rlist->rl_rgrps == rlist->rl_space) {
1548 new_space = rlist->rl_space + 10;
1550 tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *),
1551 GFP_NOFS | __GFP_NOFAIL);
1553 if (rlist->rl_rgd) {
1554 memcpy(tmp, rlist->rl_rgd,
1555 rlist->rl_space * sizeof(struct gfs2_rgrpd *));
1556 kfree(rlist->rl_rgd);
1559 rlist->rl_space = new_space;
1560 rlist->rl_rgd = tmp;
1563 rlist->rl_rgd[rlist->rl_rgrps++] = rgd;
1567 * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate
1568 * and initialize an array of glock holders for them
1569 * @rlist: the list of resource groups
1570 * @state: the lock state to acquire the RG lock in
1571 * @flags: the modifier flags for the holder structures
1573 * FIXME: Don't use NOFAIL
1577 void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state,
1582 rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder),
1583 GFP_NOFS | __GFP_NOFAIL);
1584 for (x = 0; x < rlist->rl_rgrps; x++)
1585 gfs2_holder_init(rlist->rl_rgd[x]->rd_gl,
1591 * gfs2_rlist_free - free a resource group list
1592 * @list: the list of resource groups
1596 void gfs2_rlist_free(struct gfs2_rgrp_list *rlist)
1600 kfree(rlist->rl_rgd);
1602 if (rlist->rl_ghs) {
1603 for (x = 0; x < rlist->rl_rgrps; x++)
1604 gfs2_holder_uninit(&rlist->rl_ghs[x]);
1605 kfree(rlist->rl_ghs);