2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2005 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 v.2.
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
16 #include <linux/gfs2_ondisk.h>
17 #include <asm/semaphore.h>
20 #include "lm_interface.h"
35 * gfs2_rgrp_verify - Verify that a resource group is consistent
36 * @sdp: the filesystem
41 void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd)
43 struct gfs2_sbd *sdp = rgd->rd_sbd;
44 struct gfs2_bitmap *bi = NULL;
45 uint32_t length = rgd->rd_ri.ri_length;
46 uint32_t count[4], tmp;
49 memset(count, 0, 4 * sizeof(uint32_t));
51 /* Count # blocks in each of 4 possible allocation states */
52 for (buf = 0; buf < length; buf++) {
53 bi = rgd->rd_bits + buf;
54 for (x = 0; x < 4; x++)
55 count[x] += gfs2_bitcount(rgd,
61 if (count[0] != rgd->rd_rg.rg_free) {
62 if (gfs2_consist_rgrpd(rgd))
63 fs_err(sdp, "free data mismatch: %u != %u\n",
64 count[0], rgd->rd_rg.rg_free);
68 tmp = rgd->rd_ri.ri_data -
70 rgd->rd_rg.rg_dinodes;
71 if (count[1] != tmp) {
72 if (gfs2_consist_rgrpd(rgd))
73 fs_err(sdp, "used data mismatch: %u != %u\n",
79 if (gfs2_consist_rgrpd(rgd))
80 fs_err(sdp, "free metadata mismatch: %u != 0\n",
85 if (count[3] != rgd->rd_rg.rg_dinodes) {
86 if (gfs2_consist_rgrpd(rgd))
87 fs_err(sdp, "used metadata mismatch: %u != %u\n",
88 count[3], rgd->rd_rg.rg_dinodes);
93 static inline int rgrp_contains_block(struct gfs2_rindex *ri, uint64_t block)
95 uint64_t first = ri->ri_data0;
96 uint64_t last = first + ri->ri_data;
97 return !!(first <= block && block < last);
101 * gfs2_blk2rgrpd - Find resource group for a given data/meta block number
102 * @sdp: The GFS2 superblock
103 * @n: The data block number
105 * Returns: The resource group, or NULL if not found
108 struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, uint64_t blk)
110 struct gfs2_rgrpd *rgd;
112 spin_lock(&sdp->sd_rindex_spin);
114 list_for_each_entry(rgd, &sdp->sd_rindex_mru_list, rd_list_mru) {
115 if (rgrp_contains_block(&rgd->rd_ri, blk)) {
116 list_move(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list);
117 spin_unlock(&sdp->sd_rindex_spin);
122 spin_unlock(&sdp->sd_rindex_spin);
128 * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem
129 * @sdp: The GFS2 superblock
131 * Returns: The first rgrp in the filesystem
134 struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp)
136 gfs2_assert(sdp, !list_empty(&sdp->sd_rindex_list));
137 return list_entry(sdp->sd_rindex_list.next, struct gfs2_rgrpd, rd_list);
141 * gfs2_rgrpd_get_next - get the next RG
144 * Returns: The next rgrp
147 struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd)
149 if (rgd->rd_list.next == &rgd->rd_sbd->sd_rindex_list)
151 return list_entry(rgd->rd_list.next, struct gfs2_rgrpd, rd_list);
154 static void clear_rgrpdi(struct gfs2_sbd *sdp)
156 struct list_head *head;
157 struct gfs2_rgrpd *rgd;
158 struct gfs2_glock *gl;
160 spin_lock(&sdp->sd_rindex_spin);
161 sdp->sd_rindex_forward = NULL;
162 head = &sdp->sd_rindex_recent_list;
163 while (!list_empty(head)) {
164 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_recent);
165 list_del(&rgd->rd_recent);
167 spin_unlock(&sdp->sd_rindex_spin);
169 head = &sdp->sd_rindex_list;
170 while (!list_empty(head)) {
171 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_list);
174 list_del(&rgd->rd_list);
175 list_del(&rgd->rd_list_mru);
178 gl->gl_object = NULL;
187 void gfs2_clear_rgrpd(struct gfs2_sbd *sdp)
189 mutex_lock(&sdp->sd_rindex_mutex);
191 mutex_unlock(&sdp->sd_rindex_mutex);
195 * gfs2_compute_bitstructs - Compute the bitmap sizes
196 * @rgd: The resource group descriptor
198 * Calculates bitmap descriptors, one for each block that contains bitmap data
203 static int compute_bitstructs(struct gfs2_rgrpd *rgd)
205 struct gfs2_sbd *sdp = rgd->rd_sbd;
206 struct gfs2_bitmap *bi;
207 uint32_t length = rgd->rd_ri.ri_length; /* # blocks in hdr & bitmap */
208 uint32_t bytes_left, bytes;
211 rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_KERNEL);
215 bytes_left = rgd->rd_ri.ri_bitbytes;
217 for (x = 0; x < length; x++) {
218 bi = rgd->rd_bits + x;
220 /* small rgrp; bitmap stored completely in header block */
223 bi->bi_offset = sizeof(struct gfs2_rgrp);
228 bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp);
229 bi->bi_offset = sizeof(struct gfs2_rgrp);
233 } else if (x + 1 == length) {
235 bi->bi_offset = sizeof(struct gfs2_meta_header);
236 bi->bi_start = rgd->rd_ri.ri_bitbytes - bytes_left;
240 bytes = sdp->sd_sb.sb_bsize -
241 sizeof(struct gfs2_meta_header);
242 bi->bi_offset = sizeof(struct gfs2_meta_header);
243 bi->bi_start = rgd->rd_ri.ri_bitbytes - bytes_left;
251 gfs2_consist_rgrpd(rgd);
254 bi = rgd->rd_bits + (length - 1);
255 if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_ri.ri_data) {
256 if (gfs2_consist_rgrpd(rgd)) {
257 gfs2_rindex_print(&rgd->rd_ri);
258 fs_err(sdp, "start=%u len=%u offset=%u\n",
259 bi->bi_start, bi->bi_len, bi->bi_offset);
268 * gfs2_ri_update - Pull in a new resource index from the disk
269 * @gl: The glock covering the rindex inode
271 * Returns: 0 on successful update, error code otherwise
274 static int gfs2_ri_update(struct gfs2_inode *ip)
276 struct gfs2_sbd *sdp = ip->i_sbd;
277 struct inode *inode = ip->i_vnode;
278 struct gfs2_rgrpd *rgd;
279 char buf[sizeof(struct gfs2_rindex)];
280 struct file_ra_state ra_state;
281 uint64_t junk = ip->i_di.di_size;
284 if (do_div(junk, sizeof(struct gfs2_rindex))) {
285 gfs2_consist_inode(ip);
291 file_ra_state_init(&ra_state, inode->i_mapping);
292 for (sdp->sd_rgrps = 0;; sdp->sd_rgrps++) {
293 loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex);
294 error = gfs2_internal_read(ip, &ra_state, buf, &pos,
295 sizeof(struct gfs2_rindex));
298 if (error != sizeof(struct gfs2_rindex)) {
304 rgd = kzalloc(sizeof(struct gfs2_rgrpd), GFP_KERNEL);
309 mutex_init(&rgd->rd_mutex);
310 lops_init_le(&rgd->rd_le, &gfs2_rg_lops);
313 list_add_tail(&rgd->rd_list, &sdp->sd_rindex_list);
314 list_add_tail(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list);
316 gfs2_rindex_in(&rgd->rd_ri, buf);
318 error = compute_bitstructs(rgd);
322 error = gfs2_glock_get(sdp, rgd->rd_ri.ri_addr,
323 &gfs2_rgrp_glops, CREATE, &rgd->rd_gl);
327 rgd->rd_gl->gl_object = rgd;
328 rgd->rd_rg_vn = rgd->rd_gl->gl_vn - 1;
331 sdp->sd_rindex_vn = ip->i_gl->gl_vn;
342 * gfs2_rindex_hold - Grab a lock on the rindex
343 * @sdp: The GFS2 superblock
344 * @ri_gh: the glock holder
346 * We grab a lock on the rindex inode to make sure that it doesn't
347 * change whilst we are performing an operation. We keep this lock
348 * for quite long periods of time compared to other locks. This
349 * doesn't matter, since it is shared and it is very, very rarely
350 * accessed in the exclusive mode (i.e. only when expanding the filesystem).
352 * This makes sure that we're using the latest copy of the resource index
353 * special file, which might have been updated if someone expanded the
354 * filesystem (via gfs2_grow utility), which adds new resource groups.
356 * Returns: 0 on success, error code otherwise
359 int gfs2_rindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ri_gh)
361 struct gfs2_inode *ip = sdp->sd_rindex->u.generic_ip;
362 struct gfs2_glock *gl = ip->i_gl;
365 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, ri_gh);
369 /* Read new copy from disk if we don't have the latest */
370 if (sdp->sd_rindex_vn != gl->gl_vn) {
371 mutex_lock(&sdp->sd_rindex_mutex);
372 if (sdp->sd_rindex_vn != gl->gl_vn) {
373 error = gfs2_ri_update(ip);
375 gfs2_glock_dq_uninit(ri_gh);
377 mutex_unlock(&sdp->sd_rindex_mutex);
384 * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps
385 * @rgd: the struct gfs2_rgrpd describing the RG to read in
387 * Read in all of a Resource Group's header and bitmap blocks.
388 * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps.
393 int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd)
395 struct gfs2_sbd *sdp = rgd->rd_sbd;
396 struct gfs2_glock *gl = rgd->rd_gl;
397 unsigned int length = rgd->rd_ri.ri_length;
398 struct gfs2_bitmap *bi;
402 mutex_lock(&rgd->rd_mutex);
404 spin_lock(&sdp->sd_rindex_spin);
405 if (rgd->rd_bh_count) {
407 spin_unlock(&sdp->sd_rindex_spin);
408 mutex_unlock(&rgd->rd_mutex);
411 spin_unlock(&sdp->sd_rindex_spin);
413 for (x = 0; x < length; x++) {
414 bi = rgd->rd_bits + x;
415 error = gfs2_meta_read(gl, rgd->rd_ri.ri_addr + x, DIO_START,
421 for (y = length; y--;) {
422 bi = rgd->rd_bits + y;
423 error = gfs2_meta_reread(sdp, bi->bi_bh, DIO_WAIT);
426 if (gfs2_metatype_check(sdp, bi->bi_bh,
427 (y) ? GFS2_METATYPE_RB :
434 if (rgd->rd_rg_vn != gl->gl_vn) {
435 gfs2_rgrp_in(&rgd->rd_rg, (rgd->rd_bits[0].bi_bh)->b_data);
436 rgd->rd_rg_vn = gl->gl_vn;
439 spin_lock(&sdp->sd_rindex_spin);
440 rgd->rd_free_clone = rgd->rd_rg.rg_free;
442 spin_unlock(&sdp->sd_rindex_spin);
444 mutex_unlock(&rgd->rd_mutex);
450 bi = rgd->rd_bits + x;
453 gfs2_assert_warn(sdp, !bi->bi_clone);
455 mutex_unlock(&rgd->rd_mutex);
460 void gfs2_rgrp_bh_hold(struct gfs2_rgrpd *rgd)
462 struct gfs2_sbd *sdp = rgd->rd_sbd;
464 spin_lock(&sdp->sd_rindex_spin);
465 gfs2_assert_warn(rgd->rd_sbd, rgd->rd_bh_count);
467 spin_unlock(&sdp->sd_rindex_spin);
471 * gfs2_rgrp_bh_put - Release RG bitmaps read in with gfs2_rgrp_bh_get()
472 * @rgd: the struct gfs2_rgrpd describing the RG to read in
476 void gfs2_rgrp_bh_put(struct gfs2_rgrpd *rgd)
478 struct gfs2_sbd *sdp = rgd->rd_sbd;
479 int x, length = rgd->rd_ri.ri_length;
481 spin_lock(&sdp->sd_rindex_spin);
482 gfs2_assert_warn(rgd->rd_sbd, rgd->rd_bh_count);
483 if (--rgd->rd_bh_count) {
484 spin_unlock(&sdp->sd_rindex_spin);
488 for (x = 0; x < length; x++) {
489 struct gfs2_bitmap *bi = rgd->rd_bits + x;
496 spin_unlock(&sdp->sd_rindex_spin);
499 void gfs2_rgrp_repolish_clones(struct gfs2_rgrpd *rgd)
501 struct gfs2_sbd *sdp = rgd->rd_sbd;
502 unsigned int length = rgd->rd_ri.ri_length;
505 for (x = 0; x < length; x++) {
506 struct gfs2_bitmap *bi = rgd->rd_bits + x;
509 memcpy(bi->bi_clone + bi->bi_offset,
510 bi->bi_bh->b_data + bi->bi_offset,
514 spin_lock(&sdp->sd_rindex_spin);
515 rgd->rd_free_clone = rgd->rd_rg.rg_free;
516 spin_unlock(&sdp->sd_rindex_spin);
520 * gfs2_alloc_get - get the struct gfs2_alloc structure for an inode
521 * @ip: the incore GFS2 inode structure
523 * Returns: the struct gfs2_alloc
526 struct gfs2_alloc *gfs2_alloc_get(struct gfs2_inode *ip)
528 struct gfs2_alloc *al = &ip->i_alloc;
530 /* FIXME: Should assert that the correct locks are held here... */
531 memset(al, 0, sizeof(*al));
536 * gfs2_alloc_put - throw away the struct gfs2_alloc for an inode
541 void gfs2_alloc_put(struct gfs2_inode *ip)
547 * try_rgrp_fit - See if a given reservation will fit in a given RG
549 * @al: the struct gfs2_alloc structure describing the reservation
551 * If there's room for the requested blocks to be allocated from the RG:
552 * Sets the $al_reserved_data field in @al.
553 * Sets the $al_reserved_meta field in @al.
554 * Sets the $al_rgd field in @al.
556 * Returns: 1 on success (it fits), 0 on failure (it doesn't fit)
559 static int try_rgrp_fit(struct gfs2_rgrpd *rgd, struct gfs2_alloc *al)
561 struct gfs2_sbd *sdp = rgd->rd_sbd;
564 spin_lock(&sdp->sd_rindex_spin);
565 if (rgd->rd_free_clone >= al->al_requested) {
569 spin_unlock(&sdp->sd_rindex_spin);
575 * recent_rgrp_first - get first RG from "recent" list
576 * @sdp: The GFS2 superblock
577 * @rglast: address of the rgrp used last
579 * Returns: The first rgrp in the recent list
582 static struct gfs2_rgrpd *recent_rgrp_first(struct gfs2_sbd *sdp,
585 struct gfs2_rgrpd *rgd = NULL;
587 spin_lock(&sdp->sd_rindex_spin);
589 if (list_empty(&sdp->sd_rindex_recent_list))
595 list_for_each_entry(rgd, &sdp->sd_rindex_recent_list, rd_recent) {
596 if (rgd->rd_ri.ri_addr == rglast)
601 rgd = list_entry(sdp->sd_rindex_recent_list.next, struct gfs2_rgrpd,
605 spin_unlock(&sdp->sd_rindex_spin);
611 * recent_rgrp_next - get next RG from "recent" list
612 * @cur_rgd: current rgrp
615 * Returns: The next rgrp in the recent list
618 static struct gfs2_rgrpd *recent_rgrp_next(struct gfs2_rgrpd *cur_rgd,
621 struct gfs2_sbd *sdp = cur_rgd->rd_sbd;
622 struct list_head *head;
623 struct gfs2_rgrpd *rgd;
625 spin_lock(&sdp->sd_rindex_spin);
627 head = &sdp->sd_rindex_recent_list;
629 list_for_each_entry(rgd, head, rd_recent) {
630 if (rgd == cur_rgd) {
631 if (cur_rgd->rd_recent.next != head)
632 rgd = list_entry(cur_rgd->rd_recent.next,
633 struct gfs2_rgrpd, rd_recent);
638 list_del(&cur_rgd->rd_recent);
645 if (!list_empty(head))
646 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_recent);
649 spin_unlock(&sdp->sd_rindex_spin);
655 * recent_rgrp_add - add an RG to tail of "recent" list
656 * @new_rgd: The rgrp to add
660 static void recent_rgrp_add(struct gfs2_rgrpd *new_rgd)
662 struct gfs2_sbd *sdp = new_rgd->rd_sbd;
663 struct gfs2_rgrpd *rgd;
664 unsigned int count = 0;
665 unsigned int max = sdp->sd_rgrps / gfs2_jindex_size(sdp);
667 spin_lock(&sdp->sd_rindex_spin);
669 list_for_each_entry(rgd, &sdp->sd_rindex_recent_list, rd_recent) {
676 list_add_tail(&new_rgd->rd_recent, &sdp->sd_rindex_recent_list);
679 spin_unlock(&sdp->sd_rindex_spin);
683 * forward_rgrp_get - get an rgrp to try next from full list
684 * @sdp: The GFS2 superblock
686 * Returns: The rgrp to try next
689 static struct gfs2_rgrpd *forward_rgrp_get(struct gfs2_sbd *sdp)
691 struct gfs2_rgrpd *rgd;
692 unsigned int journals = gfs2_jindex_size(sdp);
693 unsigned int rg = 0, x;
695 spin_lock(&sdp->sd_rindex_spin);
697 rgd = sdp->sd_rindex_forward;
699 if (sdp->sd_rgrps >= journals)
700 rg = sdp->sd_rgrps * sdp->sd_jdesc->jd_jid / journals;
702 for (x = 0, rgd = gfs2_rgrpd_get_first(sdp);
704 x++, rgd = gfs2_rgrpd_get_next(rgd))
707 sdp->sd_rindex_forward = rgd;
710 spin_unlock(&sdp->sd_rindex_spin);
716 * forward_rgrp_set - set the forward rgrp pointer
717 * @sdp: the filesystem
718 * @rgd: The new forward rgrp
722 static void forward_rgrp_set(struct gfs2_sbd *sdp, struct gfs2_rgrpd *rgd)
724 spin_lock(&sdp->sd_rindex_spin);
725 sdp->sd_rindex_forward = rgd;
726 spin_unlock(&sdp->sd_rindex_spin);
730 * get_local_rgrp - Choose and lock a rgrp for allocation
731 * @ip: the inode to reserve space for
732 * @rgp: the chosen and locked rgrp
734 * Try to acquire rgrp in way which avoids contending with others.
739 static int get_local_rgrp(struct gfs2_inode *ip)
741 struct gfs2_sbd *sdp = ip->i_sbd;
742 struct gfs2_rgrpd *rgd, *begin = NULL;
743 struct gfs2_alloc *al = &ip->i_alloc;
744 int flags = LM_FLAG_TRY;
749 /* Try recently successful rgrps */
751 rgd = recent_rgrp_first(sdp, ip->i_last_rg_alloc);
754 error = gfs2_glock_nq_init(rgd->rd_gl,
755 LM_ST_EXCLUSIVE, LM_FLAG_TRY,
759 if (try_rgrp_fit(rgd, al))
761 gfs2_glock_dq_uninit(&al->al_rgd_gh);
762 rgd = recent_rgrp_next(rgd, 1);
766 rgd = recent_rgrp_next(rgd, 0);
774 /* Go through full list of rgrps */
776 begin = rgd = forward_rgrp_get(sdp);
779 error = gfs2_glock_nq_init(rgd->rd_gl,
780 LM_ST_EXCLUSIVE, flags,
784 if (try_rgrp_fit(rgd, al))
786 gfs2_glock_dq_uninit(&al->al_rgd_gh);
797 rgd = gfs2_rgrpd_get_next(rgd);
799 rgd = gfs2_rgrpd_get_first(sdp);
802 if (++loops >= 2 || !skipped)
809 ip->i_last_rg_alloc = rgd->rd_ri.ri_addr;
812 recent_rgrp_add(rgd);
813 rgd = gfs2_rgrpd_get_next(rgd);
815 rgd = gfs2_rgrpd_get_first(sdp);
816 forward_rgrp_set(sdp, rgd);
823 * gfs2_inplace_reserve_i - Reserve space in the filesystem
824 * @ip: the inode to reserve space for
829 int gfs2_inplace_reserve_i(struct gfs2_inode *ip, char *file, unsigned int line)
831 struct gfs2_sbd *sdp = ip->i_sbd;
832 struct gfs2_alloc *al = &ip->i_alloc;
835 if (gfs2_assert_warn(sdp, al->al_requested))
838 error = gfs2_rindex_hold(sdp, &al->al_ri_gh);
842 error = get_local_rgrp(ip);
844 gfs2_glock_dq_uninit(&al->al_ri_gh);
855 * gfs2_inplace_release - release an inplace reservation
856 * @ip: the inode the reservation was taken out on
858 * Release a reservation made by gfs2_inplace_reserve().
861 void gfs2_inplace_release(struct gfs2_inode *ip)
863 struct gfs2_sbd *sdp = ip->i_sbd;
864 struct gfs2_alloc *al = &ip->i_alloc;
866 if (gfs2_assert_warn(sdp, al->al_alloced <= al->al_requested) == -1)
867 fs_warn(sdp, "al_alloced = %u, al_requested = %u "
868 "al_file = %s, al_line = %u\n",
869 al->al_alloced, al->al_requested, al->al_file,
873 gfs2_glock_dq_uninit(&al->al_rgd_gh);
874 gfs2_glock_dq_uninit(&al->al_ri_gh);
878 * gfs2_get_block_type - Check a block in a RG is of given type
879 * @rgd: the resource group holding the block
880 * @block: the block number
882 * Returns: The block type (GFS2_BLKST_*)
885 unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, uint64_t block)
887 struct gfs2_bitmap *bi = NULL;
888 uint32_t length, rgrp_block, buf_block;
892 length = rgd->rd_ri.ri_length;
893 rgrp_block = block - rgd->rd_ri.ri_data0;
895 for (buf = 0; buf < length; buf++) {
896 bi = rgd->rd_bits + buf;
897 if (rgrp_block < (bi->bi_start + bi->bi_len) * GFS2_NBBY)
901 gfs2_assert(rgd->rd_sbd, buf < length);
902 buf_block = rgrp_block - bi->bi_start * GFS2_NBBY;
904 type = gfs2_testbit(rgd,
905 bi->bi_bh->b_data + bi->bi_offset,
906 bi->bi_len, buf_block);
912 * rgblk_search - find a block in @old_state, change allocation
913 * state to @new_state
914 * @rgd: the resource group descriptor
915 * @goal: the goal block within the RG (start here to search for avail block)
916 * @old_state: GFS2_BLKST_XXX the before-allocation state to find
917 * @new_state: GFS2_BLKST_XXX the after-allocation block state
919 * Walk rgrp's bitmap to find bits that represent a block in @old_state.
920 * Add the found bitmap buffer to the transaction.
921 * Set the found bits to @new_state to change block's allocation state.
923 * This function never fails, because we wouldn't call it unless we
924 * know (from reservation results, etc.) that a block is available.
926 * Scope of @goal and returned block is just within rgrp, not the whole
929 * Returns: the block number allocated
932 static uint32_t rgblk_search(struct gfs2_rgrpd *rgd, uint32_t goal,
933 unsigned char old_state, unsigned char new_state)
935 struct gfs2_bitmap *bi = NULL;
936 uint32_t length = rgd->rd_ri.ri_length;
940 /* Find bitmap block that contains bits for goal block */
941 for (buf = 0; buf < length; buf++) {
942 bi = rgd->rd_bits + buf;
943 if (goal < (bi->bi_start + bi->bi_len) * GFS2_NBBY)
947 gfs2_assert(rgd->rd_sbd, buf < length);
949 /* Convert scope of "goal" from rgrp-wide to within found bit block */
950 goal -= bi->bi_start * GFS2_NBBY;
952 /* Search (up to entire) bitmap in this rgrp for allocatable block.
953 "x <= length", instead of "x < length", because we typically start
954 the search in the middle of a bit block, but if we can't find an
955 allocatable block anywhere else, we want to be able wrap around and
956 search in the first part of our first-searched bit block. */
957 for (x = 0; x <= length; x++) {
959 blk = gfs2_bitfit(rgd,
960 bi->bi_clone + bi->bi_offset,
961 bi->bi_len, goal, old_state);
963 blk = gfs2_bitfit(rgd,
964 bi->bi_bh->b_data + bi->bi_offset,
965 bi->bi_len, goal, old_state);
966 if (blk != BFITNOENT)
969 /* Try next bitmap block (wrap back to rgrp header if at end) */
970 buf = (buf + 1) % length;
971 bi = rgd->rd_bits + buf;
975 if (gfs2_assert_withdraw(rgd->rd_sbd, x <= length))
978 gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1);
980 bi->bi_bh->b_data + bi->bi_offset,
981 bi->bi_len, blk, new_state);
984 bi->bi_clone + bi->bi_offset,
985 bi->bi_len, blk, new_state);
987 return bi->bi_start * GFS2_NBBY + blk;
991 * rgblk_free - Change alloc state of given block(s)
992 * @sdp: the filesystem
993 * @bstart: the start of a run of blocks to free
994 * @blen: the length of the block run (all must lie within ONE RG!)
995 * @new_state: GFS2_BLKST_XXX the after-allocation block state
997 * Returns: Resource group containing the block(s)
1000 static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, uint64_t bstart,
1001 uint32_t blen, unsigned char new_state)
1003 struct gfs2_rgrpd *rgd;
1004 struct gfs2_bitmap *bi = NULL;
1005 uint32_t length, rgrp_blk, buf_blk;
1008 rgd = gfs2_blk2rgrpd(sdp, bstart);
1010 if (gfs2_consist(sdp))
1011 fs_err(sdp, "block = %llu\n", bstart);
1015 length = rgd->rd_ri.ri_length;
1017 rgrp_blk = bstart - rgd->rd_ri.ri_data0;
1020 for (buf = 0; buf < length; buf++) {
1021 bi = rgd->rd_bits + buf;
1022 if (rgrp_blk < (bi->bi_start + bi->bi_len) * GFS2_NBBY)
1026 gfs2_assert(rgd->rd_sbd, buf < length);
1028 buf_blk = rgrp_blk - bi->bi_start * GFS2_NBBY;
1031 if (!bi->bi_clone) {
1032 bi->bi_clone = kmalloc(bi->bi_bh->b_size,
1033 GFP_KERNEL | __GFP_NOFAIL);
1034 memcpy(bi->bi_clone + bi->bi_offset,
1035 bi->bi_bh->b_data + bi->bi_offset,
1038 gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1);
1040 bi->bi_bh->b_data + bi->bi_offset,
1041 bi->bi_len, buf_blk, new_state);
1048 * gfs2_alloc_data - Allocate a data block
1049 * @ip: the inode to allocate the data block for
1051 * Returns: the allocated block
1054 uint64_t gfs2_alloc_data(struct gfs2_inode *ip)
1056 struct gfs2_sbd *sdp = ip->i_sbd;
1057 struct gfs2_alloc *al = &ip->i_alloc;
1058 struct gfs2_rgrpd *rgd = al->al_rgd;
1062 if (rgrp_contains_block(&rgd->rd_ri, ip->i_di.di_goal_data))
1063 goal = ip->i_di.di_goal_data - rgd->rd_ri.ri_data0;
1065 goal = rgd->rd_last_alloc_data;
1067 blk = rgblk_search(rgd, goal,
1068 GFS2_BLKST_FREE, GFS2_BLKST_USED);
1069 rgd->rd_last_alloc_data = blk;
1071 block = rgd->rd_ri.ri_data0 + blk;
1072 ip->i_di.di_goal_data = block;
1074 gfs2_assert_withdraw(sdp, rgd->rd_rg.rg_free);
1075 rgd->rd_rg.rg_free--;
1077 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1078 gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data);
1082 gfs2_statfs_change(sdp, 0, -1, 0);
1083 gfs2_quota_change(ip, +1, ip->i_di.di_uid, ip->i_di.di_gid);
1085 spin_lock(&sdp->sd_rindex_spin);
1086 rgd->rd_free_clone--;
1087 spin_unlock(&sdp->sd_rindex_spin);
1093 * gfs2_alloc_meta - Allocate a metadata block
1094 * @ip: the inode to allocate the metadata block for
1096 * Returns: the allocated block
1099 uint64_t gfs2_alloc_meta(struct gfs2_inode *ip)
1101 struct gfs2_sbd *sdp = ip->i_sbd;
1102 struct gfs2_alloc *al = &ip->i_alloc;
1103 struct gfs2_rgrpd *rgd = al->al_rgd;
1107 if (rgrp_contains_block(&rgd->rd_ri, ip->i_di.di_goal_meta))
1108 goal = ip->i_di.di_goal_meta - rgd->rd_ri.ri_data0;
1110 goal = rgd->rd_last_alloc_meta;
1112 blk = rgblk_search(rgd, goal,
1113 GFS2_BLKST_FREE, GFS2_BLKST_USED);
1114 rgd->rd_last_alloc_meta = blk;
1116 block = rgd->rd_ri.ri_data0 + blk;
1117 ip->i_di.di_goal_meta = block;
1119 gfs2_assert_withdraw(sdp, rgd->rd_rg.rg_free);
1120 rgd->rd_rg.rg_free--;
1122 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1123 gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data);
1127 gfs2_statfs_change(sdp, 0, -1, 0);
1128 gfs2_quota_change(ip, +1, ip->i_di.di_uid, ip->i_di.di_gid);
1129 gfs2_trans_add_unrevoke(sdp, block);
1131 spin_lock(&sdp->sd_rindex_spin);
1132 rgd->rd_free_clone--;
1133 spin_unlock(&sdp->sd_rindex_spin);
1139 * gfs2_alloc_di - Allocate a dinode
1140 * @dip: the directory that the inode is going in
1142 * Returns: the block allocated
1145 uint64_t gfs2_alloc_di(struct gfs2_inode *dip)
1147 struct gfs2_sbd *sdp = dip->i_sbd;
1148 struct gfs2_alloc *al = &dip->i_alloc;
1149 struct gfs2_rgrpd *rgd = al->al_rgd;
1153 blk = rgblk_search(rgd, rgd->rd_last_alloc_meta,
1154 GFS2_BLKST_FREE, GFS2_BLKST_DINODE);
1156 rgd->rd_last_alloc_meta = blk;
1158 block = rgd->rd_ri.ri_data0 + blk;
1160 gfs2_assert_withdraw(sdp, rgd->rd_rg.rg_free);
1161 rgd->rd_rg.rg_free--;
1162 rgd->rd_rg.rg_dinodes++;
1164 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1165 gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data);
1169 gfs2_statfs_change(sdp, 0, -1, +1);
1170 gfs2_trans_add_unrevoke(sdp, block);
1172 spin_lock(&sdp->sd_rindex_spin);
1173 rgd->rd_free_clone--;
1174 spin_unlock(&sdp->sd_rindex_spin);
1180 * gfs2_free_data - free a contiguous run of data block(s)
1181 * @ip: the inode these blocks are being freed from
1182 * @bstart: first block of a run of contiguous blocks
1183 * @blen: the length of the block run
1187 void gfs2_free_data(struct gfs2_inode *ip, uint64_t bstart, uint32_t blen)
1189 struct gfs2_sbd *sdp = ip->i_sbd;
1190 struct gfs2_rgrpd *rgd;
1192 rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE);
1196 rgd->rd_rg.rg_free += blen;
1198 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1199 gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data);
1201 gfs2_trans_add_rg(rgd);
1203 gfs2_statfs_change(sdp, 0, +blen, 0);
1204 gfs2_quota_change(ip, -(int64_t)blen,
1205 ip->i_di.di_uid, ip->i_di.di_gid);
1209 * gfs2_free_meta - free a contiguous run of data block(s)
1210 * @ip: the inode these blocks are being freed from
1211 * @bstart: first block of a run of contiguous blocks
1212 * @blen: the length of the block run
1216 void gfs2_free_meta(struct gfs2_inode *ip, uint64_t bstart, uint32_t blen)
1218 struct gfs2_sbd *sdp = ip->i_sbd;
1219 struct gfs2_rgrpd *rgd;
1221 rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE);
1225 rgd->rd_rg.rg_free += blen;
1227 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1228 gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data);
1230 gfs2_trans_add_rg(rgd);
1232 gfs2_statfs_change(sdp, 0, +blen, 0);
1233 gfs2_quota_change(ip, -(int64_t)blen,
1234 ip->i_di.di_uid, ip->i_di.di_gid);
1235 gfs2_meta_wipe(ip, bstart, blen);
1238 void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, uint64_t blkno)
1240 struct gfs2_sbd *sdp = rgd->rd_sbd;
1241 struct gfs2_rgrpd *tmp_rgd;
1243 tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE);
1246 gfs2_assert_withdraw(sdp, rgd == tmp_rgd);
1248 if (!rgd->rd_rg.rg_dinodes)
1249 gfs2_consist_rgrpd(rgd);
1250 rgd->rd_rg.rg_dinodes--;
1251 rgd->rd_rg.rg_free++;
1253 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1254 gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data);
1256 gfs2_statfs_change(sdp, 0, +1, -1);
1257 gfs2_trans_add_rg(rgd);
1261 * gfs2_free_uninit_di - free a dinode block
1262 * @rgd: the resource group that contains the dinode
1267 void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip)
1269 gfs2_free_uninit_di(rgd, ip->i_num.no_addr);
1270 gfs2_quota_change(ip, -1, ip->i_di.di_uid, ip->i_di.di_gid);
1271 gfs2_meta_wipe(ip, ip->i_num.no_addr, 1);
1275 * gfs2_rlist_add - add a RG to a list of RGs
1276 * @sdp: the filesystem
1277 * @rlist: the list of resource groups
1280 * Figure out what RG a block belongs to and add that RG to the list
1282 * FIXME: Don't use NOFAIL
1286 void gfs2_rlist_add(struct gfs2_sbd *sdp, struct gfs2_rgrp_list *rlist,
1289 struct gfs2_rgrpd *rgd;
1290 struct gfs2_rgrpd **tmp;
1291 unsigned int new_space;
1294 if (gfs2_assert_warn(sdp, !rlist->rl_ghs))
1297 rgd = gfs2_blk2rgrpd(sdp, block);
1299 if (gfs2_consist(sdp))
1300 fs_err(sdp, "block = %llu\n", block);
1304 for (x = 0; x < rlist->rl_rgrps; x++)
1305 if (rlist->rl_rgd[x] == rgd)
1308 if (rlist->rl_rgrps == rlist->rl_space) {
1309 new_space = rlist->rl_space + 10;
1311 tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *),
1312 GFP_KERNEL | __GFP_NOFAIL);
1314 if (rlist->rl_rgd) {
1315 memcpy(tmp, rlist->rl_rgd,
1316 rlist->rl_space * sizeof(struct gfs2_rgrpd *));
1317 kfree(rlist->rl_rgd);
1320 rlist->rl_space = new_space;
1321 rlist->rl_rgd = tmp;
1324 rlist->rl_rgd[rlist->rl_rgrps++] = rgd;
1328 * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate
1329 * and initialize an array of glock holders for them
1330 * @rlist: the list of resource groups
1331 * @state: the lock state to acquire the RG lock in
1332 * @flags: the modifier flags for the holder structures
1334 * FIXME: Don't use NOFAIL
1338 void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state,
1343 rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder),
1344 GFP_KERNEL | __GFP_NOFAIL);
1345 for (x = 0; x < rlist->rl_rgrps; x++)
1346 gfs2_holder_init(rlist->rl_rgd[x]->rd_gl,
1352 * gfs2_rlist_free - free a resource group list
1353 * @list: the list of resource groups
1357 void gfs2_rlist_free(struct gfs2_rgrp_list *rlist)
1361 kfree(rlist->rl_rgd);
1363 if (rlist->rl_ghs) {
1364 for (x = 0; x < rlist->rl_rgrps; x++)
1365 gfs2_holder_uninit(&rlist->rl_ghs[x]);
1366 kfree(rlist->rl_ghs);