2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
15 #include <linux/gfs2_ondisk.h>
16 #include <linux/prefetch.h>
17 #include <linux/blkdev.h>
18 #include <linux/rbtree.h>
33 #include "trace_gfs2.h"
35 #define BFITNOENT ((u32)~0)
36 #define NO_BLOCK ((u64)~0)
38 #if BITS_PER_LONG == 32
39 #define LBITMASK (0x55555555UL)
40 #define LBITSKIP55 (0x55555555UL)
41 #define LBITSKIP00 (0x00000000UL)
43 #define LBITMASK (0x5555555555555555UL)
44 #define LBITSKIP55 (0x5555555555555555UL)
45 #define LBITSKIP00 (0x0000000000000000UL)
49 * These routines are used by the resource group routines (rgrp.c)
50 * to keep track of block allocation. Each block is represented by two
51 * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks.
54 * 1 = Used (not metadata)
55 * 2 = Unlinked (still in use) inode
59 static const char valid_change[16] = {
67 static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal,
68 unsigned char old_state,
69 struct gfs2_bitmap **rbi);
72 * gfs2_setbit - Set a bit in the bitmaps
73 * @rgd: the resource group descriptor
74 * @buf2: the clone buffer that holds the bitmaps
75 * @bi: the bitmap structure
76 * @block: the block to set
77 * @new_state: the new state of the block
81 static inline void gfs2_setbit(struct gfs2_rgrpd *rgd, unsigned char *buf2,
82 struct gfs2_bitmap *bi, u32 block,
83 unsigned char new_state)
85 unsigned char *byte1, *byte2, *end, cur_state;
86 unsigned int buflen = bi->bi_len;
87 const unsigned int bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE;
89 byte1 = bi->bi_bh->b_data + bi->bi_offset + (block / GFS2_NBBY);
90 end = bi->bi_bh->b_data + bi->bi_offset + buflen;
94 cur_state = (*byte1 >> bit) & GFS2_BIT_MASK;
96 if (unlikely(!valid_change[new_state * 4 + cur_state])) {
97 printk(KERN_WARNING "GFS2: buf_blk = 0x%llx old_state=%d, "
99 (unsigned long long)block, cur_state, new_state);
100 printk(KERN_WARNING "GFS2: rgrp=0x%llx bi_start=0x%lx\n",
101 (unsigned long long)rgd->rd_addr,
102 (unsigned long)bi->bi_start);
103 printk(KERN_WARNING "GFS2: bi_offset=0x%lx bi_len=0x%lx\n",
104 (unsigned long)bi->bi_offset,
105 (unsigned long)bi->bi_len);
107 gfs2_consist_rgrpd(rgd);
110 *byte1 ^= (cur_state ^ new_state) << bit;
113 byte2 = buf2 + bi->bi_offset + (block / GFS2_NBBY);
114 cur_state = (*byte2 >> bit) & GFS2_BIT_MASK;
115 *byte2 ^= (cur_state ^ new_state) << bit;
120 * gfs2_testbit - test a bit in the bitmaps
121 * @rgd: the resource group descriptor
122 * @buffer: the buffer that holds the bitmaps
123 * @buflen: the length (in bytes) of the buffer
124 * @block: the block to read
128 static inline unsigned char gfs2_testbit(struct gfs2_rgrpd *rgd,
129 const unsigned char *buffer,
130 unsigned int buflen, u32 block)
132 const unsigned char *byte, *end;
133 unsigned char cur_state;
136 byte = buffer + (block / GFS2_NBBY);
137 bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE;
138 end = buffer + buflen;
140 gfs2_assert(rgd->rd_sbd, byte < end);
142 cur_state = (*byte >> bit) & GFS2_BIT_MASK;
149 * @ptr: Pointer to bitmap data
150 * @mask: Mask to use (normally 0x55555.... but adjusted for search start)
151 * @state: The state we are searching for
153 * We xor the bitmap data with a patter which is the bitwise opposite
154 * of what we are looking for, this gives rise to a pattern of ones
155 * wherever there is a match. Since we have two bits per entry, we
156 * take this pattern, shift it down by one place and then and it with
157 * the original. All the even bit positions (0,2,4, etc) then represent
158 * successful matches, so we mask with 0x55555..... to remove the unwanted
161 * This allows searching of a whole u64 at once (32 blocks) with a
162 * single test (on 64 bit arches).
165 static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state)
168 static const u64 search[] = {
169 [0] = 0xffffffffffffffffULL,
170 [1] = 0xaaaaaaaaaaaaaaaaULL,
171 [2] = 0x5555555555555555ULL,
172 [3] = 0x0000000000000000ULL,
174 tmp = le64_to_cpu(*ptr) ^ search[state];
181 * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing
182 * a block in a given allocation state.
183 * @buf: the buffer that holds the bitmaps
184 * @len: the length (in bytes) of the buffer
185 * @goal: start search at this block's bit-pair (within @buffer)
186 * @state: GFS2_BLKST_XXX the state of the block we're looking for.
188 * Scope of @goal and returned block number is only within this bitmap buffer,
189 * not entire rgrp or filesystem. @buffer will be offset from the actual
190 * beginning of a bitmap block buffer, skipping any header structures, but
191 * headers are always a multiple of 64 bits long so that the buffer is
192 * always aligned to a 64 bit boundary.
194 * The size of the buffer is in bytes, but is it assumed that it is
195 * always ok to read a complete multiple of 64 bits at the end
196 * of the block in case the end is no aligned to a natural boundary.
198 * Return: the block number (bitmap buffer scope) that was found
201 static u32 gfs2_bitfit(const u8 *buf, const unsigned int len,
204 u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1);
205 const __le64 *ptr = ((__le64 *)buf) + (goal >> 5);
206 const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64)));
208 u64 mask = 0x5555555555555555ULL;
213 /* Mask off bits we don't care about at the start of the search */
215 tmp = gfs2_bit_search(ptr, mask, state);
217 while(tmp == 0 && ptr < end) {
218 tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state);
221 /* Mask off any bits which are more than len bytes from the start */
222 if (ptr == end && (len & (sizeof(u64) - 1)))
223 tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1))));
224 /* Didn't find anything, so return */
229 bit /= 2; /* two bits per entry in the bitmap */
230 return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit;
234 * gfs2_bitcount - count the number of bits in a certain state
235 * @rgd: the resource group descriptor
236 * @buffer: the buffer that holds the bitmaps
237 * @buflen: the length (in bytes) of the buffer
238 * @state: the state of the block we're looking for
240 * Returns: The number of bits
243 static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer,
244 unsigned int buflen, u8 state)
246 const u8 *byte = buffer;
247 const u8 *end = buffer + buflen;
248 const u8 state1 = state << 2;
249 const u8 state2 = state << 4;
250 const u8 state3 = state << 6;
253 for (; byte < end; byte++) {
254 if (((*byte) & 0x03) == state)
256 if (((*byte) & 0x0C) == state1)
258 if (((*byte) & 0x30) == state2)
260 if (((*byte) & 0xC0) == state3)
268 * gfs2_rgrp_verify - Verify that a resource group is consistent
273 void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd)
275 struct gfs2_sbd *sdp = rgd->rd_sbd;
276 struct gfs2_bitmap *bi = NULL;
277 u32 length = rgd->rd_length;
281 memset(count, 0, 4 * sizeof(u32));
283 /* Count # blocks in each of 4 possible allocation states */
284 for (buf = 0; buf < length; buf++) {
285 bi = rgd->rd_bits + buf;
286 for (x = 0; x < 4; x++)
287 count[x] += gfs2_bitcount(rgd,
293 if (count[0] != rgd->rd_free) {
294 if (gfs2_consist_rgrpd(rgd))
295 fs_err(sdp, "free data mismatch: %u != %u\n",
296 count[0], rgd->rd_free);
300 tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes;
301 if (count[1] != tmp) {
302 if (gfs2_consist_rgrpd(rgd))
303 fs_err(sdp, "used data mismatch: %u != %u\n",
308 if (count[2] + count[3] != rgd->rd_dinodes) {
309 if (gfs2_consist_rgrpd(rgd))
310 fs_err(sdp, "used metadata mismatch: %u != %u\n",
311 count[2] + count[3], rgd->rd_dinodes);
316 static inline int rgrp_contains_block(struct gfs2_rgrpd *rgd, u64 block)
318 u64 first = rgd->rd_data0;
319 u64 last = first + rgd->rd_data;
320 return first <= block && block < last;
324 * gfs2_blk2rgrpd - Find resource group for a given data/meta block number
325 * @sdp: The GFS2 superblock
326 * @blk: The data block number
327 * @exact: True if this needs to be an exact match
329 * Returns: The resource group, or NULL if not found
332 struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk, bool exact)
334 struct rb_node *n, *next;
335 struct gfs2_rgrpd *cur;
337 spin_lock(&sdp->sd_rindex_spin);
338 n = sdp->sd_rindex_tree.rb_node;
340 cur = rb_entry(n, struct gfs2_rgrpd, rd_node);
342 if (blk < cur->rd_addr)
344 else if (blk >= cur->rd_data0 + cur->rd_data)
347 spin_unlock(&sdp->sd_rindex_spin);
349 if (blk < cur->rd_addr)
351 if (blk >= cur->rd_data0 + cur->rd_data)
358 spin_unlock(&sdp->sd_rindex_spin);
364 * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem
365 * @sdp: The GFS2 superblock
367 * Returns: The first rgrp in the filesystem
370 struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp)
372 const struct rb_node *n;
373 struct gfs2_rgrpd *rgd;
375 spin_lock(&sdp->sd_rindex_spin);
376 n = rb_first(&sdp->sd_rindex_tree);
377 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node);
378 spin_unlock(&sdp->sd_rindex_spin);
384 * gfs2_rgrpd_get_next - get the next RG
385 * @rgd: the resource group descriptor
387 * Returns: The next rgrp
390 struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd)
392 struct gfs2_sbd *sdp = rgd->rd_sbd;
393 const struct rb_node *n;
395 spin_lock(&sdp->sd_rindex_spin);
396 n = rb_next(&rgd->rd_node);
398 n = rb_first(&sdp->sd_rindex_tree);
400 if (unlikely(&rgd->rd_node == n)) {
401 spin_unlock(&sdp->sd_rindex_spin);
404 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node);
405 spin_unlock(&sdp->sd_rindex_spin);
409 void gfs2_free_clones(struct gfs2_rgrpd *rgd)
413 for (x = 0; x < rgd->rd_length; x++) {
414 struct gfs2_bitmap *bi = rgd->rd_bits + x;
421 * gfs2_rs_alloc - make sure we have a reservation assigned to the inode
422 * @ip: the inode for this reservation
424 int gfs2_rs_alloc(struct gfs2_inode *ip)
428 down_write(&ip->i_rw_mutex);
430 ip->i_res = kmem_cache_zalloc(gfs2_rsrv_cachep, GFP_NOFS);
434 up_write(&ip->i_rw_mutex);
439 * gfs2_rs_delete - delete a reservation
440 * @ip: The inode for this reservation
443 void gfs2_rs_delete(struct gfs2_inode *ip)
445 down_write(&ip->i_rw_mutex);
447 kmem_cache_free(gfs2_rsrv_cachep, ip->i_res);
450 up_write(&ip->i_rw_mutex);
453 void gfs2_clear_rgrpd(struct gfs2_sbd *sdp)
456 struct gfs2_rgrpd *rgd;
457 struct gfs2_glock *gl;
459 while ((n = rb_first(&sdp->sd_rindex_tree))) {
460 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node);
463 rb_erase(n, &sdp->sd_rindex_tree);
466 spin_lock(&gl->gl_spin);
467 gl->gl_object = NULL;
468 spin_unlock(&gl->gl_spin);
469 gfs2_glock_add_to_lru(gl);
473 gfs2_free_clones(rgd);
475 kmem_cache_free(gfs2_rgrpd_cachep, rgd);
479 static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd)
481 printk(KERN_INFO " ri_addr = %llu\n", (unsigned long long)rgd->rd_addr);
482 printk(KERN_INFO " ri_length = %u\n", rgd->rd_length);
483 printk(KERN_INFO " ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0);
484 printk(KERN_INFO " ri_data = %u\n", rgd->rd_data);
485 printk(KERN_INFO " ri_bitbytes = %u\n", rgd->rd_bitbytes);
489 * gfs2_compute_bitstructs - Compute the bitmap sizes
490 * @rgd: The resource group descriptor
492 * Calculates bitmap descriptors, one for each block that contains bitmap data
497 static int compute_bitstructs(struct gfs2_rgrpd *rgd)
499 struct gfs2_sbd *sdp = rgd->rd_sbd;
500 struct gfs2_bitmap *bi;
501 u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */
502 u32 bytes_left, bytes;
508 rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS);
512 bytes_left = rgd->rd_bitbytes;
514 for (x = 0; x < length; x++) {
515 bi = rgd->rd_bits + x;
518 /* small rgrp; bitmap stored completely in header block */
521 bi->bi_offset = sizeof(struct gfs2_rgrp);
526 bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp);
527 bi->bi_offset = sizeof(struct gfs2_rgrp);
531 } else if (x + 1 == length) {
533 bi->bi_offset = sizeof(struct gfs2_meta_header);
534 bi->bi_start = rgd->rd_bitbytes - bytes_left;
538 bytes = sdp->sd_sb.sb_bsize -
539 sizeof(struct gfs2_meta_header);
540 bi->bi_offset = sizeof(struct gfs2_meta_header);
541 bi->bi_start = rgd->rd_bitbytes - bytes_left;
549 gfs2_consist_rgrpd(rgd);
552 bi = rgd->rd_bits + (length - 1);
553 if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) {
554 if (gfs2_consist_rgrpd(rgd)) {
555 gfs2_rindex_print(rgd);
556 fs_err(sdp, "start=%u len=%u offset=%u\n",
557 bi->bi_start, bi->bi_len, bi->bi_offset);
566 * gfs2_ri_total - Total up the file system space, according to the rindex.
567 * @sdp: the filesystem
570 u64 gfs2_ri_total(struct gfs2_sbd *sdp)
573 struct inode *inode = sdp->sd_rindex;
574 struct gfs2_inode *ip = GFS2_I(inode);
575 char buf[sizeof(struct gfs2_rindex)];
578 for (rgrps = 0;; rgrps++) {
579 loff_t pos = rgrps * sizeof(struct gfs2_rindex);
581 if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode))
583 error = gfs2_internal_read(ip, buf, &pos,
584 sizeof(struct gfs2_rindex));
585 if (error != sizeof(struct gfs2_rindex))
587 total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data);
592 static int rgd_insert(struct gfs2_rgrpd *rgd)
594 struct gfs2_sbd *sdp = rgd->rd_sbd;
595 struct rb_node **newn = &sdp->sd_rindex_tree.rb_node, *parent = NULL;
597 /* Figure out where to put new node */
599 struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd,
603 if (rgd->rd_addr < cur->rd_addr)
604 newn = &((*newn)->rb_left);
605 else if (rgd->rd_addr > cur->rd_addr)
606 newn = &((*newn)->rb_right);
611 rb_link_node(&rgd->rd_node, parent, newn);
612 rb_insert_color(&rgd->rd_node, &sdp->sd_rindex_tree);
618 * read_rindex_entry - Pull in a new resource index entry from the disk
619 * @ip: Pointer to the rindex inode
621 * Returns: 0 on success, > 0 on EOF, error code otherwise
624 static int read_rindex_entry(struct gfs2_inode *ip)
626 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
627 loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex);
628 struct gfs2_rindex buf;
630 struct gfs2_rgrpd *rgd;
632 if (pos >= i_size_read(&ip->i_inode))
635 error = gfs2_internal_read(ip, (char *)&buf, &pos,
636 sizeof(struct gfs2_rindex));
638 if (error != sizeof(struct gfs2_rindex))
639 return (error == 0) ? 1 : error;
641 rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS);
647 rgd->rd_addr = be64_to_cpu(buf.ri_addr);
648 rgd->rd_length = be32_to_cpu(buf.ri_length);
649 rgd->rd_data0 = be64_to_cpu(buf.ri_data0);
650 rgd->rd_data = be32_to_cpu(buf.ri_data);
651 rgd->rd_bitbytes = be32_to_cpu(buf.ri_bitbytes);
653 error = compute_bitstructs(rgd);
657 error = gfs2_glock_get(sdp, rgd->rd_addr,
658 &gfs2_rgrp_glops, CREATE, &rgd->rd_gl);
662 rgd->rd_gl->gl_object = rgd;
663 rgd->rd_rgl = (struct gfs2_rgrp_lvb *)rgd->rd_gl->gl_lvb;
664 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
665 if (rgd->rd_data > sdp->sd_max_rg_data)
666 sdp->sd_max_rg_data = rgd->rd_data;
667 spin_lock(&sdp->sd_rindex_spin);
668 error = rgd_insert(rgd);
669 spin_unlock(&sdp->sd_rindex_spin);
673 error = 0; /* someone else read in the rgrp; free it and ignore it */
674 gfs2_glock_put(rgd->rd_gl);
678 kmem_cache_free(gfs2_rgrpd_cachep, rgd);
683 * gfs2_ri_update - Pull in a new resource index from the disk
684 * @ip: pointer to the rindex inode
686 * Returns: 0 on successful update, error code otherwise
689 static int gfs2_ri_update(struct gfs2_inode *ip)
691 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
695 error = read_rindex_entry(ip);
696 } while (error == 0);
701 sdp->sd_rindex_uptodate = 1;
706 * gfs2_rindex_update - Update the rindex if required
707 * @sdp: The GFS2 superblock
709 * We grab a lock on the rindex inode to make sure that it doesn't
710 * change whilst we are performing an operation. We keep this lock
711 * for quite long periods of time compared to other locks. This
712 * doesn't matter, since it is shared and it is very, very rarely
713 * accessed in the exclusive mode (i.e. only when expanding the filesystem).
715 * This makes sure that we're using the latest copy of the resource index
716 * special file, which might have been updated if someone expanded the
717 * filesystem (via gfs2_grow utility), which adds new resource groups.
719 * Returns: 0 on succeess, error code otherwise
722 int gfs2_rindex_update(struct gfs2_sbd *sdp)
724 struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex);
725 struct gfs2_glock *gl = ip->i_gl;
726 struct gfs2_holder ri_gh;
728 int unlock_required = 0;
730 /* Read new copy from disk if we don't have the latest */
731 if (!sdp->sd_rindex_uptodate) {
732 if (!gfs2_glock_is_locked_by_me(gl)) {
733 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &ri_gh);
738 if (!sdp->sd_rindex_uptodate)
739 error = gfs2_ri_update(ip);
741 gfs2_glock_dq_uninit(&ri_gh);
747 static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf)
749 const struct gfs2_rgrp *str = buf;
752 rg_flags = be32_to_cpu(str->rg_flags);
753 rg_flags &= ~GFS2_RDF_MASK;
754 rgd->rd_flags &= GFS2_RDF_MASK;
755 rgd->rd_flags |= rg_flags;
756 rgd->rd_free = be32_to_cpu(str->rg_free);
757 rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes);
758 rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration);
761 static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf)
763 struct gfs2_rgrp *str = buf;
765 str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK);
766 str->rg_free = cpu_to_be32(rgd->rd_free);
767 str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes);
768 str->__pad = cpu_to_be32(0);
769 str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration);
770 memset(&str->rg_reserved, 0, sizeof(str->rg_reserved));
773 static int gfs2_rgrp_lvb_valid(struct gfs2_rgrpd *rgd)
775 struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl;
776 struct gfs2_rgrp *str = (struct gfs2_rgrp *)rgd->rd_bits[0].bi_bh->b_data;
778 if (rgl->rl_flags != str->rg_flags || rgl->rl_free != str->rg_free ||
779 rgl->rl_dinodes != str->rg_dinodes ||
780 rgl->rl_igeneration != str->rg_igeneration)
785 static void gfs2_rgrp_ondisk2lvb(struct gfs2_rgrp_lvb *rgl, const void *buf)
787 const struct gfs2_rgrp *str = buf;
789 rgl->rl_magic = cpu_to_be32(GFS2_MAGIC);
790 rgl->rl_flags = str->rg_flags;
791 rgl->rl_free = str->rg_free;
792 rgl->rl_dinodes = str->rg_dinodes;
793 rgl->rl_igeneration = str->rg_igeneration;
797 static void update_rgrp_lvb_unlinked(struct gfs2_rgrpd *rgd, u32 change)
799 struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl;
800 u32 unlinked = be32_to_cpu(rgl->rl_unlinked) + change;
801 rgl->rl_unlinked = cpu_to_be32(unlinked);
804 static u32 count_unlinked(struct gfs2_rgrpd *rgd)
806 struct gfs2_bitmap *bi;
807 const u32 length = rgd->rd_length;
808 const u8 *buffer = NULL;
809 u32 i, goal, count = 0;
811 for (i = 0, bi = rgd->rd_bits; i < length; i++, bi++) {
813 buffer = bi->bi_bh->b_data + bi->bi_offset;
814 WARN_ON(!buffer_uptodate(bi->bi_bh));
815 while (goal < bi->bi_len * GFS2_NBBY) {
816 goal = gfs2_bitfit(buffer, bi->bi_len, goal,
817 GFS2_BLKST_UNLINKED);
818 if (goal == BFITNOENT)
830 * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps
831 * @rgd: the struct gfs2_rgrpd describing the RG to read in
833 * Read in all of a Resource Group's header and bitmap blocks.
834 * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps.
839 int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd)
841 struct gfs2_sbd *sdp = rgd->rd_sbd;
842 struct gfs2_glock *gl = rgd->rd_gl;
843 unsigned int length = rgd->rd_length;
844 struct gfs2_bitmap *bi;
848 if (rgd->rd_bits[0].bi_bh != NULL)
851 for (x = 0; x < length; x++) {
852 bi = rgd->rd_bits + x;
853 error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, &bi->bi_bh);
858 for (y = length; y--;) {
859 bi = rgd->rd_bits + y;
860 error = gfs2_meta_wait(sdp, bi->bi_bh);
863 if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB :
870 if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) {
871 for (x = 0; x < length; x++)
872 clear_bit(GBF_FULL, &rgd->rd_bits[x].bi_flags);
873 gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data);
874 rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK);
875 rgd->rd_free_clone = rgd->rd_free;
877 if (be32_to_cpu(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) {
878 rgd->rd_rgl->rl_unlinked = cpu_to_be32(count_unlinked(rgd));
879 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl,
880 rgd->rd_bits[0].bi_bh->b_data);
882 else if (sdp->sd_args.ar_rgrplvb) {
883 if (!gfs2_rgrp_lvb_valid(rgd)){
884 gfs2_consist_rgrpd(rgd);
888 if (rgd->rd_rgl->rl_unlinked == 0)
889 rgd->rd_flags &= ~GFS2_RDF_CHECK;
895 bi = rgd->rd_bits + x;
898 gfs2_assert_warn(sdp, !bi->bi_clone);
904 int update_rgrp_lvb(struct gfs2_rgrpd *rgd)
908 if (rgd->rd_flags & GFS2_RDF_UPTODATE)
911 if (be32_to_cpu(GFS2_MAGIC) != rgd->rd_rgl->rl_magic)
912 return gfs2_rgrp_bh_get(rgd);
914 rl_flags = be32_to_cpu(rgd->rd_rgl->rl_flags);
915 rl_flags &= ~GFS2_RDF_MASK;
916 rgd->rd_flags &= GFS2_RDF_MASK;
917 rgd->rd_flags |= (rl_flags | GFS2_RDF_UPTODATE | GFS2_RDF_CHECK);
918 if (rgd->rd_rgl->rl_unlinked == 0)
919 rgd->rd_flags &= ~GFS2_RDF_CHECK;
920 rgd->rd_free = be32_to_cpu(rgd->rd_rgl->rl_free);
921 rgd->rd_free_clone = rgd->rd_free;
922 rgd->rd_dinodes = be32_to_cpu(rgd->rd_rgl->rl_dinodes);
923 rgd->rd_igeneration = be64_to_cpu(rgd->rd_rgl->rl_igeneration);
927 int gfs2_rgrp_go_lock(struct gfs2_holder *gh)
929 struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object;
930 struct gfs2_sbd *sdp = rgd->rd_sbd;
932 if (gh->gh_flags & GL_SKIP && sdp->sd_args.ar_rgrplvb)
934 return gfs2_rgrp_bh_get((struct gfs2_rgrpd *)gh->gh_gl->gl_object);
938 * gfs2_rgrp_go_unlock - Release RG bitmaps read in with gfs2_rgrp_bh_get()
939 * @gh: The glock holder for the resource group
943 void gfs2_rgrp_go_unlock(struct gfs2_holder *gh)
945 struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object;
946 int x, length = rgd->rd_length;
948 for (x = 0; x < length; x++) {
949 struct gfs2_bitmap *bi = rgd->rd_bits + x;
958 int gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset,
959 struct buffer_head *bh,
960 const struct gfs2_bitmap *bi, unsigned minlen, u64 *ptrimmed)
962 struct super_block *sb = sdp->sd_vfs;
963 struct block_device *bdev = sb->s_bdev;
964 const unsigned int sects_per_blk = sdp->sd_sb.sb_bsize /
965 bdev_logical_block_size(sb->s_bdev);
968 sector_t nr_sects = 0;
974 for (x = 0; x < bi->bi_len; x++) {
975 const u8 *clone = bi->bi_clone ? bi->bi_clone : bi->bi_bh->b_data;
976 clone += bi->bi_offset;
979 const u8 *orig = bh->b_data + bi->bi_offset + x;
980 diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1));
982 diff = ~(*clone | (*clone >> 1));
987 blk = offset + ((bi->bi_start + x) * GFS2_NBBY);
988 blk *= sects_per_blk; /* convert to sectors */
992 goto start_new_extent;
993 if ((start + nr_sects) != blk) {
994 if (nr_sects >= minlen) {
995 rv = blkdev_issue_discard(bdev,
1000 trimmed += nr_sects;
1006 nr_sects += sects_per_blk;
1009 blk += sects_per_blk;
1012 if (nr_sects >= minlen) {
1013 rv = blkdev_issue_discard(bdev, start, nr_sects, GFP_NOFS, 0);
1016 trimmed += nr_sects;
1019 *ptrimmed = trimmed;
1023 if (sdp->sd_args.ar_discard)
1024 fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv);
1025 sdp->sd_args.ar_discard = 0;
1030 * gfs2_fitrim - Generate discard requests for unused bits of the filesystem
1031 * @filp: Any file on the filesystem
1032 * @argp: Pointer to the arguments (also used to pass result)
1034 * Returns: 0 on success, otherwise error code
1037 int gfs2_fitrim(struct file *filp, void __user *argp)
1039 struct inode *inode = filp->f_dentry->d_inode;
1040 struct gfs2_sbd *sdp = GFS2_SB(inode);
1041 struct request_queue *q = bdev_get_queue(sdp->sd_vfs->s_bdev);
1042 struct buffer_head *bh;
1043 struct gfs2_rgrpd *rgd;
1044 struct gfs2_rgrpd *rgd_end;
1045 struct gfs2_holder gh;
1046 struct fstrim_range r;
1052 if (!capable(CAP_SYS_ADMIN))
1055 if (!blk_queue_discard(q))
1062 } else if (copy_from_user(&r, argp, sizeof(r)))
1065 ret = gfs2_rindex_update(sdp);
1069 rgd = gfs2_blk2rgrpd(sdp, r.start, 0);
1070 rgd_end = gfs2_blk2rgrpd(sdp, r.start + r.len, 0);
1074 ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1078 if (!(rgd->rd_flags & GFS2_RGF_TRIMMED)) {
1079 /* Trim each bitmap in the rgrp */
1080 for (x = 0; x < rgd->rd_length; x++) {
1081 struct gfs2_bitmap *bi = rgd->rd_bits + x;
1082 ret = gfs2_rgrp_send_discards(sdp, rgd->rd_data0, NULL, bi, r.minlen, &amt);
1084 gfs2_glock_dq_uninit(&gh);
1090 /* Mark rgrp as having been trimmed */
1091 ret = gfs2_trans_begin(sdp, RES_RG_HDR, 0);
1093 bh = rgd->rd_bits[0].bi_bh;
1094 rgd->rd_flags |= GFS2_RGF_TRIMMED;
1095 gfs2_trans_add_bh(rgd->rd_gl, bh, 1);
1096 gfs2_rgrp_out(rgd, bh->b_data);
1097 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, bh->b_data);
1098 gfs2_trans_end(sdp);
1101 gfs2_glock_dq_uninit(&gh);
1106 rgd = gfs2_rgrpd_get_next(rgd);
1110 r.len = trimmed << 9;
1111 if (argp && copy_to_user(argp, &r, sizeof(r)))
1118 * try_rgrp_fit - See if a given reservation will fit in a given RG
1122 * If there's room for the requested blocks to be allocated from the RG:
1124 * Returns: 1 on success (it fits), 0 on failure (it doesn't fit)
1127 static int try_rgrp_fit(const struct gfs2_rgrpd *rgd, const struct gfs2_inode *ip)
1129 const struct gfs2_blkreserv *rs = ip->i_res;
1131 if (rgd->rd_flags & (GFS2_RGF_NOALLOC | GFS2_RDF_ERROR))
1133 if (rgd->rd_free_clone >= rs->rs_requested)
1138 static inline u32 gfs2_bi2rgd_blk(struct gfs2_bitmap *bi, u32 blk)
1140 return (bi->bi_start * GFS2_NBBY) + blk;
1144 * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes
1146 * @last_unlinked: block address of the last dinode we unlinked
1147 * @skip: block address we should explicitly not unlink
1149 * Returns: 0 if no error
1150 * The inode, if one has been found, in inode.
1153 static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip)
1155 u32 goal = 0, block;
1157 struct gfs2_sbd *sdp = rgd->rd_sbd;
1158 struct gfs2_glock *gl;
1159 struct gfs2_inode *ip;
1162 struct gfs2_bitmap *bi;
1164 while (goal < rgd->rd_data) {
1165 down_write(&sdp->sd_log_flush_lock);
1166 block = rgblk_search(rgd, goal, GFS2_BLKST_UNLINKED, &bi);
1167 up_write(&sdp->sd_log_flush_lock);
1168 if (block == BFITNOENT)
1171 block = gfs2_bi2rgd_blk(bi, block);
1172 /* rgblk_search can return a block < goal, so we need to
1173 keep it marching forward. */
1174 no_addr = block + rgd->rd_data0;
1175 goal = max(block + 1, goal + 1);
1176 if (*last_unlinked != NO_BLOCK && no_addr <= *last_unlinked)
1178 if (no_addr == skip)
1180 *last_unlinked = no_addr;
1182 error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE, &gl);
1186 /* If the inode is already in cache, we can ignore it here
1187 * because the existing inode disposal code will deal with
1188 * it when all refs have gone away. Accessing gl_object like
1189 * this is not safe in general. Here it is ok because we do
1190 * not dereference the pointer, and we only need an approx
1191 * answer to whether it is NULL or not.
1195 if (ip || queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
1200 /* Limit reclaim to sensible number of tasks */
1201 if (found > NR_CPUS)
1205 rgd->rd_flags &= ~GFS2_RDF_CHECK;
1210 * gfs2_inplace_reserve - Reserve space in the filesystem
1211 * @ip: the inode to reserve space for
1212 * @requested: the number of blocks to be reserved
1217 int gfs2_inplace_reserve(struct gfs2_inode *ip, u32 requested)
1219 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1220 struct gfs2_rgrpd *rgd, *begin = NULL;
1221 struct gfs2_blkreserv *rs = ip->i_res;
1222 int error = 0, rg_locked, flags = LM_FLAG_TRY;
1223 u64 last_unlinked = NO_BLOCK;
1226 if (sdp->sd_args.ar_rgrplvb)
1229 rs->rs_requested = requested;
1230 if (gfs2_assert_warn(sdp, requested)) {
1235 if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, ip->i_goal))
1236 rgd = begin = ip->i_rgd;
1238 rgd = begin = gfs2_blk2rgrpd(sdp, ip->i_goal, 1);
1246 if (gfs2_glock_is_locked_by_me(rgd->rd_gl)) {
1250 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1251 flags, &rs->rs_rgd_gh);
1252 if (!error && sdp->sd_args.ar_rgrplvb) {
1253 error = update_rgrp_lvb(rgd);
1255 gfs2_glock_dq_uninit(&rs->rs_rgd_gh);
1262 if (try_rgrp_fit(rgd, ip)) {
1263 if (sdp->sd_args.ar_rgrplvb)
1264 gfs2_rgrp_bh_get(rgd);
1268 if (rgd->rd_flags & GFS2_RDF_CHECK) {
1269 if (sdp->sd_args.ar_rgrplvb)
1270 gfs2_rgrp_bh_get(rgd);
1271 try_rgrp_unlink(rgd, &last_unlinked,
1275 gfs2_glock_dq_uninit(&rs->rs_rgd_gh);
1278 rgd = gfs2_rgrpd_get_next(rgd);
1279 rgd = rgd ? : begin; /* if NULL, wrap */
1280 if (rgd != begin) /* If we didn't wrap */
1283 flags &= ~LM_FLAG_TRY;
1285 /* Check that fs hasn't grown if writing to rindex */
1286 if (ip == GFS2_I(sdp->sd_rindex) &&
1287 !sdp->sd_rindex_uptodate) {
1288 error = gfs2_ri_update(ip);
1291 } else if (loops == 2)
1292 /* Flushing the log may release space */
1293 gfs2_log_flush(sdp, NULL);
1303 rs->rs_requested = 0;
1308 * gfs2_inplace_release - release an inplace reservation
1309 * @ip: the inode the reservation was taken out on
1311 * Release a reservation made by gfs2_inplace_reserve().
1314 void gfs2_inplace_release(struct gfs2_inode *ip)
1316 struct gfs2_blkreserv *rs = ip->i_res;
1318 if (rs->rs_rgd_gh.gh_gl)
1319 gfs2_glock_dq_uninit(&rs->rs_rgd_gh);
1320 rs->rs_requested = 0;
1324 * gfs2_get_block_type - Check a block in a RG is of given type
1325 * @rgd: the resource group holding the block
1326 * @block: the block number
1328 * Returns: The block type (GFS2_BLKST_*)
1331 static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block)
1333 struct gfs2_bitmap *bi = NULL;
1334 u32 length, rgrp_block, buf_block;
1338 length = rgd->rd_length;
1339 rgrp_block = block - rgd->rd_data0;
1341 for (buf = 0; buf < length; buf++) {
1342 bi = rgd->rd_bits + buf;
1343 if (rgrp_block < (bi->bi_start + bi->bi_len) * GFS2_NBBY)
1347 gfs2_assert(rgd->rd_sbd, buf < length);
1348 buf_block = rgrp_block - bi->bi_start * GFS2_NBBY;
1350 type = gfs2_testbit(rgd, bi->bi_bh->b_data + bi->bi_offset,
1351 bi->bi_len, buf_block);
1357 * rgblk_search - find a block in @state
1358 * @rgd: the resource group descriptor
1359 * @goal: the goal block within the RG (start here to search for avail block)
1360 * @state: GFS2_BLKST_XXX the before-allocation state to find
1361 * @rbi: address of the pointer to the bitmap containing the block found
1363 * Walk rgrp's bitmap to find bits that represent a block in @state.
1365 * This function never fails, because we wouldn't call it unless we
1366 * know (from reservation results, etc.) that a block is available.
1368 * Scope of @goal is just within rgrp, not the whole filesystem.
1369 * Scope of @returned block is just within bitmap, not the whole filesystem.
1371 * Returns: the block number found relative to the bitmap rbi
1374 static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal, unsigned char state,
1375 struct gfs2_bitmap **rbi)
1377 struct gfs2_bitmap *bi = NULL;
1378 const u32 length = rgd->rd_length;
1379 u32 biblk = BFITNOENT;
1380 unsigned int buf, x;
1381 const u8 *buffer = NULL;
1384 /* Find bitmap block that contains bits for goal block */
1385 for (buf = 0; buf < length; buf++) {
1386 bi = rgd->rd_bits + buf;
1387 /* Convert scope of "goal" from rgrp-wide to within found bit block */
1388 if (goal < (bi->bi_start + bi->bi_len) * GFS2_NBBY) {
1389 goal -= bi->bi_start * GFS2_NBBY;
1397 /* Search (up to entire) bitmap in this rgrp for allocatable block.
1398 "x <= length", instead of "x < length", because we typically start
1399 the search in the middle of a bit block, but if we can't find an
1400 allocatable block anywhere else, we want to be able wrap around and
1401 search in the first part of our first-searched bit block. */
1402 for (x = 0; x <= length; x++) {
1403 bi = rgd->rd_bits + buf;
1405 if (test_bit(GBF_FULL, &bi->bi_flags) &&
1406 (state == GFS2_BLKST_FREE))
1409 /* The GFS2_BLKST_UNLINKED state doesn't apply to the clone
1410 bitmaps, so we must search the originals for that. */
1411 buffer = bi->bi_bh->b_data + bi->bi_offset;
1412 WARN_ON(!buffer_uptodate(bi->bi_bh));
1413 if (state != GFS2_BLKST_UNLINKED && bi->bi_clone)
1414 buffer = bi->bi_clone + bi->bi_offset;
1416 biblk = gfs2_bitfit(buffer, bi->bi_len, goal, state);
1417 if (biblk != BFITNOENT)
1420 if ((goal == 0) && (state == GFS2_BLKST_FREE))
1421 set_bit(GBF_FULL, &bi->bi_flags);
1423 /* Try next bitmap block (wrap back to rgrp header if at end) */
1430 if (biblk != BFITNOENT)
1437 * gfs2_alloc_extent - allocate an extent from a given bitmap
1438 * @rgd: the resource group descriptor
1439 * @bi: the bitmap within the rgrp
1440 * @blk: the block within the bitmap
1441 * @dinode: TRUE if the first block we allocate is for a dinode
1442 * @n: The extent length
1444 * Add the found bitmap buffer to the transaction.
1445 * Set the found bits to @new_state to change block's allocation state.
1446 * Returns: starting block number of the extent (fs scope)
1448 static u64 gfs2_alloc_extent(struct gfs2_rgrpd *rgd, struct gfs2_bitmap *bi,
1449 u32 blk, bool dinode, unsigned int *n)
1451 const unsigned int elen = *n;
1453 const u8 *buffer = NULL;
1456 buffer = bi->bi_bh->b_data + bi->bi_offset;
1457 gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1);
1458 gfs2_setbit(rgd, bi->bi_clone, bi, blk,
1459 dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED);
1464 if (goal >= (bi->bi_len * GFS2_NBBY))
1466 if (gfs2_testbit(rgd, buffer, bi->bi_len, goal) !=
1469 gfs2_setbit(rgd, bi->bi_clone, bi, goal, GFS2_BLKST_USED);
1472 blk = gfs2_bi2rgd_blk(bi, blk);
1473 rgd->rd_last_alloc = blk + *n - 1;
1474 return rgd->rd_data0 + blk;
1478 * rgblk_free - Change alloc state of given block(s)
1479 * @sdp: the filesystem
1480 * @bstart: the start of a run of blocks to free
1481 * @blen: the length of the block run (all must lie within ONE RG!)
1482 * @new_state: GFS2_BLKST_XXX the after-allocation block state
1484 * Returns: Resource group containing the block(s)
1487 static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart,
1488 u32 blen, unsigned char new_state)
1490 struct gfs2_rgrpd *rgd;
1491 struct gfs2_bitmap *bi = NULL;
1492 u32 length, rgrp_blk, buf_blk;
1495 rgd = gfs2_blk2rgrpd(sdp, bstart, 1);
1497 if (gfs2_consist(sdp))
1498 fs_err(sdp, "block = %llu\n", (unsigned long long)bstart);
1502 length = rgd->rd_length;
1504 rgrp_blk = bstart - rgd->rd_data0;
1507 for (buf = 0; buf < length; buf++) {
1508 bi = rgd->rd_bits + buf;
1509 if (rgrp_blk < (bi->bi_start + bi->bi_len) * GFS2_NBBY)
1513 gfs2_assert(rgd->rd_sbd, buf < length);
1515 buf_blk = rgrp_blk - bi->bi_start * GFS2_NBBY;
1518 if (!bi->bi_clone) {
1519 bi->bi_clone = kmalloc(bi->bi_bh->b_size,
1520 GFP_NOFS | __GFP_NOFAIL);
1521 memcpy(bi->bi_clone + bi->bi_offset,
1522 bi->bi_bh->b_data + bi->bi_offset,
1525 gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1);
1526 gfs2_setbit(rgd, NULL, bi, buf_blk, new_state);
1533 * gfs2_rgrp_dump - print out an rgrp
1534 * @seq: The iterator
1535 * @gl: The glock in question
1539 int gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl)
1541 const struct gfs2_rgrpd *rgd = gl->gl_object;
1544 gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u\n",
1545 (unsigned long long)rgd->rd_addr, rgd->rd_flags,
1546 rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes);
1550 static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd)
1552 struct gfs2_sbd *sdp = rgd->rd_sbd;
1553 fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n",
1554 (unsigned long long)rgd->rd_addr);
1555 fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n");
1556 gfs2_rgrp_dump(NULL, rgd->rd_gl);
1557 rgd->rd_flags |= GFS2_RDF_ERROR;
1561 * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode
1562 * @ip: the inode to allocate the block for
1563 * @bn: Used to return the starting block number
1564 * @ndata: requested number of blocks/extent length (value/result)
1565 * @dinode: 1 if we're allocating a dinode block, else 0
1566 * @generation: the generation number of the inode
1568 * Returns: 0 or error
1571 int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *nblocks,
1572 bool dinode, u64 *generation)
1574 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1575 struct buffer_head *dibh;
1576 struct gfs2_rgrpd *rgd;
1578 u32 goal, blk; /* block, within the rgrp scope */
1579 u64 block; /* block, within the file system scope */
1581 struct gfs2_bitmap *bi;
1583 /* Only happens if there is a bug in gfs2, return something distinctive
1584 * to ensure that it is noticed.
1586 if (ip->i_res->rs_requested == 0)
1591 if (!dinode && rgrp_contains_block(rgd, ip->i_goal))
1592 goal = ip->i_goal - rgd->rd_data0;
1594 goal = rgd->rd_last_alloc;
1596 blk = rgblk_search(rgd, goal, GFS2_BLKST_FREE, &bi);
1598 /* Since all blocks are reserved in advance, this shouldn't happen */
1599 if (blk == BFITNOENT)
1602 block = gfs2_alloc_extent(rgd, bi, blk, dinode, nblocks);
1608 ip->i_goal = block + ndata - 1;
1609 error = gfs2_meta_inode_buffer(ip, &dibh);
1611 struct gfs2_dinode *di =
1612 (struct gfs2_dinode *)dibh->b_data;
1613 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1614 di->di_goal_meta = di->di_goal_data =
1615 cpu_to_be64(ip->i_goal);
1619 if (rgd->rd_free < *nblocks)
1622 rgd->rd_free -= *nblocks;
1625 *generation = rgd->rd_igeneration++;
1626 if (*generation == 0)
1627 *generation = rgd->rd_igeneration++;
1630 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1631 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1632 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data);
1634 gfs2_statfs_change(sdp, 0, -(s64)*nblocks, dinode ? 1 : 0);
1636 gfs2_trans_add_unrevoke(sdp, block, 1);
1639 * This needs reviewing to see why we cannot do the quota change
1640 * at this point in the dinode case.
1643 gfs2_quota_change(ip, ndata, ip->i_inode.i_uid,
1646 rgd->rd_free_clone -= *nblocks;
1647 trace_gfs2_block_alloc(ip, rgd, block, *nblocks,
1648 dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED);
1653 gfs2_rgrp_error(rgd);
1658 * __gfs2_free_blocks - free a contiguous run of block(s)
1659 * @ip: the inode these blocks are being freed from
1660 * @bstart: first block of a run of contiguous blocks
1661 * @blen: the length of the block run
1662 * @meta: 1 if the blocks represent metadata
1666 void __gfs2_free_blocks(struct gfs2_inode *ip, u64 bstart, u32 blen, int meta)
1668 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1669 struct gfs2_rgrpd *rgd;
1671 rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE);
1674 trace_gfs2_block_alloc(ip, rgd, bstart, blen, GFS2_BLKST_FREE);
1675 rgd->rd_free += blen;
1676 rgd->rd_flags &= ~GFS2_RGF_TRIMMED;
1677 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1678 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1679 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data);
1681 /* Directories keep their data in the metadata address space */
1682 if (meta || ip->i_depth)
1683 gfs2_meta_wipe(ip, bstart, blen);
1687 * gfs2_free_meta - free a contiguous run of data block(s)
1688 * @ip: the inode these blocks are being freed from
1689 * @bstart: first block of a run of contiguous blocks
1690 * @blen: the length of the block run
1694 void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen)
1696 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1698 __gfs2_free_blocks(ip, bstart, blen, 1);
1699 gfs2_statfs_change(sdp, 0, +blen, 0);
1700 gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid);
1703 void gfs2_unlink_di(struct inode *inode)
1705 struct gfs2_inode *ip = GFS2_I(inode);
1706 struct gfs2_sbd *sdp = GFS2_SB(inode);
1707 struct gfs2_rgrpd *rgd;
1708 u64 blkno = ip->i_no_addr;
1710 rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED);
1713 trace_gfs2_block_alloc(ip, rgd, blkno, 1, GFS2_BLKST_UNLINKED);
1714 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1715 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1716 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data);
1717 update_rgrp_lvb_unlinked(rgd, 1);
1720 static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno)
1722 struct gfs2_sbd *sdp = rgd->rd_sbd;
1723 struct gfs2_rgrpd *tmp_rgd;
1725 tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE);
1728 gfs2_assert_withdraw(sdp, rgd == tmp_rgd);
1730 if (!rgd->rd_dinodes)
1731 gfs2_consist_rgrpd(rgd);
1735 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1736 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1737 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data);
1738 update_rgrp_lvb_unlinked(rgd, -1);
1740 gfs2_statfs_change(sdp, 0, +1, -1);
1744 void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip)
1746 gfs2_free_uninit_di(rgd, ip->i_no_addr);
1747 trace_gfs2_block_alloc(ip, rgd, ip->i_no_addr, 1, GFS2_BLKST_FREE);
1748 gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid);
1749 gfs2_meta_wipe(ip, ip->i_no_addr, 1);
1753 * gfs2_check_blk_type - Check the type of a block
1754 * @sdp: The superblock
1755 * @no_addr: The block number to check
1756 * @type: The block type we are looking for
1758 * Returns: 0 if the block type matches the expected type
1759 * -ESTALE if it doesn't match
1760 * or -ve errno if something went wrong while checking
1763 int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type)
1765 struct gfs2_rgrpd *rgd;
1766 struct gfs2_holder rgd_gh;
1767 int error = -EINVAL;
1769 rgd = gfs2_blk2rgrpd(sdp, no_addr, 1);
1773 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh);
1777 if (gfs2_get_block_type(rgd, no_addr) != type)
1780 gfs2_glock_dq_uninit(&rgd_gh);
1786 * gfs2_rlist_add - add a RG to a list of RGs
1788 * @rlist: the list of resource groups
1791 * Figure out what RG a block belongs to and add that RG to the list
1793 * FIXME: Don't use NOFAIL
1797 void gfs2_rlist_add(struct gfs2_inode *ip, struct gfs2_rgrp_list *rlist,
1800 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1801 struct gfs2_rgrpd *rgd;
1802 struct gfs2_rgrpd **tmp;
1803 unsigned int new_space;
1806 if (gfs2_assert_warn(sdp, !rlist->rl_ghs))
1809 if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, block))
1812 rgd = gfs2_blk2rgrpd(sdp, block, 1);
1814 fs_err(sdp, "rlist_add: no rgrp for block %llu\n", (unsigned long long)block);
1819 for (x = 0; x < rlist->rl_rgrps; x++)
1820 if (rlist->rl_rgd[x] == rgd)
1823 if (rlist->rl_rgrps == rlist->rl_space) {
1824 new_space = rlist->rl_space + 10;
1826 tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *),
1827 GFP_NOFS | __GFP_NOFAIL);
1829 if (rlist->rl_rgd) {
1830 memcpy(tmp, rlist->rl_rgd,
1831 rlist->rl_space * sizeof(struct gfs2_rgrpd *));
1832 kfree(rlist->rl_rgd);
1835 rlist->rl_space = new_space;
1836 rlist->rl_rgd = tmp;
1839 rlist->rl_rgd[rlist->rl_rgrps++] = rgd;
1843 * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate
1844 * and initialize an array of glock holders for them
1845 * @rlist: the list of resource groups
1846 * @state: the lock state to acquire the RG lock in
1848 * FIXME: Don't use NOFAIL
1852 void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state)
1856 rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder),
1857 GFP_NOFS | __GFP_NOFAIL);
1858 for (x = 0; x < rlist->rl_rgrps; x++)
1859 gfs2_holder_init(rlist->rl_rgd[x]->rd_gl,
1865 * gfs2_rlist_free - free a resource group list
1866 * @list: the list of resource groups
1870 void gfs2_rlist_free(struct gfs2_rgrp_list *rlist)
1874 kfree(rlist->rl_rgd);
1876 if (rlist->rl_ghs) {
1877 for (x = 0; x < rlist->rl_rgrps; x++)
1878 gfs2_holder_uninit(&rlist->rl_ghs[x]);
1879 kfree(rlist->rl_ghs);