mm-only debug patch...
[mmotm.git] / fs / gfs2 / rgrp.c
blob15332af34d991e891a1dd096f1daa978410fc100
1 /*
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.
8 */
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/fs.h>
15 #include <linux/gfs2_ondisk.h>
16 #include <linux/prefetch.h>
17 #include <linux/blkdev.h>
18 #include <trace/events/gfs2.h>
20 #include "gfs2.h"
21 #include "incore.h"
22 #include "glock.h"
23 #include "glops.h"
24 #include "lops.h"
25 #include "meta_io.h"
26 #include "quota.h"
27 #include "rgrp.h"
28 #include "super.h"
29 #include "trans.h"
30 #include "util.h"
31 #include "log.h"
32 #include "inode.h"
34 #define BFITNOENT ((u32)~0)
35 #define NO_BLOCK ((u64)~0)
37 #if BITS_PER_LONG == 32
38 #define LBITMASK (0x55555555UL)
39 #define LBITSKIP55 (0x55555555UL)
40 #define LBITSKIP00 (0x00000000UL)
41 #else
42 #define LBITMASK (0x5555555555555555UL)
43 #define LBITSKIP55 (0x5555555555555555UL)
44 #define LBITSKIP00 (0x0000000000000000UL)
45 #endif
48 * These routines are used by the resource group routines (rgrp.c)
49 * to keep track of block allocation. Each block is represented by two
50 * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks.
52 * 0 = Free
53 * 1 = Used (not metadata)
54 * 2 = Unlinked (still in use) inode
55 * 3 = Used (metadata)
58 static const char valid_change[16] = {
59 /* current */
60 /* n */ 0, 1, 1, 1,
61 /* e */ 1, 0, 0, 0,
62 /* w */ 0, 0, 0, 1,
63 1, 0, 0, 0
66 static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal,
67 unsigned char old_state, unsigned char new_state,
68 unsigned int *n);
70 /**
71 * gfs2_setbit - Set a bit in the bitmaps
72 * @buffer: the buffer that holds the bitmaps
73 * @buflen: the length (in bytes) of the buffer
74 * @block: the block to set
75 * @new_state: the new state of the block
79 static inline void gfs2_setbit(struct gfs2_rgrpd *rgd, unsigned char *buf1,
80 unsigned char *buf2, unsigned int offset,
81 unsigned int buflen, u32 block,
82 unsigned char new_state)
84 unsigned char *byte1, *byte2, *end, cur_state;
85 const unsigned int bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE;
87 byte1 = buf1 + offset + (block / GFS2_NBBY);
88 end = buf1 + offset + buflen;
90 BUG_ON(byte1 >= end);
92 cur_state = (*byte1 >> bit) & GFS2_BIT_MASK;
94 if (unlikely(!valid_change[new_state * 4 + cur_state])) {
95 gfs2_consist_rgrpd(rgd);
96 return;
98 *byte1 ^= (cur_state ^ new_state) << bit;
100 if (buf2) {
101 byte2 = buf2 + offset + (block / GFS2_NBBY);
102 cur_state = (*byte2 >> bit) & GFS2_BIT_MASK;
103 *byte2 ^= (cur_state ^ new_state) << bit;
108 * gfs2_testbit - test a bit in the bitmaps
109 * @buffer: the buffer that holds the bitmaps
110 * @buflen: the length (in bytes) of the buffer
111 * @block: the block to read
115 static inline unsigned char gfs2_testbit(struct gfs2_rgrpd *rgd,
116 const unsigned char *buffer,
117 unsigned int buflen, u32 block)
119 const unsigned char *byte, *end;
120 unsigned char cur_state;
121 unsigned int bit;
123 byte = buffer + (block / GFS2_NBBY);
124 bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE;
125 end = buffer + buflen;
127 gfs2_assert(rgd->rd_sbd, byte < end);
129 cur_state = (*byte >> bit) & GFS2_BIT_MASK;
131 return cur_state;
135 * gfs2_bit_search
136 * @ptr: Pointer to bitmap data
137 * @mask: Mask to use (normally 0x55555.... but adjusted for search start)
138 * @state: The state we are searching for
140 * We xor the bitmap data with a patter which is the bitwise opposite
141 * of what we are looking for, this gives rise to a pattern of ones
142 * wherever there is a match. Since we have two bits per entry, we
143 * take this pattern, shift it down by one place and then and it with
144 * the original. All the even bit positions (0,2,4, etc) then represent
145 * successful matches, so we mask with 0x55555..... to remove the unwanted
146 * odd bit positions.
148 * This allows searching of a whole u64 at once (32 blocks) with a
149 * single test (on 64 bit arches).
152 static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state)
154 u64 tmp;
155 static const u64 search[] = {
156 [0] = 0xffffffffffffffffULL,
157 [1] = 0xaaaaaaaaaaaaaaaaULL,
158 [2] = 0x5555555555555555ULL,
159 [3] = 0x0000000000000000ULL,
161 tmp = le64_to_cpu(*ptr) ^ search[state];
162 tmp &= (tmp >> 1);
163 tmp &= mask;
164 return tmp;
168 * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing
169 * a block in a given allocation state.
170 * @buffer: the buffer that holds the bitmaps
171 * @len: the length (in bytes) of the buffer
172 * @goal: start search at this block's bit-pair (within @buffer)
173 * @state: GFS2_BLKST_XXX the state of the block we're looking for.
175 * Scope of @goal and returned block number is only within this bitmap buffer,
176 * not entire rgrp or filesystem. @buffer will be offset from the actual
177 * beginning of a bitmap block buffer, skipping any header structures, but
178 * headers are always a multiple of 64 bits long so that the buffer is
179 * always aligned to a 64 bit boundary.
181 * The size of the buffer is in bytes, but is it assumed that it is
182 * always ok to read a complete multiple of 64 bits at the end
183 * of the block in case the end is no aligned to a natural boundary.
185 * Return: the block number (bitmap buffer scope) that was found
188 static u32 gfs2_bitfit(const u8 *buf, const unsigned int len,
189 u32 goal, u8 state)
191 u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1);
192 const __le64 *ptr = ((__le64 *)buf) + (goal >> 5);
193 const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64)));
194 u64 tmp;
195 u64 mask = 0x5555555555555555ULL;
196 u32 bit;
198 BUG_ON(state > 3);
200 /* Mask off bits we don't care about at the start of the search */
201 mask <<= spoint;
202 tmp = gfs2_bit_search(ptr, mask, state);
203 ptr++;
204 while(tmp == 0 && ptr < end) {
205 tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state);
206 ptr++;
208 /* Mask off any bits which are more than len bytes from the start */
209 if (ptr == end && (len & (sizeof(u64) - 1)))
210 tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1))));
211 /* Didn't find anything, so return */
212 if (tmp == 0)
213 return BFITNOENT;
214 ptr--;
215 bit = __ffs64(tmp);
216 bit /= 2; /* two bits per entry in the bitmap */
217 return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit;
221 * gfs2_bitcount - count the number of bits in a certain state
222 * @buffer: the buffer that holds the bitmaps
223 * @buflen: the length (in bytes) of the buffer
224 * @state: the state of the block we're looking for
226 * Returns: The number of bits
229 static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer,
230 unsigned int buflen, u8 state)
232 const u8 *byte = buffer;
233 const u8 *end = buffer + buflen;
234 const u8 state1 = state << 2;
235 const u8 state2 = state << 4;
236 const u8 state3 = state << 6;
237 u32 count = 0;
239 for (; byte < end; byte++) {
240 if (((*byte) & 0x03) == state)
241 count++;
242 if (((*byte) & 0x0C) == state1)
243 count++;
244 if (((*byte) & 0x30) == state2)
245 count++;
246 if (((*byte) & 0xC0) == state3)
247 count++;
250 return count;
254 * gfs2_rgrp_verify - Verify that a resource group is consistent
255 * @sdp: the filesystem
256 * @rgd: the rgrp
260 void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd)
262 struct gfs2_sbd *sdp = rgd->rd_sbd;
263 struct gfs2_bitmap *bi = NULL;
264 u32 length = rgd->rd_length;
265 u32 count[4], tmp;
266 int buf, x;
268 memset(count, 0, 4 * sizeof(u32));
270 /* Count # blocks in each of 4 possible allocation states */
271 for (buf = 0; buf < length; buf++) {
272 bi = rgd->rd_bits + buf;
273 for (x = 0; x < 4; x++)
274 count[x] += gfs2_bitcount(rgd,
275 bi->bi_bh->b_data +
276 bi->bi_offset,
277 bi->bi_len, x);
280 if (count[0] != rgd->rd_free) {
281 if (gfs2_consist_rgrpd(rgd))
282 fs_err(sdp, "free data mismatch: %u != %u\n",
283 count[0], rgd->rd_free);
284 return;
287 tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes;
288 if (count[1] != tmp) {
289 if (gfs2_consist_rgrpd(rgd))
290 fs_err(sdp, "used data mismatch: %u != %u\n",
291 count[1], tmp);
292 return;
295 if (count[2] + count[3] != rgd->rd_dinodes) {
296 if (gfs2_consist_rgrpd(rgd))
297 fs_err(sdp, "used metadata mismatch: %u != %u\n",
298 count[2] + count[3], rgd->rd_dinodes);
299 return;
303 static inline int rgrp_contains_block(struct gfs2_rgrpd *rgd, u64 block)
305 u64 first = rgd->rd_data0;
306 u64 last = first + rgd->rd_data;
307 return first <= block && block < last;
311 * gfs2_blk2rgrpd - Find resource group for a given data/meta block number
312 * @sdp: The GFS2 superblock
313 * @n: The data block number
315 * Returns: The resource group, or NULL if not found
318 struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk)
320 struct gfs2_rgrpd *rgd;
322 spin_lock(&sdp->sd_rindex_spin);
324 list_for_each_entry(rgd, &sdp->sd_rindex_mru_list, rd_list_mru) {
325 if (rgrp_contains_block(rgd, blk)) {
326 list_move(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list);
327 spin_unlock(&sdp->sd_rindex_spin);
328 return rgd;
332 spin_unlock(&sdp->sd_rindex_spin);
334 return NULL;
338 * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem
339 * @sdp: The GFS2 superblock
341 * Returns: The first rgrp in the filesystem
344 struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp)
346 gfs2_assert(sdp, !list_empty(&sdp->sd_rindex_list));
347 return list_entry(sdp->sd_rindex_list.next, struct gfs2_rgrpd, rd_list);
351 * gfs2_rgrpd_get_next - get the next RG
352 * @rgd: A RG
354 * Returns: The next rgrp
357 struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd)
359 if (rgd->rd_list.next == &rgd->rd_sbd->sd_rindex_list)
360 return NULL;
361 return list_entry(rgd->rd_list.next, struct gfs2_rgrpd, rd_list);
364 static void clear_rgrpdi(struct gfs2_sbd *sdp)
366 struct list_head *head;
367 struct gfs2_rgrpd *rgd;
368 struct gfs2_glock *gl;
370 spin_lock(&sdp->sd_rindex_spin);
371 sdp->sd_rindex_forward = NULL;
372 spin_unlock(&sdp->sd_rindex_spin);
374 head = &sdp->sd_rindex_list;
375 while (!list_empty(head)) {
376 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_list);
377 gl = rgd->rd_gl;
379 list_del(&rgd->rd_list);
380 list_del(&rgd->rd_list_mru);
382 if (gl) {
383 gl->gl_object = NULL;
384 gfs2_glock_put(gl);
387 kfree(rgd->rd_bits);
388 kmem_cache_free(gfs2_rgrpd_cachep, rgd);
392 void gfs2_clear_rgrpd(struct gfs2_sbd *sdp)
394 mutex_lock(&sdp->sd_rindex_mutex);
395 clear_rgrpdi(sdp);
396 mutex_unlock(&sdp->sd_rindex_mutex);
399 static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd)
401 printk(KERN_INFO " ri_addr = %llu\n", (unsigned long long)rgd->rd_addr);
402 printk(KERN_INFO " ri_length = %u\n", rgd->rd_length);
403 printk(KERN_INFO " ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0);
404 printk(KERN_INFO " ri_data = %u\n", rgd->rd_data);
405 printk(KERN_INFO " ri_bitbytes = %u\n", rgd->rd_bitbytes);
409 * gfs2_compute_bitstructs - Compute the bitmap sizes
410 * @rgd: The resource group descriptor
412 * Calculates bitmap descriptors, one for each block that contains bitmap data
414 * Returns: errno
417 static int compute_bitstructs(struct gfs2_rgrpd *rgd)
419 struct gfs2_sbd *sdp = rgd->rd_sbd;
420 struct gfs2_bitmap *bi;
421 u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */
422 u32 bytes_left, bytes;
423 int x;
425 if (!length)
426 return -EINVAL;
428 rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS);
429 if (!rgd->rd_bits)
430 return -ENOMEM;
432 bytes_left = rgd->rd_bitbytes;
434 for (x = 0; x < length; x++) {
435 bi = rgd->rd_bits + x;
437 bi->bi_flags = 0;
438 /* small rgrp; bitmap stored completely in header block */
439 if (length == 1) {
440 bytes = bytes_left;
441 bi->bi_offset = sizeof(struct gfs2_rgrp);
442 bi->bi_start = 0;
443 bi->bi_len = bytes;
444 /* header block */
445 } else if (x == 0) {
446 bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp);
447 bi->bi_offset = sizeof(struct gfs2_rgrp);
448 bi->bi_start = 0;
449 bi->bi_len = bytes;
450 /* last block */
451 } else if (x + 1 == length) {
452 bytes = bytes_left;
453 bi->bi_offset = sizeof(struct gfs2_meta_header);
454 bi->bi_start = rgd->rd_bitbytes - bytes_left;
455 bi->bi_len = bytes;
456 /* other blocks */
457 } else {
458 bytes = sdp->sd_sb.sb_bsize -
459 sizeof(struct gfs2_meta_header);
460 bi->bi_offset = sizeof(struct gfs2_meta_header);
461 bi->bi_start = rgd->rd_bitbytes - bytes_left;
462 bi->bi_len = bytes;
465 bytes_left -= bytes;
468 if (bytes_left) {
469 gfs2_consist_rgrpd(rgd);
470 return -EIO;
472 bi = rgd->rd_bits + (length - 1);
473 if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) {
474 if (gfs2_consist_rgrpd(rgd)) {
475 gfs2_rindex_print(rgd);
476 fs_err(sdp, "start=%u len=%u offset=%u\n",
477 bi->bi_start, bi->bi_len, bi->bi_offset);
479 return -EIO;
482 return 0;
486 * gfs2_ri_total - Total up the file system space, according to the rindex.
489 u64 gfs2_ri_total(struct gfs2_sbd *sdp)
491 u64 total_data = 0;
492 struct inode *inode = sdp->sd_rindex;
493 struct gfs2_inode *ip = GFS2_I(inode);
494 char buf[sizeof(struct gfs2_rindex)];
495 struct file_ra_state ra_state;
496 int error, rgrps;
498 mutex_lock(&sdp->sd_rindex_mutex);
499 file_ra_state_init(&ra_state, inode->i_mapping);
500 for (rgrps = 0;; rgrps++) {
501 loff_t pos = rgrps * sizeof(struct gfs2_rindex);
503 if (pos + sizeof(struct gfs2_rindex) >= ip->i_disksize)
504 break;
505 error = gfs2_internal_read(ip, &ra_state, buf, &pos,
506 sizeof(struct gfs2_rindex));
507 if (error != sizeof(struct gfs2_rindex))
508 break;
509 total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data);
511 mutex_unlock(&sdp->sd_rindex_mutex);
512 return total_data;
515 static void gfs2_rindex_in(struct gfs2_rgrpd *rgd, const void *buf)
517 const struct gfs2_rindex *str = buf;
519 rgd->rd_addr = be64_to_cpu(str->ri_addr);
520 rgd->rd_length = be32_to_cpu(str->ri_length);
521 rgd->rd_data0 = be64_to_cpu(str->ri_data0);
522 rgd->rd_data = be32_to_cpu(str->ri_data);
523 rgd->rd_bitbytes = be32_to_cpu(str->ri_bitbytes);
527 * read_rindex_entry - Pull in a new resource index entry from the disk
528 * @gl: The glock covering the rindex inode
530 * Returns: 0 on success, error code otherwise
533 static int read_rindex_entry(struct gfs2_inode *ip,
534 struct file_ra_state *ra_state)
536 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
537 loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex);
538 char buf[sizeof(struct gfs2_rindex)];
539 int error;
540 struct gfs2_rgrpd *rgd;
542 error = gfs2_internal_read(ip, ra_state, buf, &pos,
543 sizeof(struct gfs2_rindex));
544 if (!error)
545 return 0;
546 if (error != sizeof(struct gfs2_rindex)) {
547 if (error > 0)
548 error = -EIO;
549 return error;
552 rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS);
553 error = -ENOMEM;
554 if (!rgd)
555 return error;
557 mutex_init(&rgd->rd_mutex);
558 lops_init_le(&rgd->rd_le, &gfs2_rg_lops);
559 rgd->rd_sbd = sdp;
561 list_add_tail(&rgd->rd_list, &sdp->sd_rindex_list);
562 list_add_tail(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list);
564 gfs2_rindex_in(rgd, buf);
565 error = compute_bitstructs(rgd);
566 if (error)
567 return error;
569 error = gfs2_glock_get(sdp, rgd->rd_addr,
570 &gfs2_rgrp_glops, CREATE, &rgd->rd_gl);
571 if (error)
572 return error;
574 rgd->rd_gl->gl_object = rgd;
575 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
576 return error;
580 * gfs2_ri_update - Pull in a new resource index from the disk
581 * @ip: pointer to the rindex inode
583 * Returns: 0 on successful update, error code otherwise
586 static int gfs2_ri_update(struct gfs2_inode *ip)
588 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
589 struct inode *inode = &ip->i_inode;
590 struct file_ra_state ra_state;
591 u64 rgrp_count = ip->i_disksize;
592 int error;
594 if (do_div(rgrp_count, sizeof(struct gfs2_rindex))) {
595 gfs2_consist_inode(ip);
596 return -EIO;
599 clear_rgrpdi(sdp);
601 file_ra_state_init(&ra_state, inode->i_mapping);
602 for (sdp->sd_rgrps = 0; sdp->sd_rgrps < rgrp_count; sdp->sd_rgrps++) {
603 error = read_rindex_entry(ip, &ra_state);
604 if (error) {
605 clear_rgrpdi(sdp);
606 return error;
610 sdp->sd_rindex_uptodate = 1;
611 return 0;
615 * gfs2_ri_update_special - Pull in a new resource index from the disk
617 * This is a special version that's safe to call from gfs2_inplace_reserve_i.
618 * In this case we know that we don't have any resource groups in memory yet.
620 * @ip: pointer to the rindex inode
622 * Returns: 0 on successful update, error code otherwise
624 static int gfs2_ri_update_special(struct gfs2_inode *ip)
626 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
627 struct inode *inode = &ip->i_inode;
628 struct file_ra_state ra_state;
629 int error;
631 file_ra_state_init(&ra_state, inode->i_mapping);
632 for (sdp->sd_rgrps = 0;; sdp->sd_rgrps++) {
633 /* Ignore partials */
634 if ((sdp->sd_rgrps + 1) * sizeof(struct gfs2_rindex) >
635 ip->i_disksize)
636 break;
637 error = read_rindex_entry(ip, &ra_state);
638 if (error) {
639 clear_rgrpdi(sdp);
640 return error;
644 sdp->sd_rindex_uptodate = 1;
645 return 0;
649 * gfs2_rindex_hold - Grab a lock on the rindex
650 * @sdp: The GFS2 superblock
651 * @ri_gh: the glock holder
653 * We grab a lock on the rindex inode to make sure that it doesn't
654 * change whilst we are performing an operation. We keep this lock
655 * for quite long periods of time compared to other locks. This
656 * doesn't matter, since it is shared and it is very, very rarely
657 * accessed in the exclusive mode (i.e. only when expanding the filesystem).
659 * This makes sure that we're using the latest copy of the resource index
660 * special file, which might have been updated if someone expanded the
661 * filesystem (via gfs2_grow utility), which adds new resource groups.
663 * Returns: 0 on success, error code otherwise
666 int gfs2_rindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ri_gh)
668 struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex);
669 struct gfs2_glock *gl = ip->i_gl;
670 int error;
672 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, ri_gh);
673 if (error)
674 return error;
676 /* Read new copy from disk if we don't have the latest */
677 if (!sdp->sd_rindex_uptodate) {
678 mutex_lock(&sdp->sd_rindex_mutex);
679 if (!sdp->sd_rindex_uptodate) {
680 error = gfs2_ri_update(ip);
681 if (error)
682 gfs2_glock_dq_uninit(ri_gh);
684 mutex_unlock(&sdp->sd_rindex_mutex);
687 return error;
690 static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf)
692 const struct gfs2_rgrp *str = buf;
693 u32 rg_flags;
695 rg_flags = be32_to_cpu(str->rg_flags);
696 rg_flags &= ~GFS2_RDF_MASK;
697 rgd->rd_flags &= GFS2_RDF_MASK;
698 rgd->rd_flags |= rg_flags;
699 rgd->rd_free = be32_to_cpu(str->rg_free);
700 rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes);
701 rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration);
704 static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf)
706 struct gfs2_rgrp *str = buf;
708 str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK);
709 str->rg_free = cpu_to_be32(rgd->rd_free);
710 str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes);
711 str->__pad = cpu_to_be32(0);
712 str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration);
713 memset(&str->rg_reserved, 0, sizeof(str->rg_reserved));
717 * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps
718 * @rgd: the struct gfs2_rgrpd describing the RG to read in
720 * Read in all of a Resource Group's header and bitmap blocks.
721 * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps.
723 * Returns: errno
726 int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd)
728 struct gfs2_sbd *sdp = rgd->rd_sbd;
729 struct gfs2_glock *gl = rgd->rd_gl;
730 unsigned int length = rgd->rd_length;
731 struct gfs2_bitmap *bi;
732 unsigned int x, y;
733 int error;
735 mutex_lock(&rgd->rd_mutex);
737 spin_lock(&sdp->sd_rindex_spin);
738 if (rgd->rd_bh_count) {
739 rgd->rd_bh_count++;
740 spin_unlock(&sdp->sd_rindex_spin);
741 mutex_unlock(&rgd->rd_mutex);
742 return 0;
744 spin_unlock(&sdp->sd_rindex_spin);
746 for (x = 0; x < length; x++) {
747 bi = rgd->rd_bits + x;
748 error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, &bi->bi_bh);
749 if (error)
750 goto fail;
753 for (y = length; y--;) {
754 bi = rgd->rd_bits + y;
755 error = gfs2_meta_wait(sdp, bi->bi_bh);
756 if (error)
757 goto fail;
758 if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB :
759 GFS2_METATYPE_RG)) {
760 error = -EIO;
761 goto fail;
765 if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) {
766 for (x = 0; x < length; x++)
767 clear_bit(GBF_FULL, &rgd->rd_bits[x].bi_flags);
768 gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data);
769 rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK);
772 spin_lock(&sdp->sd_rindex_spin);
773 rgd->rd_free_clone = rgd->rd_free;
774 rgd->rd_bh_count++;
775 spin_unlock(&sdp->sd_rindex_spin);
777 mutex_unlock(&rgd->rd_mutex);
779 return 0;
781 fail:
782 while (x--) {
783 bi = rgd->rd_bits + x;
784 brelse(bi->bi_bh);
785 bi->bi_bh = NULL;
786 gfs2_assert_warn(sdp, !bi->bi_clone);
788 mutex_unlock(&rgd->rd_mutex);
790 return error;
793 void gfs2_rgrp_bh_hold(struct gfs2_rgrpd *rgd)
795 struct gfs2_sbd *sdp = rgd->rd_sbd;
797 spin_lock(&sdp->sd_rindex_spin);
798 gfs2_assert_warn(rgd->rd_sbd, rgd->rd_bh_count);
799 rgd->rd_bh_count++;
800 spin_unlock(&sdp->sd_rindex_spin);
804 * gfs2_rgrp_bh_put - Release RG bitmaps read in with gfs2_rgrp_bh_get()
805 * @rgd: the struct gfs2_rgrpd describing the RG to read in
809 void gfs2_rgrp_bh_put(struct gfs2_rgrpd *rgd)
811 struct gfs2_sbd *sdp = rgd->rd_sbd;
812 int x, length = rgd->rd_length;
814 spin_lock(&sdp->sd_rindex_spin);
815 gfs2_assert_warn(rgd->rd_sbd, rgd->rd_bh_count);
816 if (--rgd->rd_bh_count) {
817 spin_unlock(&sdp->sd_rindex_spin);
818 return;
821 for (x = 0; x < length; x++) {
822 struct gfs2_bitmap *bi = rgd->rd_bits + x;
823 kfree(bi->bi_clone);
824 bi->bi_clone = NULL;
825 brelse(bi->bi_bh);
826 bi->bi_bh = NULL;
829 spin_unlock(&sdp->sd_rindex_spin);
832 static void gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset,
833 const struct gfs2_bitmap *bi)
835 struct super_block *sb = sdp->sd_vfs;
836 struct block_device *bdev = sb->s_bdev;
837 const unsigned int sects_per_blk = sdp->sd_sb.sb_bsize /
838 bdev_logical_block_size(sb->s_bdev);
839 u64 blk;
840 sector_t start = 0;
841 sector_t nr_sects = 0;
842 int rv;
843 unsigned int x;
845 for (x = 0; x < bi->bi_len; x++) {
846 const u8 *orig = bi->bi_bh->b_data + bi->bi_offset + x;
847 const u8 *clone = bi->bi_clone + bi->bi_offset + x;
848 u8 diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1));
849 diff &= 0x55;
850 if (diff == 0)
851 continue;
852 blk = offset + ((bi->bi_start + x) * GFS2_NBBY);
853 blk *= sects_per_blk; /* convert to sectors */
854 while(diff) {
855 if (diff & 1) {
856 if (nr_sects == 0)
857 goto start_new_extent;
858 if ((start + nr_sects) != blk) {
859 rv = blkdev_issue_discard(bdev, start,
860 nr_sects, GFP_NOFS,
861 DISCARD_FL_BARRIER);
862 if (rv)
863 goto fail;
864 nr_sects = 0;
865 start_new_extent:
866 start = blk;
868 nr_sects += sects_per_blk;
870 diff >>= 2;
871 blk += sects_per_blk;
874 if (nr_sects) {
875 rv = blkdev_issue_discard(bdev, start, nr_sects, GFP_NOFS,
876 DISCARD_FL_BARRIER);
877 if (rv)
878 goto fail;
880 return;
881 fail:
882 fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv);
883 sdp->sd_args.ar_discard = 0;
886 void gfs2_rgrp_repolish_clones(struct gfs2_rgrpd *rgd)
888 struct gfs2_sbd *sdp = rgd->rd_sbd;
889 unsigned int length = rgd->rd_length;
890 unsigned int x;
892 for (x = 0; x < length; x++) {
893 struct gfs2_bitmap *bi = rgd->rd_bits + x;
894 if (!bi->bi_clone)
895 continue;
896 if (sdp->sd_args.ar_discard)
897 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bi);
898 clear_bit(GBF_FULL, &bi->bi_flags);
899 memcpy(bi->bi_clone + bi->bi_offset,
900 bi->bi_bh->b_data + bi->bi_offset, bi->bi_len);
903 spin_lock(&sdp->sd_rindex_spin);
904 rgd->rd_free_clone = rgd->rd_free;
905 spin_unlock(&sdp->sd_rindex_spin);
909 * gfs2_alloc_get - get the struct gfs2_alloc structure for an inode
910 * @ip: the incore GFS2 inode structure
912 * Returns: the struct gfs2_alloc
915 struct gfs2_alloc *gfs2_alloc_get(struct gfs2_inode *ip)
917 BUG_ON(ip->i_alloc != NULL);
918 ip->i_alloc = kzalloc(sizeof(struct gfs2_alloc), GFP_KERNEL);
919 return ip->i_alloc;
923 * try_rgrp_fit - See if a given reservation will fit in a given RG
924 * @rgd: the RG data
925 * @al: the struct gfs2_alloc structure describing the reservation
927 * If there's room for the requested blocks to be allocated from the RG:
928 * Sets the $al_rgd field in @al.
930 * Returns: 1 on success (it fits), 0 on failure (it doesn't fit)
933 static int try_rgrp_fit(struct gfs2_rgrpd *rgd, struct gfs2_alloc *al)
935 struct gfs2_sbd *sdp = rgd->rd_sbd;
936 int ret = 0;
938 if (rgd->rd_flags & (GFS2_RGF_NOALLOC | GFS2_RDF_ERROR))
939 return 0;
941 spin_lock(&sdp->sd_rindex_spin);
942 if (rgd->rd_free_clone >= al->al_requested) {
943 al->al_rgd = rgd;
944 ret = 1;
946 spin_unlock(&sdp->sd_rindex_spin);
948 return ret;
952 * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes
953 * @rgd: The rgrp
955 * Returns: The inode, if one has been found
958 static struct inode *try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked,
959 u64 skip)
961 struct inode *inode;
962 u32 goal = 0, block;
963 u64 no_addr;
964 struct gfs2_sbd *sdp = rgd->rd_sbd;
965 unsigned int n;
967 for(;;) {
968 if (goal >= rgd->rd_data)
969 break;
970 down_write(&sdp->sd_log_flush_lock);
971 n = 1;
972 block = rgblk_search(rgd, goal, GFS2_BLKST_UNLINKED,
973 GFS2_BLKST_UNLINKED, &n);
974 up_write(&sdp->sd_log_flush_lock);
975 if (block == BFITNOENT)
976 break;
977 /* rgblk_search can return a block < goal, so we need to
978 keep it marching forward. */
979 no_addr = block + rgd->rd_data0;
980 goal++;
981 if (*last_unlinked != NO_BLOCK && no_addr <= *last_unlinked)
982 continue;
983 if (no_addr == skip)
984 continue;
985 *last_unlinked = no_addr;
986 inode = gfs2_inode_lookup(rgd->rd_sbd->sd_vfs, DT_UNKNOWN,
987 no_addr, -1, 1);
988 if (!IS_ERR(inode))
989 return inode;
992 rgd->rd_flags &= ~GFS2_RDF_CHECK;
993 return NULL;
997 * recent_rgrp_next - get next RG from "recent" list
998 * @cur_rgd: current rgrp
1000 * Returns: The next rgrp in the recent list
1003 static struct gfs2_rgrpd *recent_rgrp_next(struct gfs2_rgrpd *cur_rgd)
1005 struct gfs2_sbd *sdp = cur_rgd->rd_sbd;
1006 struct list_head *head;
1007 struct gfs2_rgrpd *rgd;
1009 spin_lock(&sdp->sd_rindex_spin);
1010 head = &sdp->sd_rindex_mru_list;
1011 if (unlikely(cur_rgd->rd_list_mru.next == head)) {
1012 spin_unlock(&sdp->sd_rindex_spin);
1013 return NULL;
1015 rgd = list_entry(cur_rgd->rd_list_mru.next, struct gfs2_rgrpd, rd_list_mru);
1016 spin_unlock(&sdp->sd_rindex_spin);
1017 return rgd;
1021 * forward_rgrp_get - get an rgrp to try next from full list
1022 * @sdp: The GFS2 superblock
1024 * Returns: The rgrp to try next
1027 static struct gfs2_rgrpd *forward_rgrp_get(struct gfs2_sbd *sdp)
1029 struct gfs2_rgrpd *rgd;
1030 unsigned int journals = gfs2_jindex_size(sdp);
1031 unsigned int rg = 0, x;
1033 spin_lock(&sdp->sd_rindex_spin);
1035 rgd = sdp->sd_rindex_forward;
1036 if (!rgd) {
1037 if (sdp->sd_rgrps >= journals)
1038 rg = sdp->sd_rgrps * sdp->sd_jdesc->jd_jid / journals;
1040 for (x = 0, rgd = gfs2_rgrpd_get_first(sdp); x < rg;
1041 x++, rgd = gfs2_rgrpd_get_next(rgd))
1042 /* Do Nothing */;
1044 sdp->sd_rindex_forward = rgd;
1047 spin_unlock(&sdp->sd_rindex_spin);
1049 return rgd;
1053 * forward_rgrp_set - set the forward rgrp pointer
1054 * @sdp: the filesystem
1055 * @rgd: The new forward rgrp
1059 static void forward_rgrp_set(struct gfs2_sbd *sdp, struct gfs2_rgrpd *rgd)
1061 spin_lock(&sdp->sd_rindex_spin);
1062 sdp->sd_rindex_forward = rgd;
1063 spin_unlock(&sdp->sd_rindex_spin);
1067 * get_local_rgrp - Choose and lock a rgrp for allocation
1068 * @ip: the inode to reserve space for
1069 * @rgp: the chosen and locked rgrp
1071 * Try to acquire rgrp in way which avoids contending with others.
1073 * Returns: errno
1076 static struct inode *get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked)
1078 struct inode *inode = NULL;
1079 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1080 struct gfs2_rgrpd *rgd, *begin = NULL;
1081 struct gfs2_alloc *al = ip->i_alloc;
1082 int flags = LM_FLAG_TRY;
1083 int skipped = 0;
1084 int loops = 0;
1085 int error, rg_locked;
1087 rgd = gfs2_blk2rgrpd(sdp, ip->i_goal);
1089 while (rgd) {
1090 rg_locked = 0;
1092 if (gfs2_glock_is_locked_by_me(rgd->rd_gl)) {
1093 rg_locked = 1;
1094 error = 0;
1095 } else {
1096 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1097 LM_FLAG_TRY, &al->al_rgd_gh);
1099 switch (error) {
1100 case 0:
1101 if (try_rgrp_fit(rgd, al))
1102 goto out;
1103 if (rgd->rd_flags & GFS2_RDF_CHECK)
1104 inode = try_rgrp_unlink(rgd, last_unlinked, ip->i_no_addr);
1105 if (!rg_locked)
1106 gfs2_glock_dq_uninit(&al->al_rgd_gh);
1107 if (inode)
1108 return inode;
1109 /* fall through */
1110 case GLR_TRYFAILED:
1111 rgd = recent_rgrp_next(rgd);
1112 break;
1114 default:
1115 return ERR_PTR(error);
1119 /* Go through full list of rgrps */
1121 begin = rgd = forward_rgrp_get(sdp);
1123 for (;;) {
1124 rg_locked = 0;
1126 if (gfs2_glock_is_locked_by_me(rgd->rd_gl)) {
1127 rg_locked = 1;
1128 error = 0;
1129 } else {
1130 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, flags,
1131 &al->al_rgd_gh);
1133 switch (error) {
1134 case 0:
1135 if (try_rgrp_fit(rgd, al))
1136 goto out;
1137 if (rgd->rd_flags & GFS2_RDF_CHECK)
1138 inode = try_rgrp_unlink(rgd, last_unlinked, ip->i_no_addr);
1139 if (!rg_locked)
1140 gfs2_glock_dq_uninit(&al->al_rgd_gh);
1141 if (inode)
1142 return inode;
1143 break;
1145 case GLR_TRYFAILED:
1146 skipped++;
1147 break;
1149 default:
1150 return ERR_PTR(error);
1153 rgd = gfs2_rgrpd_get_next(rgd);
1154 if (!rgd)
1155 rgd = gfs2_rgrpd_get_first(sdp);
1157 if (rgd == begin) {
1158 if (++loops >= 3)
1159 return ERR_PTR(-ENOSPC);
1160 if (!skipped)
1161 loops++;
1162 flags = 0;
1163 if (loops == 2)
1164 gfs2_log_flush(sdp, NULL);
1168 out:
1169 if (begin) {
1170 spin_lock(&sdp->sd_rindex_spin);
1171 list_move(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list);
1172 spin_unlock(&sdp->sd_rindex_spin);
1173 rgd = gfs2_rgrpd_get_next(rgd);
1174 if (!rgd)
1175 rgd = gfs2_rgrpd_get_first(sdp);
1176 forward_rgrp_set(sdp, rgd);
1179 return NULL;
1183 * gfs2_inplace_reserve_i - Reserve space in the filesystem
1184 * @ip: the inode to reserve space for
1186 * Returns: errno
1189 int gfs2_inplace_reserve_i(struct gfs2_inode *ip, char *file, unsigned int line)
1191 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1192 struct gfs2_alloc *al = ip->i_alloc;
1193 struct inode *inode;
1194 int error = 0;
1195 u64 last_unlinked = NO_BLOCK;
1197 if (gfs2_assert_warn(sdp, al->al_requested))
1198 return -EINVAL;
1200 try_again:
1201 /* We need to hold the rindex unless the inode we're using is
1202 the rindex itself, in which case it's already held. */
1203 if (ip != GFS2_I(sdp->sd_rindex))
1204 error = gfs2_rindex_hold(sdp, &al->al_ri_gh);
1205 else if (!sdp->sd_rgrps) /* We may not have the rindex read in, so: */
1206 error = gfs2_ri_update_special(ip);
1208 if (error)
1209 return error;
1211 inode = get_local_rgrp(ip, &last_unlinked);
1212 if (inode) {
1213 if (ip != GFS2_I(sdp->sd_rindex))
1214 gfs2_glock_dq_uninit(&al->al_ri_gh);
1215 if (IS_ERR(inode))
1216 return PTR_ERR(inode);
1217 iput(inode);
1218 gfs2_log_flush(sdp, NULL);
1219 goto try_again;
1222 al->al_file = file;
1223 al->al_line = line;
1225 return 0;
1229 * gfs2_inplace_release - release an inplace reservation
1230 * @ip: the inode the reservation was taken out on
1232 * Release a reservation made by gfs2_inplace_reserve().
1235 void gfs2_inplace_release(struct gfs2_inode *ip)
1237 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1238 struct gfs2_alloc *al = ip->i_alloc;
1240 if (gfs2_assert_warn(sdp, al->al_alloced <= al->al_requested) == -1)
1241 fs_warn(sdp, "al_alloced = %u, al_requested = %u "
1242 "al_file = %s, al_line = %u\n",
1243 al->al_alloced, al->al_requested, al->al_file,
1244 al->al_line);
1246 al->al_rgd = NULL;
1247 if (al->al_rgd_gh.gh_gl)
1248 gfs2_glock_dq_uninit(&al->al_rgd_gh);
1249 if (ip != GFS2_I(sdp->sd_rindex))
1250 gfs2_glock_dq_uninit(&al->al_ri_gh);
1254 * gfs2_get_block_type - Check a block in a RG is of given type
1255 * @rgd: the resource group holding the block
1256 * @block: the block number
1258 * Returns: The block type (GFS2_BLKST_*)
1261 static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block)
1263 struct gfs2_bitmap *bi = NULL;
1264 u32 length, rgrp_block, buf_block;
1265 unsigned int buf;
1266 unsigned char type;
1268 length = rgd->rd_length;
1269 rgrp_block = block - rgd->rd_data0;
1271 for (buf = 0; buf < length; buf++) {
1272 bi = rgd->rd_bits + buf;
1273 if (rgrp_block < (bi->bi_start + bi->bi_len) * GFS2_NBBY)
1274 break;
1277 gfs2_assert(rgd->rd_sbd, buf < length);
1278 buf_block = rgrp_block - bi->bi_start * GFS2_NBBY;
1280 type = gfs2_testbit(rgd, bi->bi_bh->b_data + bi->bi_offset,
1281 bi->bi_len, buf_block);
1283 return type;
1287 * rgblk_search - find a block in @old_state, change allocation
1288 * state to @new_state
1289 * @rgd: the resource group descriptor
1290 * @goal: the goal block within the RG (start here to search for avail block)
1291 * @old_state: GFS2_BLKST_XXX the before-allocation state to find
1292 * @new_state: GFS2_BLKST_XXX the after-allocation block state
1293 * @n: The extent length
1295 * Walk rgrp's bitmap to find bits that represent a block in @old_state.
1296 * Add the found bitmap buffer to the transaction.
1297 * Set the found bits to @new_state to change block's allocation state.
1299 * This function never fails, because we wouldn't call it unless we
1300 * know (from reservation results, etc.) that a block is available.
1302 * Scope of @goal and returned block is just within rgrp, not the whole
1303 * filesystem.
1305 * Returns: the block number allocated
1308 static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal,
1309 unsigned char old_state, unsigned char new_state,
1310 unsigned int *n)
1312 struct gfs2_bitmap *bi = NULL;
1313 const u32 length = rgd->rd_length;
1314 u32 blk = BFITNOENT;
1315 unsigned int buf, x;
1316 const unsigned int elen = *n;
1317 const u8 *buffer = NULL;
1319 *n = 0;
1320 /* Find bitmap block that contains bits for goal block */
1321 for (buf = 0; buf < length; buf++) {
1322 bi = rgd->rd_bits + buf;
1323 /* Convert scope of "goal" from rgrp-wide to within found bit block */
1324 if (goal < (bi->bi_start + bi->bi_len) * GFS2_NBBY) {
1325 goal -= bi->bi_start * GFS2_NBBY;
1326 goto do_search;
1329 buf = 0;
1330 goal = 0;
1332 do_search:
1333 /* Search (up to entire) bitmap in this rgrp for allocatable block.
1334 "x <= length", instead of "x < length", because we typically start
1335 the search in the middle of a bit block, but if we can't find an
1336 allocatable block anywhere else, we want to be able wrap around and
1337 search in the first part of our first-searched bit block. */
1338 for (x = 0; x <= length; x++) {
1339 bi = rgd->rd_bits + buf;
1341 if (test_bit(GBF_FULL, &bi->bi_flags) &&
1342 (old_state == GFS2_BLKST_FREE))
1343 goto skip;
1345 /* The GFS2_BLKST_UNLINKED state doesn't apply to the clone
1346 bitmaps, so we must search the originals for that. */
1347 buffer = bi->bi_bh->b_data + bi->bi_offset;
1348 if (old_state != GFS2_BLKST_UNLINKED && bi->bi_clone)
1349 buffer = bi->bi_clone + bi->bi_offset;
1351 blk = gfs2_bitfit(buffer, bi->bi_len, goal, old_state);
1352 if (blk != BFITNOENT)
1353 break;
1355 if ((goal == 0) && (old_state == GFS2_BLKST_FREE))
1356 set_bit(GBF_FULL, &bi->bi_flags);
1358 /* Try next bitmap block (wrap back to rgrp header if at end) */
1359 skip:
1360 buf++;
1361 buf %= length;
1362 goal = 0;
1365 if (blk == BFITNOENT)
1366 return blk;
1367 *n = 1;
1368 if (old_state == new_state)
1369 goto out;
1371 gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1);
1372 gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, bi->bi_offset,
1373 bi->bi_len, blk, new_state);
1374 goal = blk;
1375 while (*n < elen) {
1376 goal++;
1377 if (goal >= (bi->bi_len * GFS2_NBBY))
1378 break;
1379 if (gfs2_testbit(rgd, buffer, bi->bi_len, goal) !=
1380 GFS2_BLKST_FREE)
1381 break;
1382 gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, bi->bi_offset,
1383 bi->bi_len, goal, new_state);
1384 (*n)++;
1386 out:
1387 return (bi->bi_start * GFS2_NBBY) + blk;
1391 * rgblk_free - Change alloc state of given block(s)
1392 * @sdp: the filesystem
1393 * @bstart: the start of a run of blocks to free
1394 * @blen: the length of the block run (all must lie within ONE RG!)
1395 * @new_state: GFS2_BLKST_XXX the after-allocation block state
1397 * Returns: Resource group containing the block(s)
1400 static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart,
1401 u32 blen, unsigned char new_state)
1403 struct gfs2_rgrpd *rgd;
1404 struct gfs2_bitmap *bi = NULL;
1405 u32 length, rgrp_blk, buf_blk;
1406 unsigned int buf;
1408 rgd = gfs2_blk2rgrpd(sdp, bstart);
1409 if (!rgd) {
1410 if (gfs2_consist(sdp))
1411 fs_err(sdp, "block = %llu\n", (unsigned long long)bstart);
1412 return NULL;
1415 length = rgd->rd_length;
1417 rgrp_blk = bstart - rgd->rd_data0;
1419 while (blen--) {
1420 for (buf = 0; buf < length; buf++) {
1421 bi = rgd->rd_bits + buf;
1422 if (rgrp_blk < (bi->bi_start + bi->bi_len) * GFS2_NBBY)
1423 break;
1426 gfs2_assert(rgd->rd_sbd, buf < length);
1428 buf_blk = rgrp_blk - bi->bi_start * GFS2_NBBY;
1429 rgrp_blk++;
1431 if (!bi->bi_clone) {
1432 bi->bi_clone = kmalloc(bi->bi_bh->b_size,
1433 GFP_NOFS | __GFP_NOFAIL);
1434 memcpy(bi->bi_clone + bi->bi_offset,
1435 bi->bi_bh->b_data + bi->bi_offset,
1436 bi->bi_len);
1438 gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1);
1439 gfs2_setbit(rgd, bi->bi_bh->b_data, NULL, bi->bi_offset,
1440 bi->bi_len, buf_blk, new_state);
1443 return rgd;
1447 * gfs2_rgrp_dump - print out an rgrp
1448 * @seq: The iterator
1449 * @gl: The glock in question
1453 int gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl)
1455 const struct gfs2_rgrpd *rgd = gl->gl_object;
1456 if (rgd == NULL)
1457 return 0;
1458 gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u\n",
1459 (unsigned long long)rgd->rd_addr, rgd->rd_flags,
1460 rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes);
1461 return 0;
1464 static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd)
1466 struct gfs2_sbd *sdp = rgd->rd_sbd;
1467 fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n",
1468 (unsigned long long)rgd->rd_addr);
1469 fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n");
1470 gfs2_rgrp_dump(NULL, rgd->rd_gl);
1471 rgd->rd_flags |= GFS2_RDF_ERROR;
1475 * gfs2_alloc_block - Allocate one or more blocks
1476 * @ip: the inode to allocate the block for
1477 * @bn: Used to return the starting block number
1478 * @n: requested number of blocks/extent length (value/result)
1480 * Returns: 0 or error
1483 int gfs2_alloc_block(struct gfs2_inode *ip, u64 *bn, unsigned int *n)
1485 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1486 struct buffer_head *dibh;
1487 struct gfs2_alloc *al = ip->i_alloc;
1488 struct gfs2_rgrpd *rgd = al->al_rgd;
1489 u32 goal, blk;
1490 u64 block;
1491 int error;
1493 if (rgrp_contains_block(rgd, ip->i_goal))
1494 goal = ip->i_goal - rgd->rd_data0;
1495 else
1496 goal = rgd->rd_last_alloc;
1498 blk = rgblk_search(rgd, goal, GFS2_BLKST_FREE, GFS2_BLKST_USED, n);
1500 /* Since all blocks are reserved in advance, this shouldn't happen */
1501 if (blk == BFITNOENT)
1502 goto rgrp_error;
1504 rgd->rd_last_alloc = blk;
1505 block = rgd->rd_data0 + blk;
1506 ip->i_goal = block;
1507 error = gfs2_meta_inode_buffer(ip, &dibh);
1508 if (error == 0) {
1509 struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
1510 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1511 di->di_goal_meta = di->di_goal_data = cpu_to_be64(ip->i_goal);
1512 brelse(dibh);
1514 if (rgd->rd_free < *n)
1515 goto rgrp_error;
1517 rgd->rd_free -= *n;
1519 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1520 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1522 al->al_alloced += *n;
1524 gfs2_statfs_change(sdp, 0, -(s64)*n, 0);
1525 gfs2_quota_change(ip, *n, ip->i_inode.i_uid, ip->i_inode.i_gid);
1527 spin_lock(&sdp->sd_rindex_spin);
1528 rgd->rd_free_clone -= *n;
1529 spin_unlock(&sdp->sd_rindex_spin);
1530 trace_gfs2_block_alloc(ip, block, *n, GFS2_BLKST_USED);
1531 *bn = block;
1532 return 0;
1534 rgrp_error:
1535 gfs2_rgrp_error(rgd);
1536 return -EIO;
1540 * gfs2_alloc_di - Allocate a dinode
1541 * @dip: the directory that the inode is going in
1542 * @bn: the block number which is allocated
1543 * @generation: the generation number of the inode
1545 * Returns: 0 on success or error
1548 int gfs2_alloc_di(struct gfs2_inode *dip, u64 *bn, u64 *generation)
1550 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1551 struct gfs2_alloc *al = dip->i_alloc;
1552 struct gfs2_rgrpd *rgd = al->al_rgd;
1553 u32 blk;
1554 u64 block;
1555 unsigned int n = 1;
1557 blk = rgblk_search(rgd, rgd->rd_last_alloc,
1558 GFS2_BLKST_FREE, GFS2_BLKST_DINODE, &n);
1560 /* Since all blocks are reserved in advance, this shouldn't happen */
1561 if (blk == BFITNOENT)
1562 goto rgrp_error;
1564 rgd->rd_last_alloc = blk;
1565 block = rgd->rd_data0 + blk;
1566 if (rgd->rd_free == 0)
1567 goto rgrp_error;
1569 rgd->rd_free--;
1570 rgd->rd_dinodes++;
1571 *generation = rgd->rd_igeneration++;
1572 if (*generation == 0)
1573 *generation = rgd->rd_igeneration++;
1574 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1575 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1577 al->al_alloced++;
1579 gfs2_statfs_change(sdp, 0, -1, +1);
1580 gfs2_trans_add_unrevoke(sdp, block, 1);
1582 spin_lock(&sdp->sd_rindex_spin);
1583 rgd->rd_free_clone--;
1584 spin_unlock(&sdp->sd_rindex_spin);
1585 trace_gfs2_block_alloc(dip, block, 1, GFS2_BLKST_DINODE);
1586 *bn = block;
1587 return 0;
1589 rgrp_error:
1590 gfs2_rgrp_error(rgd);
1591 return -EIO;
1595 * gfs2_free_data - free a contiguous run of data block(s)
1596 * @ip: the inode these blocks are being freed from
1597 * @bstart: first block of a run of contiguous blocks
1598 * @blen: the length of the block run
1602 void gfs2_free_data(struct gfs2_inode *ip, u64 bstart, u32 blen)
1604 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1605 struct gfs2_rgrpd *rgd;
1607 rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE);
1608 if (!rgd)
1609 return;
1610 trace_gfs2_block_alloc(ip, bstart, blen, GFS2_BLKST_FREE);
1611 rgd->rd_free += blen;
1613 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1614 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1616 gfs2_trans_add_rg(rgd);
1618 gfs2_statfs_change(sdp, 0, +blen, 0);
1619 gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid);
1623 * gfs2_free_meta - free a contiguous run of data block(s)
1624 * @ip: the inode these blocks are being freed from
1625 * @bstart: first block of a run of contiguous blocks
1626 * @blen: the length of the block run
1630 void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen)
1632 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1633 struct gfs2_rgrpd *rgd;
1635 rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE);
1636 if (!rgd)
1637 return;
1638 trace_gfs2_block_alloc(ip, bstart, blen, GFS2_BLKST_FREE);
1639 rgd->rd_free += blen;
1641 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1642 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1644 gfs2_trans_add_rg(rgd);
1646 gfs2_statfs_change(sdp, 0, +blen, 0);
1647 gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid);
1648 gfs2_meta_wipe(ip, bstart, blen);
1651 void gfs2_unlink_di(struct inode *inode)
1653 struct gfs2_inode *ip = GFS2_I(inode);
1654 struct gfs2_sbd *sdp = GFS2_SB(inode);
1655 struct gfs2_rgrpd *rgd;
1656 u64 blkno = ip->i_no_addr;
1658 rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED);
1659 if (!rgd)
1660 return;
1661 trace_gfs2_block_alloc(ip, blkno, 1, GFS2_BLKST_UNLINKED);
1662 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1663 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1664 gfs2_trans_add_rg(rgd);
1667 static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno)
1669 struct gfs2_sbd *sdp = rgd->rd_sbd;
1670 struct gfs2_rgrpd *tmp_rgd;
1672 tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE);
1673 if (!tmp_rgd)
1674 return;
1675 gfs2_assert_withdraw(sdp, rgd == tmp_rgd);
1677 if (!rgd->rd_dinodes)
1678 gfs2_consist_rgrpd(rgd);
1679 rgd->rd_dinodes--;
1680 rgd->rd_free++;
1682 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1);
1683 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
1685 gfs2_statfs_change(sdp, 0, +1, -1);
1686 gfs2_trans_add_rg(rgd);
1690 void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip)
1692 gfs2_free_uninit_di(rgd, ip->i_no_addr);
1693 trace_gfs2_block_alloc(ip, ip->i_no_addr, 1, GFS2_BLKST_FREE);
1694 gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid);
1695 gfs2_meta_wipe(ip, ip->i_no_addr, 1);
1699 * gfs2_check_blk_type - Check the type of a block
1700 * @sdp: The superblock
1701 * @no_addr: The block number to check
1702 * @type: The block type we are looking for
1704 * Returns: 0 if the block type matches the expected type
1705 * -ESTALE if it doesn't match
1706 * or -ve errno if something went wrong while checking
1709 int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type)
1711 struct gfs2_rgrpd *rgd;
1712 struct gfs2_holder ri_gh, rgd_gh;
1713 int error;
1715 error = gfs2_rindex_hold(sdp, &ri_gh);
1716 if (error)
1717 goto fail;
1719 error = -EINVAL;
1720 rgd = gfs2_blk2rgrpd(sdp, no_addr);
1721 if (!rgd)
1722 goto fail_rindex;
1724 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh);
1725 if (error)
1726 goto fail_rindex;
1728 if (gfs2_get_block_type(rgd, no_addr) != type)
1729 error = -ESTALE;
1731 gfs2_glock_dq_uninit(&rgd_gh);
1732 fail_rindex:
1733 gfs2_glock_dq_uninit(&ri_gh);
1734 fail:
1735 return error;
1739 * gfs2_rlist_add - add a RG to a list of RGs
1740 * @sdp: the filesystem
1741 * @rlist: the list of resource groups
1742 * @block: the block
1744 * Figure out what RG a block belongs to and add that RG to the list
1746 * FIXME: Don't use NOFAIL
1750 void gfs2_rlist_add(struct gfs2_sbd *sdp, struct gfs2_rgrp_list *rlist,
1751 u64 block)
1753 struct gfs2_rgrpd *rgd;
1754 struct gfs2_rgrpd **tmp;
1755 unsigned int new_space;
1756 unsigned int x;
1758 if (gfs2_assert_warn(sdp, !rlist->rl_ghs))
1759 return;
1761 rgd = gfs2_blk2rgrpd(sdp, block);
1762 if (!rgd) {
1763 if (gfs2_consist(sdp))
1764 fs_err(sdp, "block = %llu\n", (unsigned long long)block);
1765 return;
1768 for (x = 0; x < rlist->rl_rgrps; x++)
1769 if (rlist->rl_rgd[x] == rgd)
1770 return;
1772 if (rlist->rl_rgrps == rlist->rl_space) {
1773 new_space = rlist->rl_space + 10;
1775 tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *),
1776 GFP_NOFS | __GFP_NOFAIL);
1778 if (rlist->rl_rgd) {
1779 memcpy(tmp, rlist->rl_rgd,
1780 rlist->rl_space * sizeof(struct gfs2_rgrpd *));
1781 kfree(rlist->rl_rgd);
1784 rlist->rl_space = new_space;
1785 rlist->rl_rgd = tmp;
1788 rlist->rl_rgd[rlist->rl_rgrps++] = rgd;
1792 * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate
1793 * and initialize an array of glock holders for them
1794 * @rlist: the list of resource groups
1795 * @state: the lock state to acquire the RG lock in
1796 * @flags: the modifier flags for the holder structures
1798 * FIXME: Don't use NOFAIL
1802 void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state)
1804 unsigned int x;
1806 rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder),
1807 GFP_NOFS | __GFP_NOFAIL);
1808 for (x = 0; x < rlist->rl_rgrps; x++)
1809 gfs2_holder_init(rlist->rl_rgd[x]->rd_gl,
1810 state, 0,
1811 &rlist->rl_ghs[x]);
1815 * gfs2_rlist_free - free a resource group list
1816 * @list: the list of resource groups
1820 void gfs2_rlist_free(struct gfs2_rgrp_list *rlist)
1822 unsigned int x;
1824 kfree(rlist->rl_rgd);
1826 if (rlist->rl_ghs) {
1827 for (x = 0; x < rlist->rl_rgrps; x++)
1828 gfs2_holder_uninit(&rlist->rl_ghs[x]);
1829 kfree(rlist->rl_ghs);