Loading...
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
4 *
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 */
9
10#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11
12#include <linux/slab.h>
13#include <linux/spinlock.h>
14#include <linux/completion.h>
15#include <linux/buffer_head.h>
16#include <linux/fs.h>
17#include <linux/gfs2_ondisk.h>
18#include <linux/prefetch.h>
19#include <linux/blkdev.h>
20#include <linux/rbtree.h>
21#include <linux/random.h>
22
23#include "gfs2.h"
24#include "incore.h"
25#include "glock.h"
26#include "glops.h"
27#include "lops.h"
28#include "meta_io.h"
29#include "quota.h"
30#include "rgrp.h"
31#include "super.h"
32#include "trans.h"
33#include "util.h"
34#include "log.h"
35#include "inode.h"
36#include "trace_gfs2.h"
37
38#define BFITNOENT ((u32)~0)
39#define NO_BLOCK ((u64)~0)
40
41#if BITS_PER_LONG == 32
42#define LBITMASK (0x55555555UL)
43#define LBITSKIP55 (0x55555555UL)
44#define LBITSKIP00 (0x00000000UL)
45#else
46#define LBITMASK (0x5555555555555555UL)
47#define LBITSKIP55 (0x5555555555555555UL)
48#define LBITSKIP00 (0x0000000000000000UL)
49#endif
50
51/*
52 * These routines are used by the resource group routines (rgrp.c)
53 * to keep track of block allocation. Each block is represented by two
54 * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks.
55 *
56 * 0 = Free
57 * 1 = Used (not metadata)
58 * 2 = Unlinked (still in use) inode
59 * 3 = Used (metadata)
60 */
61
62struct gfs2_extent {
63 struct gfs2_rbm rbm;
64 u32 len;
65};
66
67static const char valid_change[16] = {
68 /* current */
69 /* n */ 0, 1, 1, 1,
70 /* e */ 1, 0, 0, 0,
71 /* w */ 0, 0, 0, 1,
72 1, 0, 0, 0
73};
74
75static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 *minext,
76 const struct gfs2_inode *ip, bool nowrap,
77 const struct gfs2_alloc_parms *ap);
78
79
80/**
81 * gfs2_setbit - Set a bit in the bitmaps
82 * @rbm: The position of the bit to set
83 * @do_clone: Also set the clone bitmap, if it exists
84 * @new_state: the new state of the block
85 *
86 */
87
88static inline void gfs2_setbit(const struct gfs2_rbm *rbm, bool do_clone,
89 unsigned char new_state)
90{
91 unsigned char *byte1, *byte2, *end, cur_state;
92 struct gfs2_bitmap *bi = rbm_bi(rbm);
93 unsigned int buflen = bi->bi_len;
94 const unsigned int bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE;
95
96 byte1 = bi->bi_bh->b_data + bi->bi_offset + (rbm->offset / GFS2_NBBY);
97 end = bi->bi_bh->b_data + bi->bi_offset + buflen;
98
99 BUG_ON(byte1 >= end);
100
101 cur_state = (*byte1 >> bit) & GFS2_BIT_MASK;
102
103 if (unlikely(!valid_change[new_state * 4 + cur_state])) {
104 pr_warn("buf_blk = 0x%x old_state=%d, new_state=%d\n",
105 rbm->offset, cur_state, new_state);
106 pr_warn("rgrp=0x%llx bi_start=0x%x\n",
107 (unsigned long long)rbm->rgd->rd_addr, bi->bi_start);
108 pr_warn("bi_offset=0x%x bi_len=0x%x\n",
109 bi->bi_offset, bi->bi_len);
110 dump_stack();
111 gfs2_consist_rgrpd(rbm->rgd);
112 return;
113 }
114 *byte1 ^= (cur_state ^ new_state) << bit;
115
116 if (do_clone && bi->bi_clone) {
117 byte2 = bi->bi_clone + bi->bi_offset + (rbm->offset / GFS2_NBBY);
118 cur_state = (*byte2 >> bit) & GFS2_BIT_MASK;
119 *byte2 ^= (cur_state ^ new_state) << bit;
120 }
121}
122
123/**
124 * gfs2_testbit - test a bit in the bitmaps
125 * @rbm: The bit to test
126 *
127 * Returns: The two bit block state of the requested bit
128 */
129
130static inline u8 gfs2_testbit(const struct gfs2_rbm *rbm)
131{
132 struct gfs2_bitmap *bi = rbm_bi(rbm);
133 const u8 *buffer = bi->bi_bh->b_data + bi->bi_offset;
134 const u8 *byte;
135 unsigned int bit;
136
137 byte = buffer + (rbm->offset / GFS2_NBBY);
138 bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE;
139
140 return (*byte >> bit) & GFS2_BIT_MASK;
141}
142
143/**
144 * gfs2_bit_search
145 * @ptr: Pointer to bitmap data
146 * @mask: Mask to use (normally 0x55555.... but adjusted for search start)
147 * @state: The state we are searching for
148 *
149 * We xor the bitmap data with a patter which is the bitwise opposite
150 * of what we are looking for, this gives rise to a pattern of ones
151 * wherever there is a match. Since we have two bits per entry, we
152 * take this pattern, shift it down by one place and then and it with
153 * the original. All the even bit positions (0,2,4, etc) then represent
154 * successful matches, so we mask with 0x55555..... to remove the unwanted
155 * odd bit positions.
156 *
157 * This allows searching of a whole u64 at once (32 blocks) with a
158 * single test (on 64 bit arches).
159 */
160
161static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state)
162{
163 u64 tmp;
164 static const u64 search[] = {
165 [0] = 0xffffffffffffffffULL,
166 [1] = 0xaaaaaaaaaaaaaaaaULL,
167 [2] = 0x5555555555555555ULL,
168 [3] = 0x0000000000000000ULL,
169 };
170 tmp = le64_to_cpu(*ptr) ^ search[state];
171 tmp &= (tmp >> 1);
172 tmp &= mask;
173 return tmp;
174}
175
176/**
177 * rs_cmp - multi-block reservation range compare
178 * @blk: absolute file system block number of the new reservation
179 * @len: number of blocks in the new reservation
180 * @rs: existing reservation to compare against
181 *
182 * returns: 1 if the block range is beyond the reach of the reservation
183 * -1 if the block range is before the start of the reservation
184 * 0 if the block range overlaps with the reservation
185 */
186static inline int rs_cmp(u64 blk, u32 len, struct gfs2_blkreserv *rs)
187{
188 u64 startblk = gfs2_rbm_to_block(&rs->rs_rbm);
189
190 if (blk >= startblk + rs->rs_free)
191 return 1;
192 if (blk + len - 1 < startblk)
193 return -1;
194 return 0;
195}
196
197/**
198 * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing
199 * a block in a given allocation state.
200 * @buf: the buffer that holds the bitmaps
201 * @len: the length (in bytes) of the buffer
202 * @goal: start search at this block's bit-pair (within @buffer)
203 * @state: GFS2_BLKST_XXX the state of the block we're looking for.
204 *
205 * Scope of @goal and returned block number is only within this bitmap buffer,
206 * not entire rgrp or filesystem. @buffer will be offset from the actual
207 * beginning of a bitmap block buffer, skipping any header structures, but
208 * headers are always a multiple of 64 bits long so that the buffer is
209 * always aligned to a 64 bit boundary.
210 *
211 * The size of the buffer is in bytes, but is it assumed that it is
212 * always ok to read a complete multiple of 64 bits at the end
213 * of the block in case the end is no aligned to a natural boundary.
214 *
215 * Return: the block number (bitmap buffer scope) that was found
216 */
217
218static u32 gfs2_bitfit(const u8 *buf, const unsigned int len,
219 u32 goal, u8 state)
220{
221 u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1);
222 const __le64 *ptr = ((__le64 *)buf) + (goal >> 5);
223 const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64)));
224 u64 tmp;
225 u64 mask = 0x5555555555555555ULL;
226 u32 bit;
227
228 /* Mask off bits we don't care about at the start of the search */
229 mask <<= spoint;
230 tmp = gfs2_bit_search(ptr, mask, state);
231 ptr++;
232 while(tmp == 0 && ptr < end) {
233 tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state);
234 ptr++;
235 }
236 /* Mask off any bits which are more than len bytes from the start */
237 if (ptr == end && (len & (sizeof(u64) - 1)))
238 tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1))));
239 /* Didn't find anything, so return */
240 if (tmp == 0)
241 return BFITNOENT;
242 ptr--;
243 bit = __ffs64(tmp);
244 bit /= 2; /* two bits per entry in the bitmap */
245 return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit;
246}
247
248/**
249 * gfs2_rbm_from_block - Set the rbm based upon rgd and block number
250 * @rbm: The rbm with rgd already set correctly
251 * @block: The block number (filesystem relative)
252 *
253 * This sets the bi and offset members of an rbm based on a
254 * resource group and a filesystem relative block number. The
255 * resource group must be set in the rbm on entry, the bi and
256 * offset members will be set by this function.
257 *
258 * Returns: 0 on success, or an error code
259 */
260
261static int gfs2_rbm_from_block(struct gfs2_rbm *rbm, u64 block)
262{
263 u64 rblock = block - rbm->rgd->rd_data0;
264
265 if (WARN_ON_ONCE(rblock > UINT_MAX))
266 return -EINVAL;
267 if (block >= rbm->rgd->rd_data0 + rbm->rgd->rd_data)
268 return -E2BIG;
269
270 rbm->bii = 0;
271 rbm->offset = (u32)(rblock);
272 /* Check if the block is within the first block */
273 if (rbm->offset < rbm_bi(rbm)->bi_blocks)
274 return 0;
275
276 /* Adjust for the size diff between gfs2_meta_header and gfs2_rgrp */
277 rbm->offset += (sizeof(struct gfs2_rgrp) -
278 sizeof(struct gfs2_meta_header)) * GFS2_NBBY;
279 rbm->bii = rbm->offset / rbm->rgd->rd_sbd->sd_blocks_per_bitmap;
280 rbm->offset -= rbm->bii * rbm->rgd->rd_sbd->sd_blocks_per_bitmap;
281 return 0;
282}
283
284/**
285 * gfs2_rbm_incr - increment an rbm structure
286 * @rbm: The rbm with rgd already set correctly
287 *
288 * This function takes an existing rbm structure and increments it to the next
289 * viable block offset.
290 *
291 * Returns: If incrementing the offset would cause the rbm to go past the
292 * end of the rgrp, true is returned, otherwise false.
293 *
294 */
295
296static bool gfs2_rbm_incr(struct gfs2_rbm *rbm)
297{
298 if (rbm->offset + 1 < rbm_bi(rbm)->bi_blocks) { /* in the same bitmap */
299 rbm->offset++;
300 return false;
301 }
302 if (rbm->bii == rbm->rgd->rd_length - 1) /* at the last bitmap */
303 return true;
304
305 rbm->offset = 0;
306 rbm->bii++;
307 return false;
308}
309
310/**
311 * gfs2_unaligned_extlen - Look for free blocks which are not byte aligned
312 * @rbm: Position to search (value/result)
313 * @n_unaligned: Number of unaligned blocks to check
314 * @len: Decremented for each block found (terminate on zero)
315 *
316 * Returns: true if a non-free block is encountered
317 */
318
319static bool gfs2_unaligned_extlen(struct gfs2_rbm *rbm, u32 n_unaligned, u32 *len)
320{
321 u32 n;
322 u8 res;
323
324 for (n = 0; n < n_unaligned; n++) {
325 res = gfs2_testbit(rbm);
326 if (res != GFS2_BLKST_FREE)
327 return true;
328 (*len)--;
329 if (*len == 0)
330 return true;
331 if (gfs2_rbm_incr(rbm))
332 return true;
333 }
334
335 return false;
336}
337
338/**
339 * gfs2_free_extlen - Return extent length of free blocks
340 * @rrbm: Starting position
341 * @len: Max length to check
342 *
343 * Starting at the block specified by the rbm, see how many free blocks
344 * there are, not reading more than len blocks ahead. This can be done
345 * using memchr_inv when the blocks are byte aligned, but has to be done
346 * on a block by block basis in case of unaligned blocks. Also this
347 * function can cope with bitmap boundaries (although it must stop on
348 * a resource group boundary)
349 *
350 * Returns: Number of free blocks in the extent
351 */
352
353static u32 gfs2_free_extlen(const struct gfs2_rbm *rrbm, u32 len)
354{
355 struct gfs2_rbm rbm = *rrbm;
356 u32 n_unaligned = rbm.offset & 3;
357 u32 size = len;
358 u32 bytes;
359 u32 chunk_size;
360 u8 *ptr, *start, *end;
361 u64 block;
362 struct gfs2_bitmap *bi;
363
364 if (n_unaligned &&
365 gfs2_unaligned_extlen(&rbm, 4 - n_unaligned, &len))
366 goto out;
367
368 n_unaligned = len & 3;
369 /* Start is now byte aligned */
370 while (len > 3) {
371 bi = rbm_bi(&rbm);
372 start = bi->bi_bh->b_data;
373 if (bi->bi_clone)
374 start = bi->bi_clone;
375 end = start + bi->bi_bh->b_size;
376 start += bi->bi_offset;
377 BUG_ON(rbm.offset & 3);
378 start += (rbm.offset / GFS2_NBBY);
379 bytes = min_t(u32, len / GFS2_NBBY, (end - start));
380 ptr = memchr_inv(start, 0, bytes);
381 chunk_size = ((ptr == NULL) ? bytes : (ptr - start));
382 chunk_size *= GFS2_NBBY;
383 BUG_ON(len < chunk_size);
384 len -= chunk_size;
385 block = gfs2_rbm_to_block(&rbm);
386 if (gfs2_rbm_from_block(&rbm, block + chunk_size)) {
387 n_unaligned = 0;
388 break;
389 }
390 if (ptr) {
391 n_unaligned = 3;
392 break;
393 }
394 n_unaligned = len & 3;
395 }
396
397 /* Deal with any bits left over at the end */
398 if (n_unaligned)
399 gfs2_unaligned_extlen(&rbm, n_unaligned, &len);
400out:
401 return size - len;
402}
403
404/**
405 * gfs2_bitcount - count the number of bits in a certain state
406 * @rgd: the resource group descriptor
407 * @buffer: the buffer that holds the bitmaps
408 * @buflen: the length (in bytes) of the buffer
409 * @state: the state of the block we're looking for
410 *
411 * Returns: The number of bits
412 */
413
414static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer,
415 unsigned int buflen, u8 state)
416{
417 const u8 *byte = buffer;
418 const u8 *end = buffer + buflen;
419 const u8 state1 = state << 2;
420 const u8 state2 = state << 4;
421 const u8 state3 = state << 6;
422 u32 count = 0;
423
424 for (; byte < end; byte++) {
425 if (((*byte) & 0x03) == state)
426 count++;
427 if (((*byte) & 0x0C) == state1)
428 count++;
429 if (((*byte) & 0x30) == state2)
430 count++;
431 if (((*byte) & 0xC0) == state3)
432 count++;
433 }
434
435 return count;
436}
437
438/**
439 * gfs2_rgrp_verify - Verify that a resource group is consistent
440 * @rgd: the rgrp
441 *
442 */
443
444void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd)
445{
446 struct gfs2_sbd *sdp = rgd->rd_sbd;
447 struct gfs2_bitmap *bi = NULL;
448 u32 length = rgd->rd_length;
449 u32 count[4], tmp;
450 int buf, x;
451
452 memset(count, 0, 4 * sizeof(u32));
453
454 /* Count # blocks in each of 4 possible allocation states */
455 for (buf = 0; buf < length; buf++) {
456 bi = rgd->rd_bits + buf;
457 for (x = 0; x < 4; x++)
458 count[x] += gfs2_bitcount(rgd,
459 bi->bi_bh->b_data +
460 bi->bi_offset,
461 bi->bi_len, x);
462 }
463
464 if (count[0] != rgd->rd_free) {
465 if (gfs2_consist_rgrpd(rgd))
466 fs_err(sdp, "free data mismatch: %u != %u\n",
467 count[0], rgd->rd_free);
468 return;
469 }
470
471 tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes;
472 if (count[1] != tmp) {
473 if (gfs2_consist_rgrpd(rgd))
474 fs_err(sdp, "used data mismatch: %u != %u\n",
475 count[1], tmp);
476 return;
477 }
478
479 if (count[2] + count[3] != rgd->rd_dinodes) {
480 if (gfs2_consist_rgrpd(rgd))
481 fs_err(sdp, "used metadata mismatch: %u != %u\n",
482 count[2] + count[3], rgd->rd_dinodes);
483 return;
484 }
485}
486
487static inline int rgrp_contains_block(struct gfs2_rgrpd *rgd, u64 block)
488{
489 u64 first = rgd->rd_data0;
490 u64 last = first + rgd->rd_data;
491 return first <= block && block < last;
492}
493
494/**
495 * gfs2_blk2rgrpd - Find resource group for a given data/meta block number
496 * @sdp: The GFS2 superblock
497 * @blk: The data block number
498 * @exact: True if this needs to be an exact match
499 *
500 * Returns: The resource group, or NULL if not found
501 */
502
503struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk, bool exact)
504{
505 struct rb_node *n, *next;
506 struct gfs2_rgrpd *cur;
507
508 spin_lock(&sdp->sd_rindex_spin);
509 n = sdp->sd_rindex_tree.rb_node;
510 while (n) {
511 cur = rb_entry(n, struct gfs2_rgrpd, rd_node);
512 next = NULL;
513 if (blk < cur->rd_addr)
514 next = n->rb_left;
515 else if (blk >= cur->rd_data0 + cur->rd_data)
516 next = n->rb_right;
517 if (next == NULL) {
518 spin_unlock(&sdp->sd_rindex_spin);
519 if (exact) {
520 if (blk < cur->rd_addr)
521 return NULL;
522 if (blk >= cur->rd_data0 + cur->rd_data)
523 return NULL;
524 }
525 return cur;
526 }
527 n = next;
528 }
529 spin_unlock(&sdp->sd_rindex_spin);
530
531 return NULL;
532}
533
534/**
535 * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem
536 * @sdp: The GFS2 superblock
537 *
538 * Returns: The first rgrp in the filesystem
539 */
540
541struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp)
542{
543 const struct rb_node *n;
544 struct gfs2_rgrpd *rgd;
545
546 spin_lock(&sdp->sd_rindex_spin);
547 n = rb_first(&sdp->sd_rindex_tree);
548 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node);
549 spin_unlock(&sdp->sd_rindex_spin);
550
551 return rgd;
552}
553
554/**
555 * gfs2_rgrpd_get_next - get the next RG
556 * @rgd: the resource group descriptor
557 *
558 * Returns: The next rgrp
559 */
560
561struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd)
562{
563 struct gfs2_sbd *sdp = rgd->rd_sbd;
564 const struct rb_node *n;
565
566 spin_lock(&sdp->sd_rindex_spin);
567 n = rb_next(&rgd->rd_node);
568 if (n == NULL)
569 n = rb_first(&sdp->sd_rindex_tree);
570
571 if (unlikely(&rgd->rd_node == n)) {
572 spin_unlock(&sdp->sd_rindex_spin);
573 return NULL;
574 }
575 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node);
576 spin_unlock(&sdp->sd_rindex_spin);
577 return rgd;
578}
579
580void check_and_update_goal(struct gfs2_inode *ip)
581{
582 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
583 if (!ip->i_goal || gfs2_blk2rgrpd(sdp, ip->i_goal, 1) == NULL)
584 ip->i_goal = ip->i_no_addr;
585}
586
587void gfs2_free_clones(struct gfs2_rgrpd *rgd)
588{
589 int x;
590
591 for (x = 0; x < rgd->rd_length; x++) {
592 struct gfs2_bitmap *bi = rgd->rd_bits + x;
593 kfree(bi->bi_clone);
594 bi->bi_clone = NULL;
595 }
596}
597
598/**
599 * gfs2_rsqa_alloc - make sure we have a reservation assigned to the inode
600 * plus a quota allocations data structure, if necessary
601 * @ip: the inode for this reservation
602 */
603int gfs2_rsqa_alloc(struct gfs2_inode *ip)
604{
605 return gfs2_qa_alloc(ip);
606}
607
608static void dump_rs(struct seq_file *seq, const struct gfs2_blkreserv *rs)
609{
610 gfs2_print_dbg(seq, " B: n:%llu s:%llu b:%u f:%u\n",
611 (unsigned long long)rs->rs_inum,
612 (unsigned long long)gfs2_rbm_to_block(&rs->rs_rbm),
613 rs->rs_rbm.offset, rs->rs_free);
614}
615
616/**
617 * __rs_deltree - remove a multi-block reservation from the rgd tree
618 * @rs: The reservation to remove
619 *
620 */
621static void __rs_deltree(struct gfs2_blkreserv *rs)
622{
623 struct gfs2_rgrpd *rgd;
624
625 if (!gfs2_rs_active(rs))
626 return;
627
628 rgd = rs->rs_rbm.rgd;
629 trace_gfs2_rs(rs, TRACE_RS_TREEDEL);
630 rb_erase(&rs->rs_node, &rgd->rd_rstree);
631 RB_CLEAR_NODE(&rs->rs_node);
632
633 if (rs->rs_free) {
634 struct gfs2_bitmap *bi = rbm_bi(&rs->rs_rbm);
635
636 /* return reserved blocks to the rgrp */
637 BUG_ON(rs->rs_rbm.rgd->rd_reserved < rs->rs_free);
638 rs->rs_rbm.rgd->rd_reserved -= rs->rs_free;
639 /* The rgrp extent failure point is likely not to increase;
640 it will only do so if the freed blocks are somehow
641 contiguous with a span of free blocks that follows. Still,
642 it will force the number to be recalculated later. */
643 rgd->rd_extfail_pt += rs->rs_free;
644 rs->rs_free = 0;
645 clear_bit(GBF_FULL, &bi->bi_flags);
646 }
647}
648
649/**
650 * gfs2_rs_deltree - remove a multi-block reservation from the rgd tree
651 * @rs: The reservation to remove
652 *
653 */
654void gfs2_rs_deltree(struct gfs2_blkreserv *rs)
655{
656 struct gfs2_rgrpd *rgd;
657
658 rgd = rs->rs_rbm.rgd;
659 if (rgd) {
660 spin_lock(&rgd->rd_rsspin);
661 __rs_deltree(rs);
662 spin_unlock(&rgd->rd_rsspin);
663 }
664}
665
666/**
667 * gfs2_rsqa_delete - delete a multi-block reservation and quota allocation
668 * @ip: The inode for this reservation
669 * @wcount: The inode's write count, or NULL
670 *
671 */
672void gfs2_rsqa_delete(struct gfs2_inode *ip, atomic_t *wcount)
673{
674 down_write(&ip->i_rw_mutex);
675 if ((wcount == NULL) || (atomic_read(wcount) <= 1)) {
676 gfs2_rs_deltree(&ip->i_res);
677 BUG_ON(ip->i_res.rs_free);
678 }
679 up_write(&ip->i_rw_mutex);
680 gfs2_qa_delete(ip, wcount);
681}
682
683/**
684 * return_all_reservations - return all reserved blocks back to the rgrp.
685 * @rgd: the rgrp that needs its space back
686 *
687 * We previously reserved a bunch of blocks for allocation. Now we need to
688 * give them back. This leave the reservation structures in tact, but removes
689 * all of their corresponding "no-fly zones".
690 */
691static void return_all_reservations(struct gfs2_rgrpd *rgd)
692{
693 struct rb_node *n;
694 struct gfs2_blkreserv *rs;
695
696 spin_lock(&rgd->rd_rsspin);
697 while ((n = rb_first(&rgd->rd_rstree))) {
698 rs = rb_entry(n, struct gfs2_blkreserv, rs_node);
699 __rs_deltree(rs);
700 }
701 spin_unlock(&rgd->rd_rsspin);
702}
703
704void gfs2_clear_rgrpd(struct gfs2_sbd *sdp)
705{
706 struct rb_node *n;
707 struct gfs2_rgrpd *rgd;
708 struct gfs2_glock *gl;
709
710 while ((n = rb_first(&sdp->sd_rindex_tree))) {
711 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node);
712 gl = rgd->rd_gl;
713
714 rb_erase(n, &sdp->sd_rindex_tree);
715
716 if (gl) {
717 spin_lock(&gl->gl_lockref.lock);
718 gl->gl_object = NULL;
719 spin_unlock(&gl->gl_lockref.lock);
720 gfs2_glock_add_to_lru(gl);
721 gfs2_glock_put(gl);
722 }
723
724 gfs2_free_clones(rgd);
725 kfree(rgd->rd_bits);
726 return_all_reservations(rgd);
727 kmem_cache_free(gfs2_rgrpd_cachep, rgd);
728 }
729}
730
731static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd)
732{
733 pr_info("ri_addr = %llu\n", (unsigned long long)rgd->rd_addr);
734 pr_info("ri_length = %u\n", rgd->rd_length);
735 pr_info("ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0);
736 pr_info("ri_data = %u\n", rgd->rd_data);
737 pr_info("ri_bitbytes = %u\n", rgd->rd_bitbytes);
738}
739
740/**
741 * gfs2_compute_bitstructs - Compute the bitmap sizes
742 * @rgd: The resource group descriptor
743 *
744 * Calculates bitmap descriptors, one for each block that contains bitmap data
745 *
746 * Returns: errno
747 */
748
749static int compute_bitstructs(struct gfs2_rgrpd *rgd)
750{
751 struct gfs2_sbd *sdp = rgd->rd_sbd;
752 struct gfs2_bitmap *bi;
753 u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */
754 u32 bytes_left, bytes;
755 int x;
756
757 if (!length)
758 return -EINVAL;
759
760 rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS);
761 if (!rgd->rd_bits)
762 return -ENOMEM;
763
764 bytes_left = rgd->rd_bitbytes;
765
766 for (x = 0; x < length; x++) {
767 bi = rgd->rd_bits + x;
768
769 bi->bi_flags = 0;
770 /* small rgrp; bitmap stored completely in header block */
771 if (length == 1) {
772 bytes = bytes_left;
773 bi->bi_offset = sizeof(struct gfs2_rgrp);
774 bi->bi_start = 0;
775 bi->bi_len = bytes;
776 bi->bi_blocks = bytes * GFS2_NBBY;
777 /* header block */
778 } else if (x == 0) {
779 bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp);
780 bi->bi_offset = sizeof(struct gfs2_rgrp);
781 bi->bi_start = 0;
782 bi->bi_len = bytes;
783 bi->bi_blocks = bytes * GFS2_NBBY;
784 /* last block */
785 } else if (x + 1 == length) {
786 bytes = bytes_left;
787 bi->bi_offset = sizeof(struct gfs2_meta_header);
788 bi->bi_start = rgd->rd_bitbytes - bytes_left;
789 bi->bi_len = bytes;
790 bi->bi_blocks = bytes * GFS2_NBBY;
791 /* other blocks */
792 } else {
793 bytes = sdp->sd_sb.sb_bsize -
794 sizeof(struct gfs2_meta_header);
795 bi->bi_offset = sizeof(struct gfs2_meta_header);
796 bi->bi_start = rgd->rd_bitbytes - bytes_left;
797 bi->bi_len = bytes;
798 bi->bi_blocks = bytes * GFS2_NBBY;
799 }
800
801 bytes_left -= bytes;
802 }
803
804 if (bytes_left) {
805 gfs2_consist_rgrpd(rgd);
806 return -EIO;
807 }
808 bi = rgd->rd_bits + (length - 1);
809 if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) {
810 if (gfs2_consist_rgrpd(rgd)) {
811 gfs2_rindex_print(rgd);
812 fs_err(sdp, "start=%u len=%u offset=%u\n",
813 bi->bi_start, bi->bi_len, bi->bi_offset);
814 }
815 return -EIO;
816 }
817
818 return 0;
819}
820
821/**
822 * gfs2_ri_total - Total up the file system space, according to the rindex.
823 * @sdp: the filesystem
824 *
825 */
826u64 gfs2_ri_total(struct gfs2_sbd *sdp)
827{
828 u64 total_data = 0;
829 struct inode *inode = sdp->sd_rindex;
830 struct gfs2_inode *ip = GFS2_I(inode);
831 char buf[sizeof(struct gfs2_rindex)];
832 int error, rgrps;
833
834 for (rgrps = 0;; rgrps++) {
835 loff_t pos = rgrps * sizeof(struct gfs2_rindex);
836
837 if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode))
838 break;
839 error = gfs2_internal_read(ip, buf, &pos,
840 sizeof(struct gfs2_rindex));
841 if (error != sizeof(struct gfs2_rindex))
842 break;
843 total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data);
844 }
845 return total_data;
846}
847
848static int rgd_insert(struct gfs2_rgrpd *rgd)
849{
850 struct gfs2_sbd *sdp = rgd->rd_sbd;
851 struct rb_node **newn = &sdp->sd_rindex_tree.rb_node, *parent = NULL;
852
853 /* Figure out where to put new node */
854 while (*newn) {
855 struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd,
856 rd_node);
857
858 parent = *newn;
859 if (rgd->rd_addr < cur->rd_addr)
860 newn = &((*newn)->rb_left);
861 else if (rgd->rd_addr > cur->rd_addr)
862 newn = &((*newn)->rb_right);
863 else
864 return -EEXIST;
865 }
866
867 rb_link_node(&rgd->rd_node, parent, newn);
868 rb_insert_color(&rgd->rd_node, &sdp->sd_rindex_tree);
869 sdp->sd_rgrps++;
870 return 0;
871}
872
873/**
874 * read_rindex_entry - Pull in a new resource index entry from the disk
875 * @ip: Pointer to the rindex inode
876 *
877 * Returns: 0 on success, > 0 on EOF, error code otherwise
878 */
879
880static int read_rindex_entry(struct gfs2_inode *ip)
881{
882 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
883 const unsigned bsize = sdp->sd_sb.sb_bsize;
884 loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex);
885 struct gfs2_rindex buf;
886 int error;
887 struct gfs2_rgrpd *rgd;
888
889 if (pos >= i_size_read(&ip->i_inode))
890 return 1;
891
892 error = gfs2_internal_read(ip, (char *)&buf, &pos,
893 sizeof(struct gfs2_rindex));
894
895 if (error != sizeof(struct gfs2_rindex))
896 return (error == 0) ? 1 : error;
897
898 rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS);
899 error = -ENOMEM;
900 if (!rgd)
901 return error;
902
903 rgd->rd_sbd = sdp;
904 rgd->rd_addr = be64_to_cpu(buf.ri_addr);
905 rgd->rd_length = be32_to_cpu(buf.ri_length);
906 rgd->rd_data0 = be64_to_cpu(buf.ri_data0);
907 rgd->rd_data = be32_to_cpu(buf.ri_data);
908 rgd->rd_bitbytes = be32_to_cpu(buf.ri_bitbytes);
909 spin_lock_init(&rgd->rd_rsspin);
910
911 error = compute_bitstructs(rgd);
912 if (error)
913 goto fail;
914
915 error = gfs2_glock_get(sdp, rgd->rd_addr,
916 &gfs2_rgrp_glops, CREATE, &rgd->rd_gl);
917 if (error)
918 goto fail;
919
920 rgd->rd_gl->gl_object = rgd;
921 rgd->rd_gl->gl_vm.start = (rgd->rd_addr * bsize) & PAGE_MASK;
922 rgd->rd_gl->gl_vm.end = PAGE_ALIGN((rgd->rd_addr + rgd->rd_length) * bsize) - 1;
923 rgd->rd_rgl = (struct gfs2_rgrp_lvb *)rgd->rd_gl->gl_lksb.sb_lvbptr;
924 rgd->rd_flags &= ~(GFS2_RDF_UPTODATE | GFS2_RDF_PREFERRED);
925 if (rgd->rd_data > sdp->sd_max_rg_data)
926 sdp->sd_max_rg_data = rgd->rd_data;
927 spin_lock(&sdp->sd_rindex_spin);
928 error = rgd_insert(rgd);
929 spin_unlock(&sdp->sd_rindex_spin);
930 if (!error)
931 return 0;
932
933 error = 0; /* someone else read in the rgrp; free it and ignore it */
934 gfs2_glock_put(rgd->rd_gl);
935
936fail:
937 kfree(rgd->rd_bits);
938 kmem_cache_free(gfs2_rgrpd_cachep, rgd);
939 return error;
940}
941
942/**
943 * set_rgrp_preferences - Run all the rgrps, selecting some we prefer to use
944 * @sdp: the GFS2 superblock
945 *
946 * The purpose of this function is to select a subset of the resource groups
947 * and mark them as PREFERRED. We do it in such a way that each node prefers
948 * to use a unique set of rgrps to minimize glock contention.
949 */
950static void set_rgrp_preferences(struct gfs2_sbd *sdp)
951{
952 struct gfs2_rgrpd *rgd, *first;
953 int i;
954
955 /* Skip an initial number of rgrps, based on this node's journal ID.
956 That should start each node out on its own set. */
957 rgd = gfs2_rgrpd_get_first(sdp);
958 for (i = 0; i < sdp->sd_lockstruct.ls_jid; i++)
959 rgd = gfs2_rgrpd_get_next(rgd);
960 first = rgd;
961
962 do {
963 rgd->rd_flags |= GFS2_RDF_PREFERRED;
964 for (i = 0; i < sdp->sd_journals; i++) {
965 rgd = gfs2_rgrpd_get_next(rgd);
966 if (!rgd || rgd == first)
967 break;
968 }
969 } while (rgd && rgd != first);
970}
971
972/**
973 * gfs2_ri_update - Pull in a new resource index from the disk
974 * @ip: pointer to the rindex inode
975 *
976 * Returns: 0 on successful update, error code otherwise
977 */
978
979static int gfs2_ri_update(struct gfs2_inode *ip)
980{
981 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
982 int error;
983
984 do {
985 error = read_rindex_entry(ip);
986 } while (error == 0);
987
988 if (error < 0)
989 return error;
990
991 set_rgrp_preferences(sdp);
992
993 sdp->sd_rindex_uptodate = 1;
994 return 0;
995}
996
997/**
998 * gfs2_rindex_update - Update the rindex if required
999 * @sdp: The GFS2 superblock
1000 *
1001 * We grab a lock on the rindex inode to make sure that it doesn't
1002 * change whilst we are performing an operation. We keep this lock
1003 * for quite long periods of time compared to other locks. This
1004 * doesn't matter, since it is shared and it is very, very rarely
1005 * accessed in the exclusive mode (i.e. only when expanding the filesystem).
1006 *
1007 * This makes sure that we're using the latest copy of the resource index
1008 * special file, which might have been updated if someone expanded the
1009 * filesystem (via gfs2_grow utility), which adds new resource groups.
1010 *
1011 * Returns: 0 on succeess, error code otherwise
1012 */
1013
1014int gfs2_rindex_update(struct gfs2_sbd *sdp)
1015{
1016 struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex);
1017 struct gfs2_glock *gl = ip->i_gl;
1018 struct gfs2_holder ri_gh;
1019 int error = 0;
1020 int unlock_required = 0;
1021
1022 /* Read new copy from disk if we don't have the latest */
1023 if (!sdp->sd_rindex_uptodate) {
1024 if (!gfs2_glock_is_locked_by_me(gl)) {
1025 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &ri_gh);
1026 if (error)
1027 return error;
1028 unlock_required = 1;
1029 }
1030 if (!sdp->sd_rindex_uptodate)
1031 error = gfs2_ri_update(ip);
1032 if (unlock_required)
1033 gfs2_glock_dq_uninit(&ri_gh);
1034 }
1035
1036 return error;
1037}
1038
1039static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf)
1040{
1041 const struct gfs2_rgrp *str = buf;
1042 u32 rg_flags;
1043
1044 rg_flags = be32_to_cpu(str->rg_flags);
1045 rg_flags &= ~GFS2_RDF_MASK;
1046 rgd->rd_flags &= GFS2_RDF_MASK;
1047 rgd->rd_flags |= rg_flags;
1048 rgd->rd_free = be32_to_cpu(str->rg_free);
1049 rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes);
1050 rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration);
1051}
1052
1053static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf)
1054{
1055 struct gfs2_rgrp *str = buf;
1056
1057 str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK);
1058 str->rg_free = cpu_to_be32(rgd->rd_free);
1059 str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes);
1060 str->__pad = cpu_to_be32(0);
1061 str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration);
1062 memset(&str->rg_reserved, 0, sizeof(str->rg_reserved));
1063}
1064
1065static int gfs2_rgrp_lvb_valid(struct gfs2_rgrpd *rgd)
1066{
1067 struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl;
1068 struct gfs2_rgrp *str = (struct gfs2_rgrp *)rgd->rd_bits[0].bi_bh->b_data;
1069
1070 if (rgl->rl_flags != str->rg_flags || rgl->rl_free != str->rg_free ||
1071 rgl->rl_dinodes != str->rg_dinodes ||
1072 rgl->rl_igeneration != str->rg_igeneration)
1073 return 0;
1074 return 1;
1075}
1076
1077static void gfs2_rgrp_ondisk2lvb(struct gfs2_rgrp_lvb *rgl, const void *buf)
1078{
1079 const struct gfs2_rgrp *str = buf;
1080
1081 rgl->rl_magic = cpu_to_be32(GFS2_MAGIC);
1082 rgl->rl_flags = str->rg_flags;
1083 rgl->rl_free = str->rg_free;
1084 rgl->rl_dinodes = str->rg_dinodes;
1085 rgl->rl_igeneration = str->rg_igeneration;
1086 rgl->__pad = 0UL;
1087}
1088
1089static void update_rgrp_lvb_unlinked(struct gfs2_rgrpd *rgd, u32 change)
1090{
1091 struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl;
1092 u32 unlinked = be32_to_cpu(rgl->rl_unlinked) + change;
1093 rgl->rl_unlinked = cpu_to_be32(unlinked);
1094}
1095
1096static u32 count_unlinked(struct gfs2_rgrpd *rgd)
1097{
1098 struct gfs2_bitmap *bi;
1099 const u32 length = rgd->rd_length;
1100 const u8 *buffer = NULL;
1101 u32 i, goal, count = 0;
1102
1103 for (i = 0, bi = rgd->rd_bits; i < length; i++, bi++) {
1104 goal = 0;
1105 buffer = bi->bi_bh->b_data + bi->bi_offset;
1106 WARN_ON(!buffer_uptodate(bi->bi_bh));
1107 while (goal < bi->bi_len * GFS2_NBBY) {
1108 goal = gfs2_bitfit(buffer, bi->bi_len, goal,
1109 GFS2_BLKST_UNLINKED);
1110 if (goal == BFITNOENT)
1111 break;
1112 count++;
1113 goal++;
1114 }
1115 }
1116
1117 return count;
1118}
1119
1120
1121/**
1122 * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps
1123 * @rgd: the struct gfs2_rgrpd describing the RG to read in
1124 *
1125 * Read in all of a Resource Group's header and bitmap blocks.
1126 * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps.
1127 *
1128 * Returns: errno
1129 */
1130
1131static int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd)
1132{
1133 struct gfs2_sbd *sdp = rgd->rd_sbd;
1134 struct gfs2_glock *gl = rgd->rd_gl;
1135 unsigned int length = rgd->rd_length;
1136 struct gfs2_bitmap *bi;
1137 unsigned int x, y;
1138 int error;
1139
1140 if (rgd->rd_bits[0].bi_bh != NULL)
1141 return 0;
1142
1143 for (x = 0; x < length; x++) {
1144 bi = rgd->rd_bits + x;
1145 error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, 0, &bi->bi_bh);
1146 if (error)
1147 goto fail;
1148 }
1149
1150 for (y = length; y--;) {
1151 bi = rgd->rd_bits + y;
1152 error = gfs2_meta_wait(sdp, bi->bi_bh);
1153 if (error)
1154 goto fail;
1155 if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB :
1156 GFS2_METATYPE_RG)) {
1157 error = -EIO;
1158 goto fail;
1159 }
1160 }
1161
1162 if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) {
1163 for (x = 0; x < length; x++)
1164 clear_bit(GBF_FULL, &rgd->rd_bits[x].bi_flags);
1165 gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data);
1166 rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK);
1167 rgd->rd_free_clone = rgd->rd_free;
1168 /* max out the rgrp allocation failure point */
1169 rgd->rd_extfail_pt = rgd->rd_free;
1170 }
1171 if (cpu_to_be32(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) {
1172 rgd->rd_rgl->rl_unlinked = cpu_to_be32(count_unlinked(rgd));
1173 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl,
1174 rgd->rd_bits[0].bi_bh->b_data);
1175 }
1176 else if (sdp->sd_args.ar_rgrplvb) {
1177 if (!gfs2_rgrp_lvb_valid(rgd)){
1178 gfs2_consist_rgrpd(rgd);
1179 error = -EIO;
1180 goto fail;
1181 }
1182 if (rgd->rd_rgl->rl_unlinked == 0)
1183 rgd->rd_flags &= ~GFS2_RDF_CHECK;
1184 }
1185 return 0;
1186
1187fail:
1188 while (x--) {
1189 bi = rgd->rd_bits + x;
1190 brelse(bi->bi_bh);
1191 bi->bi_bh = NULL;
1192 gfs2_assert_warn(sdp, !bi->bi_clone);
1193 }
1194
1195 return error;
1196}
1197
1198static int update_rgrp_lvb(struct gfs2_rgrpd *rgd)
1199{
1200 u32 rl_flags;
1201
1202 if (rgd->rd_flags & GFS2_RDF_UPTODATE)
1203 return 0;
1204
1205 if (cpu_to_be32(GFS2_MAGIC) != rgd->rd_rgl->rl_magic)
1206 return gfs2_rgrp_bh_get(rgd);
1207
1208 rl_flags = be32_to_cpu(rgd->rd_rgl->rl_flags);
1209 rl_flags &= ~GFS2_RDF_MASK;
1210 rgd->rd_flags &= GFS2_RDF_MASK;
1211 rgd->rd_flags |= (rl_flags | GFS2_RDF_UPTODATE | GFS2_RDF_CHECK);
1212 if (rgd->rd_rgl->rl_unlinked == 0)
1213 rgd->rd_flags &= ~GFS2_RDF_CHECK;
1214 rgd->rd_free = be32_to_cpu(rgd->rd_rgl->rl_free);
1215 rgd->rd_free_clone = rgd->rd_free;
1216 rgd->rd_dinodes = be32_to_cpu(rgd->rd_rgl->rl_dinodes);
1217 rgd->rd_igeneration = be64_to_cpu(rgd->rd_rgl->rl_igeneration);
1218 return 0;
1219}
1220
1221int gfs2_rgrp_go_lock(struct gfs2_holder *gh)
1222{
1223 struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object;
1224 struct gfs2_sbd *sdp = rgd->rd_sbd;
1225
1226 if (gh->gh_flags & GL_SKIP && sdp->sd_args.ar_rgrplvb)
1227 return 0;
1228 return gfs2_rgrp_bh_get(rgd);
1229}
1230
1231/**
1232 * gfs2_rgrp_brelse - Release RG bitmaps read in with gfs2_rgrp_bh_get()
1233 * @rgd: The resource group
1234 *
1235 */
1236
1237void gfs2_rgrp_brelse(struct gfs2_rgrpd *rgd)
1238{
1239 int x, length = rgd->rd_length;
1240
1241 for (x = 0; x < length; x++) {
1242 struct gfs2_bitmap *bi = rgd->rd_bits + x;
1243 if (bi->bi_bh) {
1244 brelse(bi->bi_bh);
1245 bi->bi_bh = NULL;
1246 }
1247 }
1248
1249}
1250
1251/**
1252 * gfs2_rgrp_go_unlock - Unlock a rgrp glock
1253 * @gh: The glock holder for the resource group
1254 *
1255 */
1256
1257void gfs2_rgrp_go_unlock(struct gfs2_holder *gh)
1258{
1259 struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object;
1260 int demote_requested = test_bit(GLF_DEMOTE, &gh->gh_gl->gl_flags) |
1261 test_bit(GLF_PENDING_DEMOTE, &gh->gh_gl->gl_flags);
1262
1263 if (rgd && demote_requested)
1264 gfs2_rgrp_brelse(rgd);
1265}
1266
1267int gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset,
1268 struct buffer_head *bh,
1269 const struct gfs2_bitmap *bi, unsigned minlen, u64 *ptrimmed)
1270{
1271 struct super_block *sb = sdp->sd_vfs;
1272 u64 blk;
1273 sector_t start = 0;
1274 sector_t nr_blks = 0;
1275 int rv;
1276 unsigned int x;
1277 u32 trimmed = 0;
1278 u8 diff;
1279
1280 for (x = 0; x < bi->bi_len; x++) {
1281 const u8 *clone = bi->bi_clone ? bi->bi_clone : bi->bi_bh->b_data;
1282 clone += bi->bi_offset;
1283 clone += x;
1284 if (bh) {
1285 const u8 *orig = bh->b_data + bi->bi_offset + x;
1286 diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1));
1287 } else {
1288 diff = ~(*clone | (*clone >> 1));
1289 }
1290 diff &= 0x55;
1291 if (diff == 0)
1292 continue;
1293 blk = offset + ((bi->bi_start + x) * GFS2_NBBY);
1294 while(diff) {
1295 if (diff & 1) {
1296 if (nr_blks == 0)
1297 goto start_new_extent;
1298 if ((start + nr_blks) != blk) {
1299 if (nr_blks >= minlen) {
1300 rv = sb_issue_discard(sb,
1301 start, nr_blks,
1302 GFP_NOFS, 0);
1303 if (rv)
1304 goto fail;
1305 trimmed += nr_blks;
1306 }
1307 nr_blks = 0;
1308start_new_extent:
1309 start = blk;
1310 }
1311 nr_blks++;
1312 }
1313 diff >>= 2;
1314 blk++;
1315 }
1316 }
1317 if (nr_blks >= minlen) {
1318 rv = sb_issue_discard(sb, start, nr_blks, GFP_NOFS, 0);
1319 if (rv)
1320 goto fail;
1321 trimmed += nr_blks;
1322 }
1323 if (ptrimmed)
1324 *ptrimmed = trimmed;
1325 return 0;
1326
1327fail:
1328 if (sdp->sd_args.ar_discard)
1329 fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv);
1330 sdp->sd_args.ar_discard = 0;
1331 return -EIO;
1332}
1333
1334/**
1335 * gfs2_fitrim - Generate discard requests for unused bits of the filesystem
1336 * @filp: Any file on the filesystem
1337 * @argp: Pointer to the arguments (also used to pass result)
1338 *
1339 * Returns: 0 on success, otherwise error code
1340 */
1341
1342int gfs2_fitrim(struct file *filp, void __user *argp)
1343{
1344 struct inode *inode = file_inode(filp);
1345 struct gfs2_sbd *sdp = GFS2_SB(inode);
1346 struct request_queue *q = bdev_get_queue(sdp->sd_vfs->s_bdev);
1347 struct buffer_head *bh;
1348 struct gfs2_rgrpd *rgd;
1349 struct gfs2_rgrpd *rgd_end;
1350 struct gfs2_holder gh;
1351 struct fstrim_range r;
1352 int ret = 0;
1353 u64 amt;
1354 u64 trimmed = 0;
1355 u64 start, end, minlen;
1356 unsigned int x;
1357 unsigned bs_shift = sdp->sd_sb.sb_bsize_shift;
1358
1359 if (!capable(CAP_SYS_ADMIN))
1360 return -EPERM;
1361
1362 if (!blk_queue_discard(q))
1363 return -EOPNOTSUPP;
1364
1365 if (copy_from_user(&r, argp, sizeof(r)))
1366 return -EFAULT;
1367
1368 ret = gfs2_rindex_update(sdp);
1369 if (ret)
1370 return ret;
1371
1372 start = r.start >> bs_shift;
1373 end = start + (r.len >> bs_shift);
1374 minlen = max_t(u64, r.minlen,
1375 q->limits.discard_granularity) >> bs_shift;
1376
1377 if (end <= start || minlen > sdp->sd_max_rg_data)
1378 return -EINVAL;
1379
1380 rgd = gfs2_blk2rgrpd(sdp, start, 0);
1381 rgd_end = gfs2_blk2rgrpd(sdp, end, 0);
1382
1383 if ((gfs2_rgrpd_get_first(sdp) == gfs2_rgrpd_get_next(rgd_end))
1384 && (start > rgd_end->rd_data0 + rgd_end->rd_data))
1385 return -EINVAL; /* start is beyond the end of the fs */
1386
1387 while (1) {
1388
1389 ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1390 if (ret)
1391 goto out;
1392
1393 if (!(rgd->rd_flags & GFS2_RGF_TRIMMED)) {
1394 /* Trim each bitmap in the rgrp */
1395 for (x = 0; x < rgd->rd_length; x++) {
1396 struct gfs2_bitmap *bi = rgd->rd_bits + x;
1397 ret = gfs2_rgrp_send_discards(sdp,
1398 rgd->rd_data0, NULL, bi, minlen,
1399 &amt);
1400 if (ret) {
1401 gfs2_glock_dq_uninit(&gh);
1402 goto out;
1403 }
1404 trimmed += amt;
1405 }
1406
1407 /* Mark rgrp as having been trimmed */
1408 ret = gfs2_trans_begin(sdp, RES_RG_HDR, 0);
1409 if (ret == 0) {
1410 bh = rgd->rd_bits[0].bi_bh;
1411 rgd->rd_flags |= GFS2_RGF_TRIMMED;
1412 gfs2_trans_add_meta(rgd->rd_gl, bh);
1413 gfs2_rgrp_out(rgd, bh->b_data);
1414 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, bh->b_data);
1415 gfs2_trans_end(sdp);
1416 }
1417 }
1418 gfs2_glock_dq_uninit(&gh);
1419
1420 if (rgd == rgd_end)
1421 break;
1422
1423 rgd = gfs2_rgrpd_get_next(rgd);
1424 }
1425
1426out:
1427 r.len = trimmed << bs_shift;
1428 if (copy_to_user(argp, &r, sizeof(r)))
1429 return -EFAULT;
1430
1431 return ret;
1432}
1433
1434/**
1435 * rs_insert - insert a new multi-block reservation into the rgrp's rb_tree
1436 * @ip: the inode structure
1437 *
1438 */
1439static void rs_insert(struct gfs2_inode *ip)
1440{
1441 struct rb_node **newn, *parent = NULL;
1442 int rc;
1443 struct gfs2_blkreserv *rs = &ip->i_res;
1444 struct gfs2_rgrpd *rgd = rs->rs_rbm.rgd;
1445 u64 fsblock = gfs2_rbm_to_block(&rs->rs_rbm);
1446
1447 BUG_ON(gfs2_rs_active(rs));
1448
1449 spin_lock(&rgd->rd_rsspin);
1450 newn = &rgd->rd_rstree.rb_node;
1451 while (*newn) {
1452 struct gfs2_blkreserv *cur =
1453 rb_entry(*newn, struct gfs2_blkreserv, rs_node);
1454
1455 parent = *newn;
1456 rc = rs_cmp(fsblock, rs->rs_free, cur);
1457 if (rc > 0)
1458 newn = &((*newn)->rb_right);
1459 else if (rc < 0)
1460 newn = &((*newn)->rb_left);
1461 else {
1462 spin_unlock(&rgd->rd_rsspin);
1463 WARN_ON(1);
1464 return;
1465 }
1466 }
1467
1468 rb_link_node(&rs->rs_node, parent, newn);
1469 rb_insert_color(&rs->rs_node, &rgd->rd_rstree);
1470
1471 /* Do our rgrp accounting for the reservation */
1472 rgd->rd_reserved += rs->rs_free; /* blocks reserved */
1473 spin_unlock(&rgd->rd_rsspin);
1474 trace_gfs2_rs(rs, TRACE_RS_INSERT);
1475}
1476
1477/**
1478 * rg_mblk_search - find a group of multiple free blocks to form a reservation
1479 * @rgd: the resource group descriptor
1480 * @ip: pointer to the inode for which we're reserving blocks
1481 * @ap: the allocation parameters
1482 *
1483 */
1484
1485static void rg_mblk_search(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip,
1486 const struct gfs2_alloc_parms *ap)
1487{
1488 struct gfs2_rbm rbm = { .rgd = rgd, };
1489 u64 goal;
1490 struct gfs2_blkreserv *rs = &ip->i_res;
1491 u32 extlen;
1492 u32 free_blocks = rgd->rd_free_clone - rgd->rd_reserved;
1493 int ret;
1494 struct inode *inode = &ip->i_inode;
1495
1496 if (S_ISDIR(inode->i_mode))
1497 extlen = 1;
1498 else {
1499 extlen = max_t(u32, atomic_read(&rs->rs_sizehint), ap->target);
1500 extlen = clamp(extlen, RGRP_RSRV_MINBLKS, free_blocks);
1501 }
1502 if ((rgd->rd_free_clone < rgd->rd_reserved) || (free_blocks < extlen))
1503 return;
1504
1505 /* Find bitmap block that contains bits for goal block */
1506 if (rgrp_contains_block(rgd, ip->i_goal))
1507 goal = ip->i_goal;
1508 else
1509 goal = rgd->rd_last_alloc + rgd->rd_data0;
1510
1511 if (WARN_ON(gfs2_rbm_from_block(&rbm, goal)))
1512 return;
1513
1514 ret = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, &extlen, ip, true, ap);
1515 if (ret == 0) {
1516 rs->rs_rbm = rbm;
1517 rs->rs_free = extlen;
1518 rs->rs_inum = ip->i_no_addr;
1519 rs_insert(ip);
1520 } else {
1521 if (goal == rgd->rd_last_alloc + rgd->rd_data0)
1522 rgd->rd_last_alloc = 0;
1523 }
1524}
1525
1526/**
1527 * gfs2_next_unreserved_block - Return next block that is not reserved
1528 * @rgd: The resource group
1529 * @block: The starting block
1530 * @length: The required length
1531 * @ip: Ignore any reservations for this inode
1532 *
1533 * If the block does not appear in any reservation, then return the
1534 * block number unchanged. If it does appear in the reservation, then
1535 * keep looking through the tree of reservations in order to find the
1536 * first block number which is not reserved.
1537 */
1538
1539static u64 gfs2_next_unreserved_block(struct gfs2_rgrpd *rgd, u64 block,
1540 u32 length,
1541 const struct gfs2_inode *ip)
1542{
1543 struct gfs2_blkreserv *rs;
1544 struct rb_node *n;
1545 int rc;
1546
1547 spin_lock(&rgd->rd_rsspin);
1548 n = rgd->rd_rstree.rb_node;
1549 while (n) {
1550 rs = rb_entry(n, struct gfs2_blkreserv, rs_node);
1551 rc = rs_cmp(block, length, rs);
1552 if (rc < 0)
1553 n = n->rb_left;
1554 else if (rc > 0)
1555 n = n->rb_right;
1556 else
1557 break;
1558 }
1559
1560 if (n) {
1561 while ((rs_cmp(block, length, rs) == 0) && (&ip->i_res != rs)) {
1562 block = gfs2_rbm_to_block(&rs->rs_rbm) + rs->rs_free;
1563 n = n->rb_right;
1564 if (n == NULL)
1565 break;
1566 rs = rb_entry(n, struct gfs2_blkreserv, rs_node);
1567 }
1568 }
1569
1570 spin_unlock(&rgd->rd_rsspin);
1571 return block;
1572}
1573
1574/**
1575 * gfs2_reservation_check_and_update - Check for reservations during block alloc
1576 * @rbm: The current position in the resource group
1577 * @ip: The inode for which we are searching for blocks
1578 * @minext: The minimum extent length
1579 * @maxext: A pointer to the maximum extent structure
1580 *
1581 * This checks the current position in the rgrp to see whether there is
1582 * a reservation covering this block. If not then this function is a
1583 * no-op. If there is, then the position is moved to the end of the
1584 * contiguous reservation(s) so that we are pointing at the first
1585 * non-reserved block.
1586 *
1587 * Returns: 0 if no reservation, 1 if @rbm has changed, otherwise an error
1588 */
1589
1590static int gfs2_reservation_check_and_update(struct gfs2_rbm *rbm,
1591 const struct gfs2_inode *ip,
1592 u32 minext,
1593 struct gfs2_extent *maxext)
1594{
1595 u64 block = gfs2_rbm_to_block(rbm);
1596 u32 extlen = 1;
1597 u64 nblock;
1598 int ret;
1599
1600 /*
1601 * If we have a minimum extent length, then skip over any extent
1602 * which is less than the min extent length in size.
1603 */
1604 if (minext) {
1605 extlen = gfs2_free_extlen(rbm, minext);
1606 if (extlen <= maxext->len)
1607 goto fail;
1608 }
1609
1610 /*
1611 * Check the extent which has been found against the reservations
1612 * and skip if parts of it are already reserved
1613 */
1614 nblock = gfs2_next_unreserved_block(rbm->rgd, block, extlen, ip);
1615 if (nblock == block) {
1616 if (!minext || extlen >= minext)
1617 return 0;
1618
1619 if (extlen > maxext->len) {
1620 maxext->len = extlen;
1621 maxext->rbm = *rbm;
1622 }
1623fail:
1624 nblock = block + extlen;
1625 }
1626 ret = gfs2_rbm_from_block(rbm, nblock);
1627 if (ret < 0)
1628 return ret;
1629 return 1;
1630}
1631
1632/**
1633 * gfs2_rbm_find - Look for blocks of a particular state
1634 * @rbm: Value/result starting position and final position
1635 * @state: The state which we want to find
1636 * @minext: Pointer to the requested extent length (NULL for a single block)
1637 * This is updated to be the actual reservation size.
1638 * @ip: If set, check for reservations
1639 * @nowrap: Stop looking at the end of the rgrp, rather than wrapping
1640 * around until we've reached the starting point.
1641 * @ap: the allocation parameters
1642 *
1643 * Side effects:
1644 * - If looking for free blocks, we set GBF_FULL on each bitmap which
1645 * has no free blocks in it.
1646 * - If looking for free blocks, we set rd_extfail_pt on each rgrp which
1647 * has come up short on a free block search.
1648 *
1649 * Returns: 0 on success, -ENOSPC if there is no block of the requested state
1650 */
1651
1652static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 *minext,
1653 const struct gfs2_inode *ip, bool nowrap,
1654 const struct gfs2_alloc_parms *ap)
1655{
1656 struct buffer_head *bh;
1657 int initial_bii;
1658 u32 initial_offset;
1659 int first_bii = rbm->bii;
1660 u32 first_offset = rbm->offset;
1661 u32 offset;
1662 u8 *buffer;
1663 int n = 0;
1664 int iters = rbm->rgd->rd_length;
1665 int ret;
1666 struct gfs2_bitmap *bi;
1667 struct gfs2_extent maxext = { .rbm.rgd = rbm->rgd, };
1668
1669 /* If we are not starting at the beginning of a bitmap, then we
1670 * need to add one to the bitmap count to ensure that we search
1671 * the starting bitmap twice.
1672 */
1673 if (rbm->offset != 0)
1674 iters++;
1675
1676 while(1) {
1677 bi = rbm_bi(rbm);
1678 if (test_bit(GBF_FULL, &bi->bi_flags) &&
1679 (state == GFS2_BLKST_FREE))
1680 goto next_bitmap;
1681
1682 bh = bi->bi_bh;
1683 buffer = bh->b_data + bi->bi_offset;
1684 WARN_ON(!buffer_uptodate(bh));
1685 if (state != GFS2_BLKST_UNLINKED && bi->bi_clone)
1686 buffer = bi->bi_clone + bi->bi_offset;
1687 initial_offset = rbm->offset;
1688 offset = gfs2_bitfit(buffer, bi->bi_len, rbm->offset, state);
1689 if (offset == BFITNOENT)
1690 goto bitmap_full;
1691 rbm->offset = offset;
1692 if (ip == NULL)
1693 return 0;
1694
1695 initial_bii = rbm->bii;
1696 ret = gfs2_reservation_check_and_update(rbm, ip,
1697 minext ? *minext : 0,
1698 &maxext);
1699 if (ret == 0)
1700 return 0;
1701 if (ret > 0) {
1702 n += (rbm->bii - initial_bii);
1703 goto next_iter;
1704 }
1705 if (ret == -E2BIG) {
1706 rbm->bii = 0;
1707 rbm->offset = 0;
1708 n += (rbm->bii - initial_bii);
1709 goto res_covered_end_of_rgrp;
1710 }
1711 return ret;
1712
1713bitmap_full: /* Mark bitmap as full and fall through */
1714 if ((state == GFS2_BLKST_FREE) && initial_offset == 0)
1715 set_bit(GBF_FULL, &bi->bi_flags);
1716
1717next_bitmap: /* Find next bitmap in the rgrp */
1718 rbm->offset = 0;
1719 rbm->bii++;
1720 if (rbm->bii == rbm->rgd->rd_length)
1721 rbm->bii = 0;
1722res_covered_end_of_rgrp:
1723 if ((rbm->bii == 0) && nowrap)
1724 break;
1725 n++;
1726next_iter:
1727 if (n >= iters)
1728 break;
1729 }
1730
1731 if (minext == NULL || state != GFS2_BLKST_FREE)
1732 return -ENOSPC;
1733
1734 /* If the extent was too small, and it's smaller than the smallest
1735 to have failed before, remember for future reference that it's
1736 useless to search this rgrp again for this amount or more. */
1737 if ((first_offset == 0) && (first_bii == 0) &&
1738 (*minext < rbm->rgd->rd_extfail_pt))
1739 rbm->rgd->rd_extfail_pt = *minext;
1740
1741 /* If the maximum extent we found is big enough to fulfill the
1742 minimum requirements, use it anyway. */
1743 if (maxext.len) {
1744 *rbm = maxext.rbm;
1745 *minext = maxext.len;
1746 return 0;
1747 }
1748
1749 return -ENOSPC;
1750}
1751
1752/**
1753 * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes
1754 * @rgd: The rgrp
1755 * @last_unlinked: block address of the last dinode we unlinked
1756 * @skip: block address we should explicitly not unlink
1757 *
1758 * Returns: 0 if no error
1759 * The inode, if one has been found, in inode.
1760 */
1761
1762static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip)
1763{
1764 u64 block;
1765 struct gfs2_sbd *sdp = rgd->rd_sbd;
1766 struct gfs2_glock *gl;
1767 struct gfs2_inode *ip;
1768 int error;
1769 int found = 0;
1770 struct gfs2_rbm rbm = { .rgd = rgd, .bii = 0, .offset = 0 };
1771
1772 while (1) {
1773 down_write(&sdp->sd_log_flush_lock);
1774 error = gfs2_rbm_find(&rbm, GFS2_BLKST_UNLINKED, NULL, NULL,
1775 true, NULL);
1776 up_write(&sdp->sd_log_flush_lock);
1777 if (error == -ENOSPC)
1778 break;
1779 if (WARN_ON_ONCE(error))
1780 break;
1781
1782 block = gfs2_rbm_to_block(&rbm);
1783 if (gfs2_rbm_from_block(&rbm, block + 1))
1784 break;
1785 if (*last_unlinked != NO_BLOCK && block <= *last_unlinked)
1786 continue;
1787 if (block == skip)
1788 continue;
1789 *last_unlinked = block;
1790
1791 error = gfs2_glock_get(sdp, block, &gfs2_iopen_glops, CREATE, &gl);
1792 if (error)
1793 continue;
1794
1795 /* If the inode is already in cache, we can ignore it here
1796 * because the existing inode disposal code will deal with
1797 * it when all refs have gone away. Accessing gl_object like
1798 * this is not safe in general. Here it is ok because we do
1799 * not dereference the pointer, and we only need an approx
1800 * answer to whether it is NULL or not.
1801 */
1802 ip = gl->gl_object;
1803
1804 if (ip || queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
1805 gfs2_glock_put(gl);
1806 else
1807 found++;
1808
1809 /* Limit reclaim to sensible number of tasks */
1810 if (found > NR_CPUS)
1811 return;
1812 }
1813
1814 rgd->rd_flags &= ~GFS2_RDF_CHECK;
1815 return;
1816}
1817
1818/**
1819 * gfs2_rgrp_congested - Use stats to figure out whether an rgrp is congested
1820 * @rgd: The rgrp in question
1821 * @loops: An indication of how picky we can be (0=very, 1=less so)
1822 *
1823 * This function uses the recently added glock statistics in order to
1824 * figure out whether a parciular resource group is suffering from
1825 * contention from multiple nodes. This is done purely on the basis
1826 * of timings, since this is the only data we have to work with and
1827 * our aim here is to reject a resource group which is highly contended
1828 * but (very important) not to do this too often in order to ensure that
1829 * we do not land up introducing fragmentation by changing resource
1830 * groups when not actually required.
1831 *
1832 * The calculation is fairly simple, we want to know whether the SRTTB
1833 * (i.e. smoothed round trip time for blocking operations) to acquire
1834 * the lock for this rgrp's glock is significantly greater than the
1835 * time taken for resource groups on average. We introduce a margin in
1836 * the form of the variable @var which is computed as the sum of the two
1837 * respective variences, and multiplied by a factor depending on @loops
1838 * and whether we have a lot of data to base the decision on. This is
1839 * then tested against the square difference of the means in order to
1840 * decide whether the result is statistically significant or not.
1841 *
1842 * Returns: A boolean verdict on the congestion status
1843 */
1844
1845static bool gfs2_rgrp_congested(const struct gfs2_rgrpd *rgd, int loops)
1846{
1847 const struct gfs2_glock *gl = rgd->rd_gl;
1848 const struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1849 struct gfs2_lkstats *st;
1850 u64 r_dcount, l_dcount;
1851 u64 l_srttb, a_srttb = 0;
1852 s64 srttb_diff;
1853 u64 sqr_diff;
1854 u64 var;
1855 int cpu, nonzero = 0;
1856
1857 preempt_disable();
1858 for_each_present_cpu(cpu) {
1859 st = &per_cpu_ptr(sdp->sd_lkstats, cpu)->lkstats[LM_TYPE_RGRP];
1860 if (st->stats[GFS2_LKS_SRTTB]) {
1861 a_srttb += st->stats[GFS2_LKS_SRTTB];
1862 nonzero++;
1863 }
1864 }
1865 st = &this_cpu_ptr(sdp->sd_lkstats)->lkstats[LM_TYPE_RGRP];
1866 if (nonzero)
1867 do_div(a_srttb, nonzero);
1868 r_dcount = st->stats[GFS2_LKS_DCOUNT];
1869 var = st->stats[GFS2_LKS_SRTTVARB] +
1870 gl->gl_stats.stats[GFS2_LKS_SRTTVARB];
1871 preempt_enable();
1872
1873 l_srttb = gl->gl_stats.stats[GFS2_LKS_SRTTB];
1874 l_dcount = gl->gl_stats.stats[GFS2_LKS_DCOUNT];
1875
1876 if ((l_dcount < 1) || (r_dcount < 1) || (a_srttb == 0))
1877 return false;
1878
1879 srttb_diff = a_srttb - l_srttb;
1880 sqr_diff = srttb_diff * srttb_diff;
1881
1882 var *= 2;
1883 if (l_dcount < 8 || r_dcount < 8)
1884 var *= 2;
1885 if (loops == 1)
1886 var *= 2;
1887
1888 return ((srttb_diff < 0) && (sqr_diff > var));
1889}
1890
1891/**
1892 * gfs2_rgrp_used_recently
1893 * @rs: The block reservation with the rgrp to test
1894 * @msecs: The time limit in milliseconds
1895 *
1896 * Returns: True if the rgrp glock has been used within the time limit
1897 */
1898static bool gfs2_rgrp_used_recently(const struct gfs2_blkreserv *rs,
1899 u64 msecs)
1900{
1901 u64 tdiff;
1902
1903 tdiff = ktime_to_ns(ktime_sub(ktime_get_real(),
1904 rs->rs_rbm.rgd->rd_gl->gl_dstamp));
1905
1906 return tdiff > (msecs * 1000 * 1000);
1907}
1908
1909static u32 gfs2_orlov_skip(const struct gfs2_inode *ip)
1910{
1911 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1912 u32 skip;
1913
1914 get_random_bytes(&skip, sizeof(skip));
1915 return skip % sdp->sd_rgrps;
1916}
1917
1918static bool gfs2_select_rgrp(struct gfs2_rgrpd **pos, const struct gfs2_rgrpd *begin)
1919{
1920 struct gfs2_rgrpd *rgd = *pos;
1921 struct gfs2_sbd *sdp = rgd->rd_sbd;
1922
1923 rgd = gfs2_rgrpd_get_next(rgd);
1924 if (rgd == NULL)
1925 rgd = gfs2_rgrpd_get_first(sdp);
1926 *pos = rgd;
1927 if (rgd != begin) /* If we didn't wrap */
1928 return true;
1929 return false;
1930}
1931
1932/**
1933 * fast_to_acquire - determine if a resource group will be fast to acquire
1934 *
1935 * If this is one of our preferred rgrps, it should be quicker to acquire,
1936 * because we tried to set ourselves up as dlm lock master.
1937 */
1938static inline int fast_to_acquire(struct gfs2_rgrpd *rgd)
1939{
1940 struct gfs2_glock *gl = rgd->rd_gl;
1941
1942 if (gl->gl_state != LM_ST_UNLOCKED && list_empty(&gl->gl_holders) &&
1943 !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
1944 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1945 return 1;
1946 if (rgd->rd_flags & GFS2_RDF_PREFERRED)
1947 return 1;
1948 return 0;
1949}
1950
1951/**
1952 * gfs2_inplace_reserve - Reserve space in the filesystem
1953 * @ip: the inode to reserve space for
1954 * @ap: the allocation parameters
1955 *
1956 * We try our best to find an rgrp that has at least ap->target blocks
1957 * available. After a couple of passes (loops == 2), the prospects of finding
1958 * such an rgrp diminish. At this stage, we return the first rgrp that has
1959 * atleast ap->min_target blocks available. Either way, we set ap->allowed to
1960 * the number of blocks available in the chosen rgrp.
1961 *
1962 * Returns: 0 on success,
1963 * -ENOMEM if a suitable rgrp can't be found
1964 * errno otherwise
1965 */
1966
1967int gfs2_inplace_reserve(struct gfs2_inode *ip, struct gfs2_alloc_parms *ap)
1968{
1969 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1970 struct gfs2_rgrpd *begin = NULL;
1971 struct gfs2_blkreserv *rs = &ip->i_res;
1972 int error = 0, rg_locked, flags = 0;
1973 u64 last_unlinked = NO_BLOCK;
1974 int loops = 0;
1975 u32 skip = 0;
1976
1977 if (sdp->sd_args.ar_rgrplvb)
1978 flags |= GL_SKIP;
1979 if (gfs2_assert_warn(sdp, ap->target))
1980 return -EINVAL;
1981 if (gfs2_rs_active(rs)) {
1982 begin = rs->rs_rbm.rgd;
1983 } else if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, ip->i_goal)) {
1984 rs->rs_rbm.rgd = begin = ip->i_rgd;
1985 } else {
1986 check_and_update_goal(ip);
1987 rs->rs_rbm.rgd = begin = gfs2_blk2rgrpd(sdp, ip->i_goal, 1);
1988 }
1989 if (S_ISDIR(ip->i_inode.i_mode) && (ap->aflags & GFS2_AF_ORLOV))
1990 skip = gfs2_orlov_skip(ip);
1991 if (rs->rs_rbm.rgd == NULL)
1992 return -EBADSLT;
1993
1994 while (loops < 3) {
1995 rg_locked = 1;
1996
1997 if (!gfs2_glock_is_locked_by_me(rs->rs_rbm.rgd->rd_gl)) {
1998 rg_locked = 0;
1999 if (skip && skip--)
2000 goto next_rgrp;
2001 if (!gfs2_rs_active(rs)) {
2002 if (loops == 0 &&
2003 !fast_to_acquire(rs->rs_rbm.rgd))
2004 goto next_rgrp;
2005 if ((loops < 2) &&
2006 gfs2_rgrp_used_recently(rs, 1000) &&
2007 gfs2_rgrp_congested(rs->rs_rbm.rgd, loops))
2008 goto next_rgrp;
2009 }
2010 error = gfs2_glock_nq_init(rs->rs_rbm.rgd->rd_gl,
2011 LM_ST_EXCLUSIVE, flags,
2012 &rs->rs_rgd_gh);
2013 if (unlikely(error))
2014 return error;
2015 if (!gfs2_rs_active(rs) && (loops < 2) &&
2016 gfs2_rgrp_congested(rs->rs_rbm.rgd, loops))
2017 goto skip_rgrp;
2018 if (sdp->sd_args.ar_rgrplvb) {
2019 error = update_rgrp_lvb(rs->rs_rbm.rgd);
2020 if (unlikely(error)) {
2021 gfs2_glock_dq_uninit(&rs->rs_rgd_gh);
2022 return error;
2023 }
2024 }
2025 }
2026
2027 /* Skip unuseable resource groups */
2028 if ((rs->rs_rbm.rgd->rd_flags & (GFS2_RGF_NOALLOC |
2029 GFS2_RDF_ERROR)) ||
2030 (loops == 0 && ap->target > rs->rs_rbm.rgd->rd_extfail_pt))
2031 goto skip_rgrp;
2032
2033 if (sdp->sd_args.ar_rgrplvb)
2034 gfs2_rgrp_bh_get(rs->rs_rbm.rgd);
2035
2036 /* Get a reservation if we don't already have one */
2037 if (!gfs2_rs_active(rs))
2038 rg_mblk_search(rs->rs_rbm.rgd, ip, ap);
2039
2040 /* Skip rgrps when we can't get a reservation on first pass */
2041 if (!gfs2_rs_active(rs) && (loops < 1))
2042 goto check_rgrp;
2043
2044 /* If rgrp has enough free space, use it */
2045 if (rs->rs_rbm.rgd->rd_free_clone >= ap->target ||
2046 (loops == 2 && ap->min_target &&
2047 rs->rs_rbm.rgd->rd_free_clone >= ap->min_target)) {
2048 ip->i_rgd = rs->rs_rbm.rgd;
2049 ap->allowed = ip->i_rgd->rd_free_clone;
2050 return 0;
2051 }
2052check_rgrp:
2053 /* Check for unlinked inodes which can be reclaimed */
2054 if (rs->rs_rbm.rgd->rd_flags & GFS2_RDF_CHECK)
2055 try_rgrp_unlink(rs->rs_rbm.rgd, &last_unlinked,
2056 ip->i_no_addr);
2057skip_rgrp:
2058 /* Drop reservation, if we couldn't use reserved rgrp */
2059 if (gfs2_rs_active(rs))
2060 gfs2_rs_deltree(rs);
2061
2062 /* Unlock rgrp if required */
2063 if (!rg_locked)
2064 gfs2_glock_dq_uninit(&rs->rs_rgd_gh);
2065next_rgrp:
2066 /* Find the next rgrp, and continue looking */
2067 if (gfs2_select_rgrp(&rs->rs_rbm.rgd, begin))
2068 continue;
2069 if (skip)
2070 continue;
2071
2072 /* If we've scanned all the rgrps, but found no free blocks
2073 * then this checks for some less likely conditions before
2074 * trying again.
2075 */
2076 loops++;
2077 /* Check that fs hasn't grown if writing to rindex */
2078 if (ip == GFS2_I(sdp->sd_rindex) && !sdp->sd_rindex_uptodate) {
2079 error = gfs2_ri_update(ip);
2080 if (error)
2081 return error;
2082 }
2083 /* Flushing the log may release space */
2084 if (loops == 2)
2085 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
2086 }
2087
2088 return -ENOSPC;
2089}
2090
2091/**
2092 * gfs2_inplace_release - release an inplace reservation
2093 * @ip: the inode the reservation was taken out on
2094 *
2095 * Release a reservation made by gfs2_inplace_reserve().
2096 */
2097
2098void gfs2_inplace_release(struct gfs2_inode *ip)
2099{
2100 struct gfs2_blkreserv *rs = &ip->i_res;
2101
2102 if (rs->rs_rgd_gh.gh_gl)
2103 gfs2_glock_dq_uninit(&rs->rs_rgd_gh);
2104}
2105
2106/**
2107 * gfs2_get_block_type - Check a block in a RG is of given type
2108 * @rgd: the resource group holding the block
2109 * @block: the block number
2110 *
2111 * Returns: The block type (GFS2_BLKST_*)
2112 */
2113
2114static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block)
2115{
2116 struct gfs2_rbm rbm = { .rgd = rgd, };
2117 int ret;
2118
2119 ret = gfs2_rbm_from_block(&rbm, block);
2120 WARN_ON_ONCE(ret != 0);
2121
2122 return gfs2_testbit(&rbm);
2123}
2124
2125
2126/**
2127 * gfs2_alloc_extent - allocate an extent from a given bitmap
2128 * @rbm: the resource group information
2129 * @dinode: TRUE if the first block we allocate is for a dinode
2130 * @n: The extent length (value/result)
2131 *
2132 * Add the bitmap buffer to the transaction.
2133 * Set the found bits to @new_state to change block's allocation state.
2134 */
2135static void gfs2_alloc_extent(const struct gfs2_rbm *rbm, bool dinode,
2136 unsigned int *n)
2137{
2138 struct gfs2_rbm pos = { .rgd = rbm->rgd, };
2139 const unsigned int elen = *n;
2140 u64 block;
2141 int ret;
2142
2143 *n = 1;
2144 block = gfs2_rbm_to_block(rbm);
2145 gfs2_trans_add_meta(rbm->rgd->rd_gl, rbm_bi(rbm)->bi_bh);
2146 gfs2_setbit(rbm, true, dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED);
2147 block++;
2148 while (*n < elen) {
2149 ret = gfs2_rbm_from_block(&pos, block);
2150 if (ret || gfs2_testbit(&pos) != GFS2_BLKST_FREE)
2151 break;
2152 gfs2_trans_add_meta(pos.rgd->rd_gl, rbm_bi(&pos)->bi_bh);
2153 gfs2_setbit(&pos, true, GFS2_BLKST_USED);
2154 (*n)++;
2155 block++;
2156 }
2157}
2158
2159/**
2160 * rgblk_free - Change alloc state of given block(s)
2161 * @sdp: the filesystem
2162 * @bstart: the start of a run of blocks to free
2163 * @blen: the length of the block run (all must lie within ONE RG!)
2164 * @new_state: GFS2_BLKST_XXX the after-allocation block state
2165 *
2166 * Returns: Resource group containing the block(s)
2167 */
2168
2169static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart,
2170 u32 blen, unsigned char new_state)
2171{
2172 struct gfs2_rbm rbm;
2173 struct gfs2_bitmap *bi, *bi_prev = NULL;
2174
2175 rbm.rgd = gfs2_blk2rgrpd(sdp, bstart, 1);
2176 if (!rbm.rgd) {
2177 if (gfs2_consist(sdp))
2178 fs_err(sdp, "block = %llu\n", (unsigned long long)bstart);
2179 return NULL;
2180 }
2181
2182 gfs2_rbm_from_block(&rbm, bstart);
2183 while (blen--) {
2184 bi = rbm_bi(&rbm);
2185 if (bi != bi_prev) {
2186 if (!bi->bi_clone) {
2187 bi->bi_clone = kmalloc(bi->bi_bh->b_size,
2188 GFP_NOFS | __GFP_NOFAIL);
2189 memcpy(bi->bi_clone + bi->bi_offset,
2190 bi->bi_bh->b_data + bi->bi_offset,
2191 bi->bi_len);
2192 }
2193 gfs2_trans_add_meta(rbm.rgd->rd_gl, bi->bi_bh);
2194 bi_prev = bi;
2195 }
2196 gfs2_setbit(&rbm, false, new_state);
2197 gfs2_rbm_incr(&rbm);
2198 }
2199
2200 return rbm.rgd;
2201}
2202
2203/**
2204 * gfs2_rgrp_dump - print out an rgrp
2205 * @seq: The iterator
2206 * @gl: The glock in question
2207 *
2208 */
2209
2210void gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl)
2211{
2212 struct gfs2_rgrpd *rgd = gl->gl_object;
2213 struct gfs2_blkreserv *trs;
2214 const struct rb_node *n;
2215
2216 if (rgd == NULL)
2217 return;
2218 gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u r:%u e:%u\n",
2219 (unsigned long long)rgd->rd_addr, rgd->rd_flags,
2220 rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes,
2221 rgd->rd_reserved, rgd->rd_extfail_pt);
2222 spin_lock(&rgd->rd_rsspin);
2223 for (n = rb_first(&rgd->rd_rstree); n; n = rb_next(&trs->rs_node)) {
2224 trs = rb_entry(n, struct gfs2_blkreserv, rs_node);
2225 dump_rs(seq, trs);
2226 }
2227 spin_unlock(&rgd->rd_rsspin);
2228}
2229
2230static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd)
2231{
2232 struct gfs2_sbd *sdp = rgd->rd_sbd;
2233 fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n",
2234 (unsigned long long)rgd->rd_addr);
2235 fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n");
2236 gfs2_rgrp_dump(NULL, rgd->rd_gl);
2237 rgd->rd_flags |= GFS2_RDF_ERROR;
2238}
2239
2240/**
2241 * gfs2_adjust_reservation - Adjust (or remove) a reservation after allocation
2242 * @ip: The inode we have just allocated blocks for
2243 * @rbm: The start of the allocated blocks
2244 * @len: The extent length
2245 *
2246 * Adjusts a reservation after an allocation has taken place. If the
2247 * reservation does not match the allocation, or if it is now empty
2248 * then it is removed.
2249 */
2250
2251static void gfs2_adjust_reservation(struct gfs2_inode *ip,
2252 const struct gfs2_rbm *rbm, unsigned len)
2253{
2254 struct gfs2_blkreserv *rs = &ip->i_res;
2255 struct gfs2_rgrpd *rgd = rbm->rgd;
2256 unsigned rlen;
2257 u64 block;
2258 int ret;
2259
2260 spin_lock(&rgd->rd_rsspin);
2261 if (gfs2_rs_active(rs)) {
2262 if (gfs2_rbm_eq(&rs->rs_rbm, rbm)) {
2263 block = gfs2_rbm_to_block(rbm);
2264 ret = gfs2_rbm_from_block(&rs->rs_rbm, block + len);
2265 rlen = min(rs->rs_free, len);
2266 rs->rs_free -= rlen;
2267 rgd->rd_reserved -= rlen;
2268 trace_gfs2_rs(rs, TRACE_RS_CLAIM);
2269 if (rs->rs_free && !ret)
2270 goto out;
2271 /* We used up our block reservation, so we should
2272 reserve more blocks next time. */
2273 atomic_add(RGRP_RSRV_ADDBLKS, &rs->rs_sizehint);
2274 }
2275 __rs_deltree(rs);
2276 }
2277out:
2278 spin_unlock(&rgd->rd_rsspin);
2279}
2280
2281/**
2282 * gfs2_set_alloc_start - Set starting point for block allocation
2283 * @rbm: The rbm which will be set to the required location
2284 * @ip: The gfs2 inode
2285 * @dinode: Flag to say if allocation includes a new inode
2286 *
2287 * This sets the starting point from the reservation if one is active
2288 * otherwise it falls back to guessing a start point based on the
2289 * inode's goal block or the last allocation point in the rgrp.
2290 */
2291
2292static void gfs2_set_alloc_start(struct gfs2_rbm *rbm,
2293 const struct gfs2_inode *ip, bool dinode)
2294{
2295 u64 goal;
2296
2297 if (gfs2_rs_active(&ip->i_res)) {
2298 *rbm = ip->i_res.rs_rbm;
2299 return;
2300 }
2301
2302 if (!dinode && rgrp_contains_block(rbm->rgd, ip->i_goal))
2303 goal = ip->i_goal;
2304 else
2305 goal = rbm->rgd->rd_last_alloc + rbm->rgd->rd_data0;
2306
2307 gfs2_rbm_from_block(rbm, goal);
2308}
2309
2310/**
2311 * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode
2312 * @ip: the inode to allocate the block for
2313 * @bn: Used to return the starting block number
2314 * @nblocks: requested number of blocks/extent length (value/result)
2315 * @dinode: 1 if we're allocating a dinode block, else 0
2316 * @generation: the generation number of the inode
2317 *
2318 * Returns: 0 or error
2319 */
2320
2321int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *nblocks,
2322 bool dinode, u64 *generation)
2323{
2324 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2325 struct buffer_head *dibh;
2326 struct gfs2_rbm rbm = { .rgd = ip->i_rgd, };
2327 unsigned int ndata;
2328 u64 block; /* block, within the file system scope */
2329 int error;
2330
2331 gfs2_set_alloc_start(&rbm, ip, dinode);
2332 error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, NULL, ip, false, NULL);
2333
2334 if (error == -ENOSPC) {
2335 gfs2_set_alloc_start(&rbm, ip, dinode);
2336 error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, NULL, NULL, false,
2337 NULL);
2338 }
2339
2340 /* Since all blocks are reserved in advance, this shouldn't happen */
2341 if (error) {
2342 fs_warn(sdp, "inum=%llu error=%d, nblocks=%u, full=%d fail_pt=%d\n",
2343 (unsigned long long)ip->i_no_addr, error, *nblocks,
2344 test_bit(GBF_FULL, &rbm.rgd->rd_bits->bi_flags),
2345 rbm.rgd->rd_extfail_pt);
2346 goto rgrp_error;
2347 }
2348
2349 gfs2_alloc_extent(&rbm, dinode, nblocks);
2350 block = gfs2_rbm_to_block(&rbm);
2351 rbm.rgd->rd_last_alloc = block - rbm.rgd->rd_data0;
2352 if (gfs2_rs_active(&ip->i_res))
2353 gfs2_adjust_reservation(ip, &rbm, *nblocks);
2354 ndata = *nblocks;
2355 if (dinode)
2356 ndata--;
2357
2358 if (!dinode) {
2359 ip->i_goal = block + ndata - 1;
2360 error = gfs2_meta_inode_buffer(ip, &dibh);
2361 if (error == 0) {
2362 struct gfs2_dinode *di =
2363 (struct gfs2_dinode *)dibh->b_data;
2364 gfs2_trans_add_meta(ip->i_gl, dibh);
2365 di->di_goal_meta = di->di_goal_data =
2366 cpu_to_be64(ip->i_goal);
2367 brelse(dibh);
2368 }
2369 }
2370 if (rbm.rgd->rd_free < *nblocks) {
2371 pr_warn("nblocks=%u\n", *nblocks);
2372 goto rgrp_error;
2373 }
2374
2375 rbm.rgd->rd_free -= *nblocks;
2376 if (dinode) {
2377 rbm.rgd->rd_dinodes++;
2378 *generation = rbm.rgd->rd_igeneration++;
2379 if (*generation == 0)
2380 *generation = rbm.rgd->rd_igeneration++;
2381 }
2382
2383 gfs2_trans_add_meta(rbm.rgd->rd_gl, rbm.rgd->rd_bits[0].bi_bh);
2384 gfs2_rgrp_out(rbm.rgd, rbm.rgd->rd_bits[0].bi_bh->b_data);
2385 gfs2_rgrp_ondisk2lvb(rbm.rgd->rd_rgl, rbm.rgd->rd_bits[0].bi_bh->b_data);
2386
2387 gfs2_statfs_change(sdp, 0, -(s64)*nblocks, dinode ? 1 : 0);
2388 if (dinode)
2389 gfs2_trans_add_unrevoke(sdp, block, *nblocks);
2390
2391 gfs2_quota_change(ip, *nblocks, ip->i_inode.i_uid, ip->i_inode.i_gid);
2392
2393 rbm.rgd->rd_free_clone -= *nblocks;
2394 trace_gfs2_block_alloc(ip, rbm.rgd, block, *nblocks,
2395 dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED);
2396 *bn = block;
2397 return 0;
2398
2399rgrp_error:
2400 gfs2_rgrp_error(rbm.rgd);
2401 return -EIO;
2402}
2403
2404/**
2405 * __gfs2_free_blocks - free a contiguous run of block(s)
2406 * @ip: the inode these blocks are being freed from
2407 * @bstart: first block of a run of contiguous blocks
2408 * @blen: the length of the block run
2409 * @meta: 1 if the blocks represent metadata
2410 *
2411 */
2412
2413void __gfs2_free_blocks(struct gfs2_inode *ip, u64 bstart, u32 blen, int meta)
2414{
2415 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2416 struct gfs2_rgrpd *rgd;
2417
2418 rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE);
2419 if (!rgd)
2420 return;
2421 trace_gfs2_block_alloc(ip, rgd, bstart, blen, GFS2_BLKST_FREE);
2422 rgd->rd_free += blen;
2423 rgd->rd_flags &= ~GFS2_RGF_TRIMMED;
2424 gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh);
2425 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
2426 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data);
2427
2428 /* Directories keep their data in the metadata address space */
2429 if (meta || ip->i_depth)
2430 gfs2_meta_wipe(ip, bstart, blen);
2431}
2432
2433/**
2434 * gfs2_free_meta - free a contiguous run of data block(s)
2435 * @ip: the inode these blocks are being freed from
2436 * @bstart: first block of a run of contiguous blocks
2437 * @blen: the length of the block run
2438 *
2439 */
2440
2441void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen)
2442{
2443 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2444
2445 __gfs2_free_blocks(ip, bstart, blen, 1);
2446 gfs2_statfs_change(sdp, 0, +blen, 0);
2447 gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid);
2448}
2449
2450void gfs2_unlink_di(struct inode *inode)
2451{
2452 struct gfs2_inode *ip = GFS2_I(inode);
2453 struct gfs2_sbd *sdp = GFS2_SB(inode);
2454 struct gfs2_rgrpd *rgd;
2455 u64 blkno = ip->i_no_addr;
2456
2457 rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED);
2458 if (!rgd)
2459 return;
2460 trace_gfs2_block_alloc(ip, rgd, blkno, 1, GFS2_BLKST_UNLINKED);
2461 gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh);
2462 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
2463 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data);
2464 update_rgrp_lvb_unlinked(rgd, 1);
2465}
2466
2467static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno)
2468{
2469 struct gfs2_sbd *sdp = rgd->rd_sbd;
2470 struct gfs2_rgrpd *tmp_rgd;
2471
2472 tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE);
2473 if (!tmp_rgd)
2474 return;
2475 gfs2_assert_withdraw(sdp, rgd == tmp_rgd);
2476
2477 if (!rgd->rd_dinodes)
2478 gfs2_consist_rgrpd(rgd);
2479 rgd->rd_dinodes--;
2480 rgd->rd_free++;
2481
2482 gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh);
2483 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
2484 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data);
2485 update_rgrp_lvb_unlinked(rgd, -1);
2486
2487 gfs2_statfs_change(sdp, 0, +1, -1);
2488}
2489
2490
2491void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip)
2492{
2493 gfs2_free_uninit_di(rgd, ip->i_no_addr);
2494 trace_gfs2_block_alloc(ip, rgd, ip->i_no_addr, 1, GFS2_BLKST_FREE);
2495 gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid);
2496 gfs2_meta_wipe(ip, ip->i_no_addr, 1);
2497}
2498
2499/**
2500 * gfs2_check_blk_type - Check the type of a block
2501 * @sdp: The superblock
2502 * @no_addr: The block number to check
2503 * @type: The block type we are looking for
2504 *
2505 * Returns: 0 if the block type matches the expected type
2506 * -ESTALE if it doesn't match
2507 * or -ve errno if something went wrong while checking
2508 */
2509
2510int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type)
2511{
2512 struct gfs2_rgrpd *rgd;
2513 struct gfs2_holder rgd_gh;
2514 int error = -EINVAL;
2515
2516 rgd = gfs2_blk2rgrpd(sdp, no_addr, 1);
2517 if (!rgd)
2518 goto fail;
2519
2520 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh);
2521 if (error)
2522 goto fail;
2523
2524 if (gfs2_get_block_type(rgd, no_addr) != type)
2525 error = -ESTALE;
2526
2527 gfs2_glock_dq_uninit(&rgd_gh);
2528fail:
2529 return error;
2530}
2531
2532/**
2533 * gfs2_rlist_add - add a RG to a list of RGs
2534 * @ip: the inode
2535 * @rlist: the list of resource groups
2536 * @block: the block
2537 *
2538 * Figure out what RG a block belongs to and add that RG to the list
2539 *
2540 * FIXME: Don't use NOFAIL
2541 *
2542 */
2543
2544void gfs2_rlist_add(struct gfs2_inode *ip, struct gfs2_rgrp_list *rlist,
2545 u64 block)
2546{
2547 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2548 struct gfs2_rgrpd *rgd;
2549 struct gfs2_rgrpd **tmp;
2550 unsigned int new_space;
2551 unsigned int x;
2552
2553 if (gfs2_assert_warn(sdp, !rlist->rl_ghs))
2554 return;
2555
2556 if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, block))
2557 rgd = ip->i_rgd;
2558 else
2559 rgd = gfs2_blk2rgrpd(sdp, block, 1);
2560 if (!rgd) {
2561 fs_err(sdp, "rlist_add: no rgrp for block %llu\n", (unsigned long long)block);
2562 return;
2563 }
2564 ip->i_rgd = rgd;
2565
2566 for (x = 0; x < rlist->rl_rgrps; x++)
2567 if (rlist->rl_rgd[x] == rgd)
2568 return;
2569
2570 if (rlist->rl_rgrps == rlist->rl_space) {
2571 new_space = rlist->rl_space + 10;
2572
2573 tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *),
2574 GFP_NOFS | __GFP_NOFAIL);
2575
2576 if (rlist->rl_rgd) {
2577 memcpy(tmp, rlist->rl_rgd,
2578 rlist->rl_space * sizeof(struct gfs2_rgrpd *));
2579 kfree(rlist->rl_rgd);
2580 }
2581
2582 rlist->rl_space = new_space;
2583 rlist->rl_rgd = tmp;
2584 }
2585
2586 rlist->rl_rgd[rlist->rl_rgrps++] = rgd;
2587}
2588
2589/**
2590 * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate
2591 * and initialize an array of glock holders for them
2592 * @rlist: the list of resource groups
2593 * @state: the lock state to acquire the RG lock in
2594 *
2595 * FIXME: Don't use NOFAIL
2596 *
2597 */
2598
2599void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state)
2600{
2601 unsigned int x;
2602
2603 rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder),
2604 GFP_NOFS | __GFP_NOFAIL);
2605 for (x = 0; x < rlist->rl_rgrps; x++)
2606 gfs2_holder_init(rlist->rl_rgd[x]->rd_gl,
2607 state, 0,
2608 &rlist->rl_ghs[x]);
2609}
2610
2611/**
2612 * gfs2_rlist_free - free a resource group list
2613 * @rlist: the list of resource groups
2614 *
2615 */
2616
2617void gfs2_rlist_free(struct gfs2_rgrp_list *rlist)
2618{
2619 unsigned int x;
2620
2621 kfree(rlist->rl_rgd);
2622
2623 if (rlist->rl_ghs) {
2624 for (x = 0; x < rlist->rl_rgrps; x++)
2625 gfs2_holder_uninit(&rlist->rl_ghs[x]);
2626 kfree(rlist->rl_ghs);
2627 rlist->rl_ghs = NULL;
2628 }
2629}
2630
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 */
6
7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9#include <linux/slab.h>
10#include <linux/spinlock.h>
11#include <linux/completion.h>
12#include <linux/buffer_head.h>
13#include <linux/fs.h>
14#include <linux/gfs2_ondisk.h>
15#include <linux/prefetch.h>
16#include <linux/blkdev.h>
17#include <linux/rbtree.h>
18#include <linux/random.h>
19
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"
33#include "trace_gfs2.h"
34#include "dir.h"
35
36#define BFITNOENT ((u32)~0)
37#define NO_BLOCK ((u64)~0)
38
39struct gfs2_rbm {
40 struct gfs2_rgrpd *rgd;
41 u32 offset; /* The offset is bitmap relative */
42 int bii; /* Bitmap index */
43};
44
45static inline struct gfs2_bitmap *rbm_bi(const struct gfs2_rbm *rbm)
46{
47 return rbm->rgd->rd_bits + rbm->bii;
48}
49
50static inline u64 gfs2_rbm_to_block(const struct gfs2_rbm *rbm)
51{
52 BUG_ON(rbm->offset >= rbm->rgd->rd_data);
53 return rbm->rgd->rd_data0 + (rbm_bi(rbm)->bi_start * GFS2_NBBY) +
54 rbm->offset;
55}
56
57/*
58 * These routines are used by the resource group routines (rgrp.c)
59 * to keep track of block allocation. Each block is represented by two
60 * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks.
61 *
62 * 0 = Free
63 * 1 = Used (not metadata)
64 * 2 = Unlinked (still in use) inode
65 * 3 = Used (metadata)
66 */
67
68struct gfs2_extent {
69 struct gfs2_rbm rbm;
70 u32 len;
71};
72
73static const char valid_change[16] = {
74 /* current */
75 /* n */ 0, 1, 1, 1,
76 /* e */ 1, 0, 0, 0,
77 /* w */ 0, 0, 0, 1,
78 1, 0, 0, 0
79};
80
81static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 *minext,
82 struct gfs2_blkreserv *rs, bool nowrap);
83
84
85/**
86 * gfs2_setbit - Set a bit in the bitmaps
87 * @rbm: The position of the bit to set
88 * @do_clone: Also set the clone bitmap, if it exists
89 * @new_state: the new state of the block
90 *
91 */
92
93static inline void gfs2_setbit(const struct gfs2_rbm *rbm, bool do_clone,
94 unsigned char new_state)
95{
96 unsigned char *byte1, *byte2, *end, cur_state;
97 struct gfs2_bitmap *bi = rbm_bi(rbm);
98 unsigned int buflen = bi->bi_bytes;
99 const unsigned int bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE;
100
101 byte1 = bi->bi_bh->b_data + bi->bi_offset + (rbm->offset / GFS2_NBBY);
102 end = bi->bi_bh->b_data + bi->bi_offset + buflen;
103
104 BUG_ON(byte1 >= end);
105
106 cur_state = (*byte1 >> bit) & GFS2_BIT_MASK;
107
108 if (unlikely(!valid_change[new_state * 4 + cur_state])) {
109 struct gfs2_sbd *sdp = rbm->rgd->rd_sbd;
110
111 fs_warn(sdp, "buf_blk = 0x%x old_state=%d, new_state=%d\n",
112 rbm->offset, cur_state, new_state);
113 fs_warn(sdp, "rgrp=0x%llx bi_start=0x%x biblk: 0x%llx\n",
114 (unsigned long long)rbm->rgd->rd_addr, bi->bi_start,
115 (unsigned long long)bi->bi_bh->b_blocknr);
116 fs_warn(sdp, "bi_offset=0x%x bi_bytes=0x%x block=0x%llx\n",
117 bi->bi_offset, bi->bi_bytes,
118 (unsigned long long)gfs2_rbm_to_block(rbm));
119 dump_stack();
120 gfs2_consist_rgrpd(rbm->rgd);
121 return;
122 }
123 *byte1 ^= (cur_state ^ new_state) << bit;
124
125 if (do_clone && bi->bi_clone) {
126 byte2 = bi->bi_clone + bi->bi_offset + (rbm->offset / GFS2_NBBY);
127 cur_state = (*byte2 >> bit) & GFS2_BIT_MASK;
128 *byte2 ^= (cur_state ^ new_state) << bit;
129 }
130}
131
132/**
133 * gfs2_testbit - test a bit in the bitmaps
134 * @rbm: The bit to test
135 * @use_clone: If true, test the clone bitmap, not the official bitmap.
136 *
137 * Some callers like gfs2_unaligned_extlen need to test the clone bitmaps,
138 * not the "real" bitmaps, to avoid allocating recently freed blocks.
139 *
140 * Returns: The two bit block state of the requested bit
141 */
142
143static inline u8 gfs2_testbit(const struct gfs2_rbm *rbm, bool use_clone)
144{
145 struct gfs2_bitmap *bi = rbm_bi(rbm);
146 const u8 *buffer;
147 const u8 *byte;
148 unsigned int bit;
149
150 if (use_clone && bi->bi_clone)
151 buffer = bi->bi_clone;
152 else
153 buffer = bi->bi_bh->b_data;
154 buffer += bi->bi_offset;
155 byte = buffer + (rbm->offset / GFS2_NBBY);
156 bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE;
157
158 return (*byte >> bit) & GFS2_BIT_MASK;
159}
160
161/**
162 * gfs2_bit_search - search bitmap for a state
163 * @ptr: Pointer to bitmap data
164 * @mask: Mask to use (normally 0x55555.... but adjusted for search start)
165 * @state: The state we are searching for
166 *
167 * We xor the bitmap data with a pattern which is the bitwise opposite
168 * of what we are looking for. This gives rise to a pattern of ones
169 * wherever there is a match. Since we have two bits per entry, we
170 * take this pattern, shift it down by one place and then and it with
171 * the original. All the even bit positions (0,2,4, etc) then represent
172 * successful matches, so we mask with 0x55555..... to remove the unwanted
173 * odd bit positions.
174 *
175 * This allows searching of a whole u64 at once (32 blocks) with a
176 * single test (on 64 bit arches).
177 */
178
179static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state)
180{
181 u64 tmp;
182 static const u64 search[] = {
183 [0] = 0xffffffffffffffffULL,
184 [1] = 0xaaaaaaaaaaaaaaaaULL,
185 [2] = 0x5555555555555555ULL,
186 [3] = 0x0000000000000000ULL,
187 };
188 tmp = le64_to_cpu(*ptr) ^ search[state];
189 tmp &= (tmp >> 1);
190 tmp &= mask;
191 return tmp;
192}
193
194/**
195 * rs_cmp - multi-block reservation range compare
196 * @start: start of the new reservation
197 * @len: number of blocks in the new reservation
198 * @rs: existing reservation to compare against
199 *
200 * returns: 1 if the block range is beyond the reach of the reservation
201 * -1 if the block range is before the start of the reservation
202 * 0 if the block range overlaps with the reservation
203 */
204static inline int rs_cmp(u64 start, u32 len, struct gfs2_blkreserv *rs)
205{
206 if (start >= rs->rs_start + rs->rs_requested)
207 return 1;
208 if (rs->rs_start >= start + len)
209 return -1;
210 return 0;
211}
212
213/**
214 * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing
215 * a block in a given allocation state.
216 * @buf: the buffer that holds the bitmaps
217 * @len: the length (in bytes) of the buffer
218 * @goal: start search at this block's bit-pair (within @buffer)
219 * @state: GFS2_BLKST_XXX the state of the block we're looking for.
220 *
221 * Scope of @goal and returned block number is only within this bitmap buffer,
222 * not entire rgrp or filesystem. @buffer will be offset from the actual
223 * beginning of a bitmap block buffer, skipping any header structures, but
224 * headers are always a multiple of 64 bits long so that the buffer is
225 * always aligned to a 64 bit boundary.
226 *
227 * The size of the buffer is in bytes, but is it assumed that it is
228 * always ok to read a complete multiple of 64 bits at the end
229 * of the block in case the end is no aligned to a natural boundary.
230 *
231 * Return: the block number (bitmap buffer scope) that was found
232 */
233
234static u32 gfs2_bitfit(const u8 *buf, const unsigned int len,
235 u32 goal, u8 state)
236{
237 u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1);
238 const __le64 *ptr = ((__le64 *)buf) + (goal >> 5);
239 const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64)));
240 u64 tmp;
241 u64 mask = 0x5555555555555555ULL;
242 u32 bit;
243
244 /* Mask off bits we don't care about at the start of the search */
245 mask <<= spoint;
246 tmp = gfs2_bit_search(ptr, mask, state);
247 ptr++;
248 while(tmp == 0 && ptr < end) {
249 tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state);
250 ptr++;
251 }
252 /* Mask off any bits which are more than len bytes from the start */
253 if (ptr == end && (len & (sizeof(u64) - 1)))
254 tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1))));
255 /* Didn't find anything, so return */
256 if (tmp == 0)
257 return BFITNOENT;
258 ptr--;
259 bit = __ffs64(tmp);
260 bit /= 2; /* two bits per entry in the bitmap */
261 return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit;
262}
263
264/**
265 * gfs2_rbm_from_block - Set the rbm based upon rgd and block number
266 * @rbm: The rbm with rgd already set correctly
267 * @block: The block number (filesystem relative)
268 *
269 * This sets the bi and offset members of an rbm based on a
270 * resource group and a filesystem relative block number. The
271 * resource group must be set in the rbm on entry, the bi and
272 * offset members will be set by this function.
273 *
274 * Returns: 0 on success, or an error code
275 */
276
277static int gfs2_rbm_from_block(struct gfs2_rbm *rbm, u64 block)
278{
279 if (!rgrp_contains_block(rbm->rgd, block))
280 return -E2BIG;
281 rbm->bii = 0;
282 rbm->offset = block - rbm->rgd->rd_data0;
283 /* Check if the block is within the first block */
284 if (rbm->offset < rbm_bi(rbm)->bi_blocks)
285 return 0;
286
287 /* Adjust for the size diff between gfs2_meta_header and gfs2_rgrp */
288 rbm->offset += (sizeof(struct gfs2_rgrp) -
289 sizeof(struct gfs2_meta_header)) * GFS2_NBBY;
290 rbm->bii = rbm->offset / rbm->rgd->rd_sbd->sd_blocks_per_bitmap;
291 rbm->offset -= rbm->bii * rbm->rgd->rd_sbd->sd_blocks_per_bitmap;
292 return 0;
293}
294
295/**
296 * gfs2_rbm_add - add a number of blocks to an rbm
297 * @rbm: The rbm with rgd already set correctly
298 * @blocks: The number of blocks to add to rpm
299 *
300 * This function takes an existing rbm structure and adds a number of blocks to
301 * it.
302 *
303 * Returns: True if the new rbm would point past the end of the rgrp.
304 */
305
306static bool gfs2_rbm_add(struct gfs2_rbm *rbm, u32 blocks)
307{
308 struct gfs2_rgrpd *rgd = rbm->rgd;
309 struct gfs2_bitmap *bi = rgd->rd_bits + rbm->bii;
310
311 if (rbm->offset + blocks < bi->bi_blocks) {
312 rbm->offset += blocks;
313 return false;
314 }
315 blocks -= bi->bi_blocks - rbm->offset;
316
317 for(;;) {
318 bi++;
319 if (bi == rgd->rd_bits + rgd->rd_length)
320 return true;
321 if (blocks < bi->bi_blocks) {
322 rbm->offset = blocks;
323 rbm->bii = bi - rgd->rd_bits;
324 return false;
325 }
326 blocks -= bi->bi_blocks;
327 }
328}
329
330/**
331 * gfs2_unaligned_extlen - Look for free blocks which are not byte aligned
332 * @rbm: Position to search (value/result)
333 * @n_unaligned: Number of unaligned blocks to check
334 * @len: Decremented for each block found (terminate on zero)
335 *
336 * Returns: true if a non-free block is encountered or the end of the resource
337 * group is reached.
338 */
339
340static bool gfs2_unaligned_extlen(struct gfs2_rbm *rbm, u32 n_unaligned, u32 *len)
341{
342 u32 n;
343 u8 res;
344
345 for (n = 0; n < n_unaligned; n++) {
346 res = gfs2_testbit(rbm, true);
347 if (res != GFS2_BLKST_FREE)
348 return true;
349 (*len)--;
350 if (*len == 0)
351 return true;
352 if (gfs2_rbm_add(rbm, 1))
353 return true;
354 }
355
356 return false;
357}
358
359/**
360 * gfs2_free_extlen - Return extent length of free blocks
361 * @rrbm: Starting position
362 * @len: Max length to check
363 *
364 * Starting at the block specified by the rbm, see how many free blocks
365 * there are, not reading more than len blocks ahead. This can be done
366 * using memchr_inv when the blocks are byte aligned, but has to be done
367 * on a block by block basis in case of unaligned blocks. Also this
368 * function can cope with bitmap boundaries (although it must stop on
369 * a resource group boundary)
370 *
371 * Returns: Number of free blocks in the extent
372 */
373
374static u32 gfs2_free_extlen(const struct gfs2_rbm *rrbm, u32 len)
375{
376 struct gfs2_rbm rbm = *rrbm;
377 u32 n_unaligned = rbm.offset & 3;
378 u32 size = len;
379 u32 bytes;
380 u32 chunk_size;
381 u8 *ptr, *start, *end;
382 u64 block;
383 struct gfs2_bitmap *bi;
384
385 if (n_unaligned &&
386 gfs2_unaligned_extlen(&rbm, 4 - n_unaligned, &len))
387 goto out;
388
389 n_unaligned = len & 3;
390 /* Start is now byte aligned */
391 while (len > 3) {
392 bi = rbm_bi(&rbm);
393 start = bi->bi_bh->b_data;
394 if (bi->bi_clone)
395 start = bi->bi_clone;
396 start += bi->bi_offset;
397 end = start + bi->bi_bytes;
398 BUG_ON(rbm.offset & 3);
399 start += (rbm.offset / GFS2_NBBY);
400 bytes = min_t(u32, len / GFS2_NBBY, (end - start));
401 ptr = memchr_inv(start, 0, bytes);
402 chunk_size = ((ptr == NULL) ? bytes : (ptr - start));
403 chunk_size *= GFS2_NBBY;
404 BUG_ON(len < chunk_size);
405 len -= chunk_size;
406 block = gfs2_rbm_to_block(&rbm);
407 if (gfs2_rbm_from_block(&rbm, block + chunk_size)) {
408 n_unaligned = 0;
409 break;
410 }
411 if (ptr) {
412 n_unaligned = 3;
413 break;
414 }
415 n_unaligned = len & 3;
416 }
417
418 /* Deal with any bits left over at the end */
419 if (n_unaligned)
420 gfs2_unaligned_extlen(&rbm, n_unaligned, &len);
421out:
422 return size - len;
423}
424
425/**
426 * gfs2_bitcount - count the number of bits in a certain state
427 * @rgd: the resource group descriptor
428 * @buffer: the buffer that holds the bitmaps
429 * @buflen: the length (in bytes) of the buffer
430 * @state: the state of the block we're looking for
431 *
432 * Returns: The number of bits
433 */
434
435static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer,
436 unsigned int buflen, u8 state)
437{
438 const u8 *byte = buffer;
439 const u8 *end = buffer + buflen;
440 const u8 state1 = state << 2;
441 const u8 state2 = state << 4;
442 const u8 state3 = state << 6;
443 u32 count = 0;
444
445 for (; byte < end; byte++) {
446 if (((*byte) & 0x03) == state)
447 count++;
448 if (((*byte) & 0x0C) == state1)
449 count++;
450 if (((*byte) & 0x30) == state2)
451 count++;
452 if (((*byte) & 0xC0) == state3)
453 count++;
454 }
455
456 return count;
457}
458
459/**
460 * gfs2_rgrp_verify - Verify that a resource group is consistent
461 * @rgd: the rgrp
462 *
463 */
464
465void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd)
466{
467 struct gfs2_sbd *sdp = rgd->rd_sbd;
468 struct gfs2_bitmap *bi = NULL;
469 u32 length = rgd->rd_length;
470 u32 count[4], tmp;
471 int buf, x;
472
473 memset(count, 0, 4 * sizeof(u32));
474
475 /* Count # blocks in each of 4 possible allocation states */
476 for (buf = 0; buf < length; buf++) {
477 bi = rgd->rd_bits + buf;
478 for (x = 0; x < 4; x++)
479 count[x] += gfs2_bitcount(rgd,
480 bi->bi_bh->b_data +
481 bi->bi_offset,
482 bi->bi_bytes, x);
483 }
484
485 if (count[0] != rgd->rd_free) {
486 gfs2_lm(sdp, "free data mismatch: %u != %u\n",
487 count[0], rgd->rd_free);
488 gfs2_consist_rgrpd(rgd);
489 return;
490 }
491
492 tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes;
493 if (count[1] != tmp) {
494 gfs2_lm(sdp, "used data mismatch: %u != %u\n",
495 count[1], tmp);
496 gfs2_consist_rgrpd(rgd);
497 return;
498 }
499
500 if (count[2] + count[3] != rgd->rd_dinodes) {
501 gfs2_lm(sdp, "used metadata mismatch: %u != %u\n",
502 count[2] + count[3], rgd->rd_dinodes);
503 gfs2_consist_rgrpd(rgd);
504 return;
505 }
506}
507
508/**
509 * gfs2_blk2rgrpd - Find resource group for a given data/meta block number
510 * @sdp: The GFS2 superblock
511 * @blk: The data block number
512 * @exact: True if this needs to be an exact match
513 *
514 * The @exact argument should be set to true by most callers. The exception
515 * is when we need to match blocks which are not represented by the rgrp
516 * bitmap, but which are part of the rgrp (i.e. padding blocks) which are
517 * there for alignment purposes. Another way of looking at it is that @exact
518 * matches only valid data/metadata blocks, but with @exact false, it will
519 * match any block within the extent of the rgrp.
520 *
521 * Returns: The resource group, or NULL if not found
522 */
523
524struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk, bool exact)
525{
526 struct rb_node *n, *next;
527 struct gfs2_rgrpd *cur;
528
529 spin_lock(&sdp->sd_rindex_spin);
530 n = sdp->sd_rindex_tree.rb_node;
531 while (n) {
532 cur = rb_entry(n, struct gfs2_rgrpd, rd_node);
533 next = NULL;
534 if (blk < cur->rd_addr)
535 next = n->rb_left;
536 else if (blk >= cur->rd_data0 + cur->rd_data)
537 next = n->rb_right;
538 if (next == NULL) {
539 spin_unlock(&sdp->sd_rindex_spin);
540 if (exact) {
541 if (blk < cur->rd_addr)
542 return NULL;
543 if (blk >= cur->rd_data0 + cur->rd_data)
544 return NULL;
545 }
546 return cur;
547 }
548 n = next;
549 }
550 spin_unlock(&sdp->sd_rindex_spin);
551
552 return NULL;
553}
554
555/**
556 * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem
557 * @sdp: The GFS2 superblock
558 *
559 * Returns: The first rgrp in the filesystem
560 */
561
562struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp)
563{
564 const struct rb_node *n;
565 struct gfs2_rgrpd *rgd;
566
567 spin_lock(&sdp->sd_rindex_spin);
568 n = rb_first(&sdp->sd_rindex_tree);
569 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node);
570 spin_unlock(&sdp->sd_rindex_spin);
571
572 return rgd;
573}
574
575/**
576 * gfs2_rgrpd_get_next - get the next RG
577 * @rgd: the resource group descriptor
578 *
579 * Returns: The next rgrp
580 */
581
582struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd)
583{
584 struct gfs2_sbd *sdp = rgd->rd_sbd;
585 const struct rb_node *n;
586
587 spin_lock(&sdp->sd_rindex_spin);
588 n = rb_next(&rgd->rd_node);
589 if (n == NULL)
590 n = rb_first(&sdp->sd_rindex_tree);
591
592 if (unlikely(&rgd->rd_node == n)) {
593 spin_unlock(&sdp->sd_rindex_spin);
594 return NULL;
595 }
596 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node);
597 spin_unlock(&sdp->sd_rindex_spin);
598 return rgd;
599}
600
601void check_and_update_goal(struct gfs2_inode *ip)
602{
603 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
604 if (!ip->i_goal || gfs2_blk2rgrpd(sdp, ip->i_goal, 1) == NULL)
605 ip->i_goal = ip->i_no_addr;
606}
607
608void gfs2_free_clones(struct gfs2_rgrpd *rgd)
609{
610 int x;
611
612 for (x = 0; x < rgd->rd_length; x++) {
613 struct gfs2_bitmap *bi = rgd->rd_bits + x;
614 kfree(bi->bi_clone);
615 bi->bi_clone = NULL;
616 }
617}
618
619static void dump_rs(struct seq_file *seq, const struct gfs2_blkreserv *rs,
620 const char *fs_id_buf)
621{
622 struct gfs2_inode *ip = container_of(rs, struct gfs2_inode, i_res);
623
624 gfs2_print_dbg(seq, "%s B: n:%llu s:%llu f:%u\n",
625 fs_id_buf,
626 (unsigned long long)ip->i_no_addr,
627 (unsigned long long)rs->rs_start,
628 rs->rs_requested);
629}
630
631/**
632 * __rs_deltree - remove a multi-block reservation from the rgd tree
633 * @rs: The reservation to remove
634 *
635 */
636static void __rs_deltree(struct gfs2_blkreserv *rs)
637{
638 struct gfs2_rgrpd *rgd;
639
640 if (!gfs2_rs_active(rs))
641 return;
642
643 rgd = rs->rs_rgd;
644 trace_gfs2_rs(rs, TRACE_RS_TREEDEL);
645 rb_erase(&rs->rs_node, &rgd->rd_rstree);
646 RB_CLEAR_NODE(&rs->rs_node);
647
648 if (rs->rs_requested) {
649 /* return requested blocks to the rgrp */
650 BUG_ON(rs->rs_rgd->rd_requested < rs->rs_requested);
651 rs->rs_rgd->rd_requested -= rs->rs_requested;
652
653 /* The rgrp extent failure point is likely not to increase;
654 it will only do so if the freed blocks are somehow
655 contiguous with a span of free blocks that follows. Still,
656 it will force the number to be recalculated later. */
657 rgd->rd_extfail_pt += rs->rs_requested;
658 rs->rs_requested = 0;
659 }
660}
661
662/**
663 * gfs2_rs_deltree - remove a multi-block reservation from the rgd tree
664 * @rs: The reservation to remove
665 *
666 */
667void gfs2_rs_deltree(struct gfs2_blkreserv *rs)
668{
669 struct gfs2_rgrpd *rgd;
670
671 rgd = rs->rs_rgd;
672 if (rgd) {
673 spin_lock(&rgd->rd_rsspin);
674 __rs_deltree(rs);
675 BUG_ON(rs->rs_requested);
676 spin_unlock(&rgd->rd_rsspin);
677 }
678}
679
680/**
681 * gfs2_rs_delete - delete a multi-block reservation
682 * @ip: The inode for this reservation
683 *
684 */
685void gfs2_rs_delete(struct gfs2_inode *ip)
686{
687 struct inode *inode = &ip->i_inode;
688
689 down_write(&ip->i_rw_mutex);
690 if (atomic_read(&inode->i_writecount) <= 1)
691 gfs2_rs_deltree(&ip->i_res);
692 up_write(&ip->i_rw_mutex);
693}
694
695/**
696 * return_all_reservations - return all reserved blocks back to the rgrp.
697 * @rgd: the rgrp that needs its space back
698 *
699 * We previously reserved a bunch of blocks for allocation. Now we need to
700 * give them back. This leave the reservation structures in tact, but removes
701 * all of their corresponding "no-fly zones".
702 */
703static void return_all_reservations(struct gfs2_rgrpd *rgd)
704{
705 struct rb_node *n;
706 struct gfs2_blkreserv *rs;
707
708 spin_lock(&rgd->rd_rsspin);
709 while ((n = rb_first(&rgd->rd_rstree))) {
710 rs = rb_entry(n, struct gfs2_blkreserv, rs_node);
711 __rs_deltree(rs);
712 }
713 spin_unlock(&rgd->rd_rsspin);
714}
715
716void gfs2_clear_rgrpd(struct gfs2_sbd *sdp)
717{
718 struct rb_node *n;
719 struct gfs2_rgrpd *rgd;
720 struct gfs2_glock *gl;
721
722 while ((n = rb_first(&sdp->sd_rindex_tree))) {
723 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node);
724 gl = rgd->rd_gl;
725
726 rb_erase(n, &sdp->sd_rindex_tree);
727
728 if (gl) {
729 if (gl->gl_state != LM_ST_UNLOCKED) {
730 gfs2_glock_cb(gl, LM_ST_UNLOCKED);
731 flush_delayed_work(&gl->gl_work);
732 }
733 gfs2_rgrp_brelse(rgd);
734 glock_clear_object(gl, rgd);
735 gfs2_glock_put(gl);
736 }
737
738 gfs2_free_clones(rgd);
739 return_all_reservations(rgd);
740 kfree(rgd->rd_bits);
741 rgd->rd_bits = NULL;
742 kmem_cache_free(gfs2_rgrpd_cachep, rgd);
743 }
744}
745
746/**
747 * compute_bitstructs - Compute the bitmap sizes
748 * @rgd: The resource group descriptor
749 *
750 * Calculates bitmap descriptors, one for each block that contains bitmap data
751 *
752 * Returns: errno
753 */
754
755static int compute_bitstructs(struct gfs2_rgrpd *rgd)
756{
757 struct gfs2_sbd *sdp = rgd->rd_sbd;
758 struct gfs2_bitmap *bi;
759 u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */
760 u32 bytes_left, bytes;
761 int x;
762
763 if (!length)
764 return -EINVAL;
765
766 rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS);
767 if (!rgd->rd_bits)
768 return -ENOMEM;
769
770 bytes_left = rgd->rd_bitbytes;
771
772 for (x = 0; x < length; x++) {
773 bi = rgd->rd_bits + x;
774
775 bi->bi_flags = 0;
776 /* small rgrp; bitmap stored completely in header block */
777 if (length == 1) {
778 bytes = bytes_left;
779 bi->bi_offset = sizeof(struct gfs2_rgrp);
780 bi->bi_start = 0;
781 bi->bi_bytes = bytes;
782 bi->bi_blocks = bytes * GFS2_NBBY;
783 /* header block */
784 } else if (x == 0) {
785 bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp);
786 bi->bi_offset = sizeof(struct gfs2_rgrp);
787 bi->bi_start = 0;
788 bi->bi_bytes = bytes;
789 bi->bi_blocks = bytes * GFS2_NBBY;
790 /* last block */
791 } else if (x + 1 == length) {
792 bytes = bytes_left;
793 bi->bi_offset = sizeof(struct gfs2_meta_header);
794 bi->bi_start = rgd->rd_bitbytes - bytes_left;
795 bi->bi_bytes = bytes;
796 bi->bi_blocks = bytes * GFS2_NBBY;
797 /* other blocks */
798 } else {
799 bytes = sdp->sd_sb.sb_bsize -
800 sizeof(struct gfs2_meta_header);
801 bi->bi_offset = sizeof(struct gfs2_meta_header);
802 bi->bi_start = rgd->rd_bitbytes - bytes_left;
803 bi->bi_bytes = bytes;
804 bi->bi_blocks = bytes * GFS2_NBBY;
805 }
806
807 bytes_left -= bytes;
808 }
809
810 if (bytes_left) {
811 gfs2_consist_rgrpd(rgd);
812 return -EIO;
813 }
814 bi = rgd->rd_bits + (length - 1);
815 if ((bi->bi_start + bi->bi_bytes) * GFS2_NBBY != rgd->rd_data) {
816 gfs2_lm(sdp,
817 "ri_addr=%llu "
818 "ri_length=%u "
819 "ri_data0=%llu "
820 "ri_data=%u "
821 "ri_bitbytes=%u "
822 "start=%u len=%u offset=%u\n",
823 (unsigned long long)rgd->rd_addr,
824 rgd->rd_length,
825 (unsigned long long)rgd->rd_data0,
826 rgd->rd_data,
827 rgd->rd_bitbytes,
828 bi->bi_start, bi->bi_bytes, bi->bi_offset);
829 gfs2_consist_rgrpd(rgd);
830 return -EIO;
831 }
832
833 return 0;
834}
835
836/**
837 * gfs2_ri_total - Total up the file system space, according to the rindex.
838 * @sdp: the filesystem
839 *
840 */
841u64 gfs2_ri_total(struct gfs2_sbd *sdp)
842{
843 u64 total_data = 0;
844 struct inode *inode = sdp->sd_rindex;
845 struct gfs2_inode *ip = GFS2_I(inode);
846 char buf[sizeof(struct gfs2_rindex)];
847 int error, rgrps;
848
849 for (rgrps = 0;; rgrps++) {
850 loff_t pos = rgrps * sizeof(struct gfs2_rindex);
851
852 if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode))
853 break;
854 error = gfs2_internal_read(ip, buf, &pos,
855 sizeof(struct gfs2_rindex));
856 if (error != sizeof(struct gfs2_rindex))
857 break;
858 total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data);
859 }
860 return total_data;
861}
862
863static int rgd_insert(struct gfs2_rgrpd *rgd)
864{
865 struct gfs2_sbd *sdp = rgd->rd_sbd;
866 struct rb_node **newn = &sdp->sd_rindex_tree.rb_node, *parent = NULL;
867
868 /* Figure out where to put new node */
869 while (*newn) {
870 struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd,
871 rd_node);
872
873 parent = *newn;
874 if (rgd->rd_addr < cur->rd_addr)
875 newn = &((*newn)->rb_left);
876 else if (rgd->rd_addr > cur->rd_addr)
877 newn = &((*newn)->rb_right);
878 else
879 return -EEXIST;
880 }
881
882 rb_link_node(&rgd->rd_node, parent, newn);
883 rb_insert_color(&rgd->rd_node, &sdp->sd_rindex_tree);
884 sdp->sd_rgrps++;
885 return 0;
886}
887
888/**
889 * read_rindex_entry - Pull in a new resource index entry from the disk
890 * @ip: Pointer to the rindex inode
891 *
892 * Returns: 0 on success, > 0 on EOF, error code otherwise
893 */
894
895static int read_rindex_entry(struct gfs2_inode *ip)
896{
897 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
898 loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex);
899 struct gfs2_rindex buf;
900 int error;
901 struct gfs2_rgrpd *rgd;
902
903 if (pos >= i_size_read(&ip->i_inode))
904 return 1;
905
906 error = gfs2_internal_read(ip, (char *)&buf, &pos,
907 sizeof(struct gfs2_rindex));
908
909 if (error != sizeof(struct gfs2_rindex))
910 return (error == 0) ? 1 : error;
911
912 rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS);
913 error = -ENOMEM;
914 if (!rgd)
915 return error;
916
917 rgd->rd_sbd = sdp;
918 rgd->rd_addr = be64_to_cpu(buf.ri_addr);
919 rgd->rd_length = be32_to_cpu(buf.ri_length);
920 rgd->rd_data0 = be64_to_cpu(buf.ri_data0);
921 rgd->rd_data = be32_to_cpu(buf.ri_data);
922 rgd->rd_bitbytes = be32_to_cpu(buf.ri_bitbytes);
923 spin_lock_init(&rgd->rd_rsspin);
924 mutex_init(&rgd->rd_mutex);
925
926 error = gfs2_glock_get(sdp, rgd->rd_addr,
927 &gfs2_rgrp_glops, CREATE, &rgd->rd_gl);
928 if (error)
929 goto fail;
930
931 error = compute_bitstructs(rgd);
932 if (error)
933 goto fail_glock;
934
935 rgd->rd_rgl = (struct gfs2_rgrp_lvb *)rgd->rd_gl->gl_lksb.sb_lvbptr;
936 rgd->rd_flags &= ~GFS2_RDF_PREFERRED;
937 if (rgd->rd_data > sdp->sd_max_rg_data)
938 sdp->sd_max_rg_data = rgd->rd_data;
939 spin_lock(&sdp->sd_rindex_spin);
940 error = rgd_insert(rgd);
941 spin_unlock(&sdp->sd_rindex_spin);
942 if (!error) {
943 glock_set_object(rgd->rd_gl, rgd);
944 return 0;
945 }
946
947 error = 0; /* someone else read in the rgrp; free it and ignore it */
948fail_glock:
949 gfs2_glock_put(rgd->rd_gl);
950
951fail:
952 kfree(rgd->rd_bits);
953 rgd->rd_bits = NULL;
954 kmem_cache_free(gfs2_rgrpd_cachep, rgd);
955 return error;
956}
957
958/**
959 * set_rgrp_preferences - Run all the rgrps, selecting some we prefer to use
960 * @sdp: the GFS2 superblock
961 *
962 * The purpose of this function is to select a subset of the resource groups
963 * and mark them as PREFERRED. We do it in such a way that each node prefers
964 * to use a unique set of rgrps to minimize glock contention.
965 */
966static void set_rgrp_preferences(struct gfs2_sbd *sdp)
967{
968 struct gfs2_rgrpd *rgd, *first;
969 int i;
970
971 /* Skip an initial number of rgrps, based on this node's journal ID.
972 That should start each node out on its own set. */
973 rgd = gfs2_rgrpd_get_first(sdp);
974 for (i = 0; i < sdp->sd_lockstruct.ls_jid; i++)
975 rgd = gfs2_rgrpd_get_next(rgd);
976 first = rgd;
977
978 do {
979 rgd->rd_flags |= GFS2_RDF_PREFERRED;
980 for (i = 0; i < sdp->sd_journals; i++) {
981 rgd = gfs2_rgrpd_get_next(rgd);
982 if (!rgd || rgd == first)
983 break;
984 }
985 } while (rgd && rgd != first);
986}
987
988/**
989 * gfs2_ri_update - Pull in a new resource index from the disk
990 * @ip: pointer to the rindex inode
991 *
992 * Returns: 0 on successful update, error code otherwise
993 */
994
995static int gfs2_ri_update(struct gfs2_inode *ip)
996{
997 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
998 int error;
999
1000 do {
1001 error = read_rindex_entry(ip);
1002 } while (error == 0);
1003
1004 if (error < 0)
1005 return error;
1006
1007 if (RB_EMPTY_ROOT(&sdp->sd_rindex_tree)) {
1008 fs_err(sdp, "no resource groups found in the file system.\n");
1009 return -ENOENT;
1010 }
1011 set_rgrp_preferences(sdp);
1012
1013 sdp->sd_rindex_uptodate = 1;
1014 return 0;
1015}
1016
1017/**
1018 * gfs2_rindex_update - Update the rindex if required
1019 * @sdp: The GFS2 superblock
1020 *
1021 * We grab a lock on the rindex inode to make sure that it doesn't
1022 * change whilst we are performing an operation. We keep this lock
1023 * for quite long periods of time compared to other locks. This
1024 * doesn't matter, since it is shared and it is very, very rarely
1025 * accessed in the exclusive mode (i.e. only when expanding the filesystem).
1026 *
1027 * This makes sure that we're using the latest copy of the resource index
1028 * special file, which might have been updated if someone expanded the
1029 * filesystem (via gfs2_grow utility), which adds new resource groups.
1030 *
1031 * Returns: 0 on succeess, error code otherwise
1032 */
1033
1034int gfs2_rindex_update(struct gfs2_sbd *sdp)
1035{
1036 struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex);
1037 struct gfs2_glock *gl = ip->i_gl;
1038 struct gfs2_holder ri_gh;
1039 int error = 0;
1040 int unlock_required = 0;
1041
1042 /* Read new copy from disk if we don't have the latest */
1043 if (!sdp->sd_rindex_uptodate) {
1044 if (!gfs2_glock_is_locked_by_me(gl)) {
1045 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &ri_gh);
1046 if (error)
1047 return error;
1048 unlock_required = 1;
1049 }
1050 if (!sdp->sd_rindex_uptodate)
1051 error = gfs2_ri_update(ip);
1052 if (unlock_required)
1053 gfs2_glock_dq_uninit(&ri_gh);
1054 }
1055
1056 return error;
1057}
1058
1059static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf)
1060{
1061 const struct gfs2_rgrp *str = buf;
1062 u32 rg_flags;
1063
1064 rg_flags = be32_to_cpu(str->rg_flags);
1065 rg_flags &= ~GFS2_RDF_MASK;
1066 rgd->rd_flags &= GFS2_RDF_MASK;
1067 rgd->rd_flags |= rg_flags;
1068 rgd->rd_free = be32_to_cpu(str->rg_free);
1069 rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes);
1070 rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration);
1071 /* rd_data0, rd_data and rd_bitbytes already set from rindex */
1072}
1073
1074static void gfs2_rgrp_ondisk2lvb(struct gfs2_rgrp_lvb *rgl, const void *buf)
1075{
1076 const struct gfs2_rgrp *str = buf;
1077
1078 rgl->rl_magic = cpu_to_be32(GFS2_MAGIC);
1079 rgl->rl_flags = str->rg_flags;
1080 rgl->rl_free = str->rg_free;
1081 rgl->rl_dinodes = str->rg_dinodes;
1082 rgl->rl_igeneration = str->rg_igeneration;
1083 rgl->__pad = 0UL;
1084}
1085
1086static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf)
1087{
1088 struct gfs2_rgrpd *next = gfs2_rgrpd_get_next(rgd);
1089 struct gfs2_rgrp *str = buf;
1090 u32 crc;
1091
1092 str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK);
1093 str->rg_free = cpu_to_be32(rgd->rd_free);
1094 str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes);
1095 if (next == NULL)
1096 str->rg_skip = 0;
1097 else if (next->rd_addr > rgd->rd_addr)
1098 str->rg_skip = cpu_to_be32(next->rd_addr - rgd->rd_addr);
1099 str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration);
1100 str->rg_data0 = cpu_to_be64(rgd->rd_data0);
1101 str->rg_data = cpu_to_be32(rgd->rd_data);
1102 str->rg_bitbytes = cpu_to_be32(rgd->rd_bitbytes);
1103 str->rg_crc = 0;
1104 crc = gfs2_disk_hash(buf, sizeof(struct gfs2_rgrp));
1105 str->rg_crc = cpu_to_be32(crc);
1106
1107 memset(&str->rg_reserved, 0, sizeof(str->rg_reserved));
1108 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, buf);
1109}
1110
1111static int gfs2_rgrp_lvb_valid(struct gfs2_rgrpd *rgd)
1112{
1113 struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl;
1114 struct gfs2_rgrp *str = (struct gfs2_rgrp *)rgd->rd_bits[0].bi_bh->b_data;
1115 struct gfs2_sbd *sdp = rgd->rd_sbd;
1116 int valid = 1;
1117
1118 if (rgl->rl_flags != str->rg_flags) {
1119 fs_warn(sdp, "GFS2: rgd: %llu lvb flag mismatch %u/%u",
1120 (unsigned long long)rgd->rd_addr,
1121 be32_to_cpu(rgl->rl_flags), be32_to_cpu(str->rg_flags));
1122 valid = 0;
1123 }
1124 if (rgl->rl_free != str->rg_free) {
1125 fs_warn(sdp, "GFS2: rgd: %llu lvb free mismatch %u/%u",
1126 (unsigned long long)rgd->rd_addr,
1127 be32_to_cpu(rgl->rl_free), be32_to_cpu(str->rg_free));
1128 valid = 0;
1129 }
1130 if (rgl->rl_dinodes != str->rg_dinodes) {
1131 fs_warn(sdp, "GFS2: rgd: %llu lvb dinode mismatch %u/%u",
1132 (unsigned long long)rgd->rd_addr,
1133 be32_to_cpu(rgl->rl_dinodes),
1134 be32_to_cpu(str->rg_dinodes));
1135 valid = 0;
1136 }
1137 if (rgl->rl_igeneration != str->rg_igeneration) {
1138 fs_warn(sdp, "GFS2: rgd: %llu lvb igen mismatch %llu/%llu",
1139 (unsigned long long)rgd->rd_addr,
1140 (unsigned long long)be64_to_cpu(rgl->rl_igeneration),
1141 (unsigned long long)be64_to_cpu(str->rg_igeneration));
1142 valid = 0;
1143 }
1144 return valid;
1145}
1146
1147static u32 count_unlinked(struct gfs2_rgrpd *rgd)
1148{
1149 struct gfs2_bitmap *bi;
1150 const u32 length = rgd->rd_length;
1151 const u8 *buffer = NULL;
1152 u32 i, goal, count = 0;
1153
1154 for (i = 0, bi = rgd->rd_bits; i < length; i++, bi++) {
1155 goal = 0;
1156 buffer = bi->bi_bh->b_data + bi->bi_offset;
1157 WARN_ON(!buffer_uptodate(bi->bi_bh));
1158 while (goal < bi->bi_blocks) {
1159 goal = gfs2_bitfit(buffer, bi->bi_bytes, goal,
1160 GFS2_BLKST_UNLINKED);
1161 if (goal == BFITNOENT)
1162 break;
1163 count++;
1164 goal++;
1165 }
1166 }
1167
1168 return count;
1169}
1170
1171static void rgrp_set_bitmap_flags(struct gfs2_rgrpd *rgd)
1172{
1173 struct gfs2_bitmap *bi;
1174 int x;
1175
1176 if (rgd->rd_free) {
1177 for (x = 0; x < rgd->rd_length; x++) {
1178 bi = rgd->rd_bits + x;
1179 clear_bit(GBF_FULL, &bi->bi_flags);
1180 }
1181 } else {
1182 for (x = 0; x < rgd->rd_length; x++) {
1183 bi = rgd->rd_bits + x;
1184 set_bit(GBF_FULL, &bi->bi_flags);
1185 }
1186 }
1187}
1188
1189/**
1190 * gfs2_rgrp_go_instantiate - Read in a RG's header and bitmaps
1191 * @gl: the glock representing the rgrpd to read in
1192 *
1193 * Read in all of a Resource Group's header and bitmap blocks.
1194 * Caller must eventually call gfs2_rgrp_brelse() to free the bitmaps.
1195 *
1196 * Returns: errno
1197 */
1198
1199int gfs2_rgrp_go_instantiate(struct gfs2_glock *gl)
1200{
1201 struct gfs2_rgrpd *rgd = gl->gl_object;
1202 struct gfs2_sbd *sdp = rgd->rd_sbd;
1203 unsigned int length = rgd->rd_length;
1204 struct gfs2_bitmap *bi;
1205 unsigned int x, y;
1206 int error;
1207
1208 if (rgd->rd_bits[0].bi_bh != NULL)
1209 return 0;
1210
1211 for (x = 0; x < length; x++) {
1212 bi = rgd->rd_bits + x;
1213 error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, 0, &bi->bi_bh);
1214 if (error)
1215 goto fail;
1216 }
1217
1218 for (y = length; y--;) {
1219 bi = rgd->rd_bits + y;
1220 error = gfs2_meta_wait(sdp, bi->bi_bh);
1221 if (error)
1222 goto fail;
1223 if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB :
1224 GFS2_METATYPE_RG)) {
1225 error = -EIO;
1226 goto fail;
1227 }
1228 }
1229
1230 gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data);
1231 rgrp_set_bitmap_flags(rgd);
1232 rgd->rd_flags |= GFS2_RDF_CHECK;
1233 rgd->rd_free_clone = rgd->rd_free;
1234 GLOCK_BUG_ON(rgd->rd_gl, rgd->rd_reserved);
1235 /* max out the rgrp allocation failure point */
1236 rgd->rd_extfail_pt = rgd->rd_free;
1237 if (cpu_to_be32(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) {
1238 rgd->rd_rgl->rl_unlinked = cpu_to_be32(count_unlinked(rgd));
1239 gfs2_rgrp_ondisk2lvb(rgd->rd_rgl,
1240 rgd->rd_bits[0].bi_bh->b_data);
1241 } else if (sdp->sd_args.ar_rgrplvb) {
1242 if (!gfs2_rgrp_lvb_valid(rgd)){
1243 gfs2_consist_rgrpd(rgd);
1244 error = -EIO;
1245 goto fail;
1246 }
1247 if (rgd->rd_rgl->rl_unlinked == 0)
1248 rgd->rd_flags &= ~GFS2_RDF_CHECK;
1249 }
1250 return 0;
1251
1252fail:
1253 while (x--) {
1254 bi = rgd->rd_bits + x;
1255 brelse(bi->bi_bh);
1256 bi->bi_bh = NULL;
1257 gfs2_assert_warn(sdp, !bi->bi_clone);
1258 }
1259 return error;
1260}
1261
1262static int update_rgrp_lvb(struct gfs2_rgrpd *rgd, struct gfs2_holder *gh)
1263{
1264 u32 rl_flags;
1265
1266 if (!test_bit(GLF_INSTANTIATE_NEEDED, &gh->gh_gl->gl_flags))
1267 return 0;
1268
1269 if (cpu_to_be32(GFS2_MAGIC) != rgd->rd_rgl->rl_magic)
1270 return gfs2_instantiate(gh);
1271
1272 rl_flags = be32_to_cpu(rgd->rd_rgl->rl_flags);
1273 rl_flags &= ~GFS2_RDF_MASK;
1274 rgd->rd_flags &= GFS2_RDF_MASK;
1275 rgd->rd_flags |= (rl_flags | GFS2_RDF_CHECK);
1276 if (rgd->rd_rgl->rl_unlinked == 0)
1277 rgd->rd_flags &= ~GFS2_RDF_CHECK;
1278 rgd->rd_free = be32_to_cpu(rgd->rd_rgl->rl_free);
1279 rgrp_set_bitmap_flags(rgd);
1280 rgd->rd_free_clone = rgd->rd_free;
1281 GLOCK_BUG_ON(rgd->rd_gl, rgd->rd_reserved);
1282 /* max out the rgrp allocation failure point */
1283 rgd->rd_extfail_pt = rgd->rd_free;
1284 rgd->rd_dinodes = be32_to_cpu(rgd->rd_rgl->rl_dinodes);
1285 rgd->rd_igeneration = be64_to_cpu(rgd->rd_rgl->rl_igeneration);
1286 return 0;
1287}
1288
1289/**
1290 * gfs2_rgrp_brelse - Release RG bitmaps read in with gfs2_rgrp_bh_get()
1291 * @rgd: The resource group
1292 *
1293 */
1294
1295void gfs2_rgrp_brelse(struct gfs2_rgrpd *rgd)
1296{
1297 int x, length = rgd->rd_length;
1298
1299 for (x = 0; x < length; x++) {
1300 struct gfs2_bitmap *bi = rgd->rd_bits + x;
1301 if (bi->bi_bh) {
1302 brelse(bi->bi_bh);
1303 bi->bi_bh = NULL;
1304 }
1305 }
1306 set_bit(GLF_INSTANTIATE_NEEDED, &rgd->rd_gl->gl_flags);
1307}
1308
1309int gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset,
1310 struct buffer_head *bh,
1311 const struct gfs2_bitmap *bi, unsigned minlen, u64 *ptrimmed)
1312{
1313 struct super_block *sb = sdp->sd_vfs;
1314 u64 blk;
1315 sector_t start = 0;
1316 sector_t nr_blks = 0;
1317 int rv = -EIO;
1318 unsigned int x;
1319 u32 trimmed = 0;
1320 u8 diff;
1321
1322 for (x = 0; x < bi->bi_bytes; x++) {
1323 const u8 *clone = bi->bi_clone ? bi->bi_clone : bi->bi_bh->b_data;
1324 clone += bi->bi_offset;
1325 clone += x;
1326 if (bh) {
1327 const u8 *orig = bh->b_data + bi->bi_offset + x;
1328 diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1));
1329 } else {
1330 diff = ~(*clone | (*clone >> 1));
1331 }
1332 diff &= 0x55;
1333 if (diff == 0)
1334 continue;
1335 blk = offset + ((bi->bi_start + x) * GFS2_NBBY);
1336 while(diff) {
1337 if (diff & 1) {
1338 if (nr_blks == 0)
1339 goto start_new_extent;
1340 if ((start + nr_blks) != blk) {
1341 if (nr_blks >= minlen) {
1342 rv = sb_issue_discard(sb,
1343 start, nr_blks,
1344 GFP_NOFS, 0);
1345 if (rv)
1346 goto fail;
1347 trimmed += nr_blks;
1348 }
1349 nr_blks = 0;
1350start_new_extent:
1351 start = blk;
1352 }
1353 nr_blks++;
1354 }
1355 diff >>= 2;
1356 blk++;
1357 }
1358 }
1359 if (nr_blks >= minlen) {
1360 rv = sb_issue_discard(sb, start, nr_blks, GFP_NOFS, 0);
1361 if (rv)
1362 goto fail;
1363 trimmed += nr_blks;
1364 }
1365 if (ptrimmed)
1366 *ptrimmed = trimmed;
1367 return 0;
1368
1369fail:
1370 if (sdp->sd_args.ar_discard)
1371 fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem\n", rv);
1372 sdp->sd_args.ar_discard = 0;
1373 return rv;
1374}
1375
1376/**
1377 * gfs2_fitrim - Generate discard requests for unused bits of the filesystem
1378 * @filp: Any file on the filesystem
1379 * @argp: Pointer to the arguments (also used to pass result)
1380 *
1381 * Returns: 0 on success, otherwise error code
1382 */
1383
1384int gfs2_fitrim(struct file *filp, void __user *argp)
1385{
1386 struct inode *inode = file_inode(filp);
1387 struct gfs2_sbd *sdp = GFS2_SB(inode);
1388 struct block_device *bdev = sdp->sd_vfs->s_bdev;
1389 struct buffer_head *bh;
1390 struct gfs2_rgrpd *rgd;
1391 struct gfs2_rgrpd *rgd_end;
1392 struct gfs2_holder gh;
1393 struct fstrim_range r;
1394 int ret = 0;
1395 u64 amt;
1396 u64 trimmed = 0;
1397 u64 start, end, minlen;
1398 unsigned int x;
1399 unsigned bs_shift = sdp->sd_sb.sb_bsize_shift;
1400
1401 if (!capable(CAP_SYS_ADMIN))
1402 return -EPERM;
1403
1404 if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
1405 return -EROFS;
1406
1407 if (!bdev_max_discard_sectors(bdev))
1408 return -EOPNOTSUPP;
1409
1410 if (copy_from_user(&r, argp, sizeof(r)))
1411 return -EFAULT;
1412
1413 ret = gfs2_rindex_update(sdp);
1414 if (ret)
1415 return ret;
1416
1417 start = r.start >> bs_shift;
1418 end = start + (r.len >> bs_shift);
1419 minlen = max_t(u64, r.minlen, sdp->sd_sb.sb_bsize);
1420 minlen = max_t(u64, minlen, bdev_discard_granularity(bdev)) >> bs_shift;
1421
1422 if (end <= start || minlen > sdp->sd_max_rg_data)
1423 return -EINVAL;
1424
1425 rgd = gfs2_blk2rgrpd(sdp, start, 0);
1426 rgd_end = gfs2_blk2rgrpd(sdp, end, 0);
1427
1428 if ((gfs2_rgrpd_get_first(sdp) == gfs2_rgrpd_get_next(rgd_end))
1429 && (start > rgd_end->rd_data0 + rgd_end->rd_data))
1430 return -EINVAL; /* start is beyond the end of the fs */
1431
1432 while (1) {
1433
1434 ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1435 LM_FLAG_NODE_SCOPE, &gh);
1436 if (ret)
1437 goto out;
1438
1439 if (!(rgd->rd_flags & GFS2_RGF_TRIMMED)) {
1440 /* Trim each bitmap in the rgrp */
1441 for (x = 0; x < rgd->rd_length; x++) {
1442 struct gfs2_bitmap *bi = rgd->rd_bits + x;
1443 rgrp_lock_local(rgd);
1444 ret = gfs2_rgrp_send_discards(sdp,
1445 rgd->rd_data0, NULL, bi, minlen,
1446 &amt);
1447 rgrp_unlock_local(rgd);
1448 if (ret) {
1449 gfs2_glock_dq_uninit(&gh);
1450 goto out;
1451 }
1452 trimmed += amt;
1453 }
1454
1455 /* Mark rgrp as having been trimmed */
1456 ret = gfs2_trans_begin(sdp, RES_RG_HDR, 0);
1457 if (ret == 0) {
1458 bh = rgd->rd_bits[0].bi_bh;
1459 rgrp_lock_local(rgd);
1460 rgd->rd_flags |= GFS2_RGF_TRIMMED;
1461 gfs2_trans_add_meta(rgd->rd_gl, bh);
1462 gfs2_rgrp_out(rgd, bh->b_data);
1463 rgrp_unlock_local(rgd);
1464 gfs2_trans_end(sdp);
1465 }
1466 }
1467 gfs2_glock_dq_uninit(&gh);
1468
1469 if (rgd == rgd_end)
1470 break;
1471
1472 rgd = gfs2_rgrpd_get_next(rgd);
1473 }
1474
1475out:
1476 r.len = trimmed << bs_shift;
1477 if (copy_to_user(argp, &r, sizeof(r)))
1478 return -EFAULT;
1479
1480 return ret;
1481}
1482
1483/**
1484 * rs_insert - insert a new multi-block reservation into the rgrp's rb_tree
1485 * @ip: the inode structure
1486 *
1487 */
1488static void rs_insert(struct gfs2_inode *ip)
1489{
1490 struct rb_node **newn, *parent = NULL;
1491 int rc;
1492 struct gfs2_blkreserv *rs = &ip->i_res;
1493 struct gfs2_rgrpd *rgd = rs->rs_rgd;
1494
1495 BUG_ON(gfs2_rs_active(rs));
1496
1497 spin_lock(&rgd->rd_rsspin);
1498 newn = &rgd->rd_rstree.rb_node;
1499 while (*newn) {
1500 struct gfs2_blkreserv *cur =
1501 rb_entry(*newn, struct gfs2_blkreserv, rs_node);
1502
1503 parent = *newn;
1504 rc = rs_cmp(rs->rs_start, rs->rs_requested, cur);
1505 if (rc > 0)
1506 newn = &((*newn)->rb_right);
1507 else if (rc < 0)
1508 newn = &((*newn)->rb_left);
1509 else {
1510 spin_unlock(&rgd->rd_rsspin);
1511 WARN_ON(1);
1512 return;
1513 }
1514 }
1515
1516 rb_link_node(&rs->rs_node, parent, newn);
1517 rb_insert_color(&rs->rs_node, &rgd->rd_rstree);
1518
1519 /* Do our rgrp accounting for the reservation */
1520 rgd->rd_requested += rs->rs_requested; /* blocks requested */
1521 spin_unlock(&rgd->rd_rsspin);
1522 trace_gfs2_rs(rs, TRACE_RS_INSERT);
1523}
1524
1525/**
1526 * rgd_free - return the number of free blocks we can allocate
1527 * @rgd: the resource group
1528 * @rs: The reservation to free
1529 *
1530 * This function returns the number of free blocks for an rgrp.
1531 * That's the clone-free blocks (blocks that are free, not including those
1532 * still being used for unlinked files that haven't been deleted.)
1533 *
1534 * It also subtracts any blocks reserved by someone else, but does not
1535 * include free blocks that are still part of our current reservation,
1536 * because obviously we can (and will) allocate them.
1537 */
1538static inline u32 rgd_free(struct gfs2_rgrpd *rgd, struct gfs2_blkreserv *rs)
1539{
1540 u32 tot_reserved, tot_free;
1541
1542 if (WARN_ON_ONCE(rgd->rd_requested < rs->rs_requested))
1543 return 0;
1544 tot_reserved = rgd->rd_requested - rs->rs_requested;
1545
1546 if (rgd->rd_free_clone < tot_reserved)
1547 tot_reserved = 0;
1548
1549 tot_free = rgd->rd_free_clone - tot_reserved;
1550
1551 return tot_free;
1552}
1553
1554/**
1555 * rg_mblk_search - find a group of multiple free blocks to form a reservation
1556 * @rgd: the resource group descriptor
1557 * @ip: pointer to the inode for which we're reserving blocks
1558 * @ap: the allocation parameters
1559 *
1560 */
1561
1562static void rg_mblk_search(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip,
1563 const struct gfs2_alloc_parms *ap)
1564{
1565 struct gfs2_rbm rbm = { .rgd = rgd, };
1566 u64 goal;
1567 struct gfs2_blkreserv *rs = &ip->i_res;
1568 u32 extlen;
1569 u32 free_blocks, blocks_available;
1570 int ret;
1571 struct inode *inode = &ip->i_inode;
1572
1573 spin_lock(&rgd->rd_rsspin);
1574 free_blocks = rgd_free(rgd, rs);
1575 if (rgd->rd_free_clone < rgd->rd_requested)
1576 free_blocks = 0;
1577 blocks_available = rgd->rd_free_clone - rgd->rd_reserved;
1578 if (rgd == rs->rs_rgd)
1579 blocks_available += rs->rs_reserved;
1580 spin_unlock(&rgd->rd_rsspin);
1581
1582 if (S_ISDIR(inode->i_mode))
1583 extlen = 1;
1584 else {
1585 extlen = max_t(u32, atomic_read(&ip->i_sizehint), ap->target);
1586 extlen = clamp(extlen, (u32)RGRP_RSRV_MINBLKS, free_blocks);
1587 }
1588 if (free_blocks < extlen || blocks_available < extlen)
1589 return;
1590
1591 /* Find bitmap block that contains bits for goal block */
1592 if (rgrp_contains_block(rgd, ip->i_goal))
1593 goal = ip->i_goal;
1594 else
1595 goal = rgd->rd_last_alloc + rgd->rd_data0;
1596
1597 if (WARN_ON(gfs2_rbm_from_block(&rbm, goal)))
1598 return;
1599
1600 ret = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, &extlen, &ip->i_res, true);
1601 if (ret == 0) {
1602 rs->rs_start = gfs2_rbm_to_block(&rbm);
1603 rs->rs_requested = extlen;
1604 rs_insert(ip);
1605 } else {
1606 if (goal == rgd->rd_last_alloc + rgd->rd_data0)
1607 rgd->rd_last_alloc = 0;
1608 }
1609}
1610
1611/**
1612 * gfs2_next_unreserved_block - Return next block that is not reserved
1613 * @rgd: The resource group
1614 * @block: The starting block
1615 * @length: The required length
1616 * @ignore_rs: Reservation to ignore
1617 *
1618 * If the block does not appear in any reservation, then return the
1619 * block number unchanged. If it does appear in the reservation, then
1620 * keep looking through the tree of reservations in order to find the
1621 * first block number which is not reserved.
1622 */
1623
1624static u64 gfs2_next_unreserved_block(struct gfs2_rgrpd *rgd, u64 block,
1625 u32 length,
1626 struct gfs2_blkreserv *ignore_rs)
1627{
1628 struct gfs2_blkreserv *rs;
1629 struct rb_node *n;
1630 int rc;
1631
1632 spin_lock(&rgd->rd_rsspin);
1633 n = rgd->rd_rstree.rb_node;
1634 while (n) {
1635 rs = rb_entry(n, struct gfs2_blkreserv, rs_node);
1636 rc = rs_cmp(block, length, rs);
1637 if (rc < 0)
1638 n = n->rb_left;
1639 else if (rc > 0)
1640 n = n->rb_right;
1641 else
1642 break;
1643 }
1644
1645 if (n) {
1646 while (rs_cmp(block, length, rs) == 0 && rs != ignore_rs) {
1647 block = rs->rs_start + rs->rs_requested;
1648 n = n->rb_right;
1649 if (n == NULL)
1650 break;
1651 rs = rb_entry(n, struct gfs2_blkreserv, rs_node);
1652 }
1653 }
1654
1655 spin_unlock(&rgd->rd_rsspin);
1656 return block;
1657}
1658
1659/**
1660 * gfs2_reservation_check_and_update - Check for reservations during block alloc
1661 * @rbm: The current position in the resource group
1662 * @rs: Our own reservation
1663 * @minext: The minimum extent length
1664 * @maxext: A pointer to the maximum extent structure
1665 *
1666 * This checks the current position in the rgrp to see whether there is
1667 * a reservation covering this block. If not then this function is a
1668 * no-op. If there is, then the position is moved to the end of the
1669 * contiguous reservation(s) so that we are pointing at the first
1670 * non-reserved block.
1671 *
1672 * Returns: 0 if no reservation, 1 if @rbm has changed, otherwise an error
1673 */
1674
1675static int gfs2_reservation_check_and_update(struct gfs2_rbm *rbm,
1676 struct gfs2_blkreserv *rs,
1677 u32 minext,
1678 struct gfs2_extent *maxext)
1679{
1680 u64 block = gfs2_rbm_to_block(rbm);
1681 u32 extlen = 1;
1682 u64 nblock;
1683
1684 /*
1685 * If we have a minimum extent length, then skip over any extent
1686 * which is less than the min extent length in size.
1687 */
1688 if (minext > 1) {
1689 extlen = gfs2_free_extlen(rbm, minext);
1690 if (extlen <= maxext->len)
1691 goto fail;
1692 }
1693
1694 /*
1695 * Check the extent which has been found against the reservations
1696 * and skip if parts of it are already reserved
1697 */
1698 nblock = gfs2_next_unreserved_block(rbm->rgd, block, extlen, rs);
1699 if (nblock == block) {
1700 if (!minext || extlen >= minext)
1701 return 0;
1702
1703 if (extlen > maxext->len) {
1704 maxext->len = extlen;
1705 maxext->rbm = *rbm;
1706 }
1707 } else {
1708 u64 len = nblock - block;
1709 if (len >= (u64)1 << 32)
1710 return -E2BIG;
1711 extlen = len;
1712 }
1713fail:
1714 if (gfs2_rbm_add(rbm, extlen))
1715 return -E2BIG;
1716 return 1;
1717}
1718
1719/**
1720 * gfs2_rbm_find - Look for blocks of a particular state
1721 * @rbm: Value/result starting position and final position
1722 * @state: The state which we want to find
1723 * @minext: Pointer to the requested extent length
1724 * This is updated to be the actual reservation size.
1725 * @rs: Our own reservation (NULL to skip checking for reservations)
1726 * @nowrap: Stop looking at the end of the rgrp, rather than wrapping
1727 * around until we've reached the starting point.
1728 *
1729 * Side effects:
1730 * - If looking for free blocks, we set GBF_FULL on each bitmap which
1731 * has no free blocks in it.
1732 * - If looking for free blocks, we set rd_extfail_pt on each rgrp which
1733 * has come up short on a free block search.
1734 *
1735 * Returns: 0 on success, -ENOSPC if there is no block of the requested state
1736 */
1737
1738static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 *minext,
1739 struct gfs2_blkreserv *rs, bool nowrap)
1740{
1741 bool scan_from_start = rbm->bii == 0 && rbm->offset == 0;
1742 struct buffer_head *bh;
1743 int last_bii;
1744 u32 offset;
1745 u8 *buffer;
1746 bool wrapped = false;
1747 int ret;
1748 struct gfs2_bitmap *bi;
1749 struct gfs2_extent maxext = { .rbm.rgd = rbm->rgd, };
1750
1751 /*
1752 * Determine the last bitmap to search. If we're not starting at the
1753 * beginning of a bitmap, we need to search that bitmap twice to scan
1754 * the entire resource group.
1755 */
1756 last_bii = rbm->bii - (rbm->offset == 0);
1757
1758 while(1) {
1759 bi = rbm_bi(rbm);
1760 if (test_bit(GBF_FULL, &bi->bi_flags) &&
1761 (state == GFS2_BLKST_FREE))
1762 goto next_bitmap;
1763
1764 bh = bi->bi_bh;
1765 buffer = bh->b_data + bi->bi_offset;
1766 WARN_ON(!buffer_uptodate(bh));
1767 if (state != GFS2_BLKST_UNLINKED && bi->bi_clone)
1768 buffer = bi->bi_clone + bi->bi_offset;
1769 offset = gfs2_bitfit(buffer, bi->bi_bytes, rbm->offset, state);
1770 if (offset == BFITNOENT) {
1771 if (state == GFS2_BLKST_FREE && rbm->offset == 0)
1772 set_bit(GBF_FULL, &bi->bi_flags);
1773 goto next_bitmap;
1774 }
1775 rbm->offset = offset;
1776 if (!rs || !minext)
1777 return 0;
1778
1779 ret = gfs2_reservation_check_and_update(rbm, rs, *minext,
1780 &maxext);
1781 if (ret == 0)
1782 return 0;
1783 if (ret > 0)
1784 goto next_iter;
1785 if (ret == -E2BIG) {
1786 rbm->bii = 0;
1787 rbm->offset = 0;
1788 goto res_covered_end_of_rgrp;
1789 }
1790 return ret;
1791
1792next_bitmap: /* Find next bitmap in the rgrp */
1793 rbm->offset = 0;
1794 rbm->bii++;
1795 if (rbm->bii == rbm->rgd->rd_length)
1796 rbm->bii = 0;
1797res_covered_end_of_rgrp:
1798 if (rbm->bii == 0) {
1799 if (wrapped)
1800 break;
1801 wrapped = true;
1802 if (nowrap)
1803 break;
1804 }
1805next_iter:
1806 /* Have we scanned the entire resource group? */
1807 if (wrapped && rbm->bii > last_bii)
1808 break;
1809 }
1810
1811 if (state != GFS2_BLKST_FREE)
1812 return -ENOSPC;
1813
1814 /* If the extent was too small, and it's smaller than the smallest
1815 to have failed before, remember for future reference that it's
1816 useless to search this rgrp again for this amount or more. */
1817 if (wrapped && (scan_from_start || rbm->bii > last_bii) &&
1818 *minext < rbm->rgd->rd_extfail_pt)
1819 rbm->rgd->rd_extfail_pt = *minext - 1;
1820
1821 /* If the maximum extent we found is big enough to fulfill the
1822 minimum requirements, use it anyway. */
1823 if (maxext.len) {
1824 *rbm = maxext.rbm;
1825 *minext = maxext.len;
1826 return 0;
1827 }
1828
1829 return -ENOSPC;
1830}
1831
1832/**
1833 * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes
1834 * @rgd: The rgrp
1835 * @last_unlinked: block address of the last dinode we unlinked
1836 * @skip: block address we should explicitly not unlink
1837 *
1838 * Returns: 0 if no error
1839 * The inode, if one has been found, in inode.
1840 */
1841
1842static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip)
1843{
1844 u64 block;
1845 struct gfs2_sbd *sdp = rgd->rd_sbd;
1846 struct gfs2_glock *gl;
1847 struct gfs2_inode *ip;
1848 int error;
1849 int found = 0;
1850 struct gfs2_rbm rbm = { .rgd = rgd, .bii = 0, .offset = 0 };
1851
1852 while (1) {
1853 error = gfs2_rbm_find(&rbm, GFS2_BLKST_UNLINKED, NULL, NULL,
1854 true);
1855 if (error == -ENOSPC)
1856 break;
1857 if (WARN_ON_ONCE(error))
1858 break;
1859
1860 block = gfs2_rbm_to_block(&rbm);
1861 if (gfs2_rbm_from_block(&rbm, block + 1))
1862 break;
1863 if (*last_unlinked != NO_BLOCK && block <= *last_unlinked)
1864 continue;
1865 if (block == skip)
1866 continue;
1867 *last_unlinked = block;
1868
1869 error = gfs2_glock_get(sdp, block, &gfs2_iopen_glops, CREATE, &gl);
1870 if (error)
1871 continue;
1872
1873 /* If the inode is already in cache, we can ignore it here
1874 * because the existing inode disposal code will deal with
1875 * it when all refs have gone away. Accessing gl_object like
1876 * this is not safe in general. Here it is ok because we do
1877 * not dereference the pointer, and we only need an approx
1878 * answer to whether it is NULL or not.
1879 */
1880 ip = gl->gl_object;
1881
1882 if (ip || !gfs2_queue_verify_delete(gl, false))
1883 gfs2_glock_put(gl);
1884 else
1885 found++;
1886
1887 /* Limit reclaim to sensible number of tasks */
1888 if (found > NR_CPUS)
1889 return;
1890 }
1891
1892 rgd->rd_flags &= ~GFS2_RDF_CHECK;
1893 return;
1894}
1895
1896/**
1897 * gfs2_rgrp_congested - Use stats to figure out whether an rgrp is congested
1898 * @rgd: The rgrp in question
1899 * @loops: An indication of how picky we can be (0=very, 1=less so)
1900 *
1901 * This function uses the recently added glock statistics in order to
1902 * figure out whether a parciular resource group is suffering from
1903 * contention from multiple nodes. This is done purely on the basis
1904 * of timings, since this is the only data we have to work with and
1905 * our aim here is to reject a resource group which is highly contended
1906 * but (very important) not to do this too often in order to ensure that
1907 * we do not land up introducing fragmentation by changing resource
1908 * groups when not actually required.
1909 *
1910 * The calculation is fairly simple, we want to know whether the SRTTB
1911 * (i.e. smoothed round trip time for blocking operations) to acquire
1912 * the lock for this rgrp's glock is significantly greater than the
1913 * time taken for resource groups on average. We introduce a margin in
1914 * the form of the variable @var which is computed as the sum of the two
1915 * respective variences, and multiplied by a factor depending on @loops
1916 * and whether we have a lot of data to base the decision on. This is
1917 * then tested against the square difference of the means in order to
1918 * decide whether the result is statistically significant or not.
1919 *
1920 * Returns: A boolean verdict on the congestion status
1921 */
1922
1923static bool gfs2_rgrp_congested(const struct gfs2_rgrpd *rgd, int loops)
1924{
1925 const struct gfs2_glock *gl = rgd->rd_gl;
1926 const struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1927 struct gfs2_lkstats *st;
1928 u64 r_dcount, l_dcount;
1929 u64 l_srttb, a_srttb = 0;
1930 s64 srttb_diff;
1931 u64 sqr_diff;
1932 u64 var;
1933 int cpu, nonzero = 0;
1934
1935 preempt_disable();
1936 for_each_present_cpu(cpu) {
1937 st = &per_cpu_ptr(sdp->sd_lkstats, cpu)->lkstats[LM_TYPE_RGRP];
1938 if (st->stats[GFS2_LKS_SRTTB]) {
1939 a_srttb += st->stats[GFS2_LKS_SRTTB];
1940 nonzero++;
1941 }
1942 }
1943 st = &this_cpu_ptr(sdp->sd_lkstats)->lkstats[LM_TYPE_RGRP];
1944 if (nonzero)
1945 do_div(a_srttb, nonzero);
1946 r_dcount = st->stats[GFS2_LKS_DCOUNT];
1947 var = st->stats[GFS2_LKS_SRTTVARB] +
1948 gl->gl_stats.stats[GFS2_LKS_SRTTVARB];
1949 preempt_enable();
1950
1951 l_srttb = gl->gl_stats.stats[GFS2_LKS_SRTTB];
1952 l_dcount = gl->gl_stats.stats[GFS2_LKS_DCOUNT];
1953
1954 if ((l_dcount < 1) || (r_dcount < 1) || (a_srttb == 0))
1955 return false;
1956
1957 srttb_diff = a_srttb - l_srttb;
1958 sqr_diff = srttb_diff * srttb_diff;
1959
1960 var *= 2;
1961 if (l_dcount < 8 || r_dcount < 8)
1962 var *= 2;
1963 if (loops == 1)
1964 var *= 2;
1965
1966 return ((srttb_diff < 0) && (sqr_diff > var));
1967}
1968
1969/**
1970 * gfs2_rgrp_used_recently - test if an rgrp has been used recently
1971 * @rs: The block reservation with the rgrp to test
1972 * @msecs: The time limit in milliseconds
1973 *
1974 * Returns: True if the rgrp glock has been used within the time limit
1975 */
1976static bool gfs2_rgrp_used_recently(const struct gfs2_blkreserv *rs,
1977 u64 msecs)
1978{
1979 u64 tdiff;
1980
1981 tdiff = ktime_to_ns(ktime_sub(ktime_get_real(),
1982 rs->rs_rgd->rd_gl->gl_dstamp));
1983
1984 return tdiff > (msecs * 1000 * 1000);
1985}
1986
1987static u32 gfs2_orlov_skip(const struct gfs2_inode *ip)
1988{
1989 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1990
1991 return get_random_u32() % sdp->sd_rgrps;
1992}
1993
1994static bool gfs2_select_rgrp(struct gfs2_rgrpd **pos, const struct gfs2_rgrpd *begin)
1995{
1996 struct gfs2_rgrpd *rgd = *pos;
1997 struct gfs2_sbd *sdp = rgd->rd_sbd;
1998
1999 rgd = gfs2_rgrpd_get_next(rgd);
2000 if (rgd == NULL)
2001 rgd = gfs2_rgrpd_get_first(sdp);
2002 *pos = rgd;
2003 if (rgd != begin) /* If we didn't wrap */
2004 return true;
2005 return false;
2006}
2007
2008/**
2009 * fast_to_acquire - determine if a resource group will be fast to acquire
2010 * @rgd: The rgrp
2011 *
2012 * If this is one of our preferred rgrps, it should be quicker to acquire,
2013 * because we tried to set ourselves up as dlm lock master.
2014 */
2015static inline int fast_to_acquire(struct gfs2_rgrpd *rgd)
2016{
2017 struct gfs2_glock *gl = rgd->rd_gl;
2018
2019 if (gl->gl_state != LM_ST_UNLOCKED && list_empty(&gl->gl_holders) &&
2020 !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
2021 !test_bit(GLF_DEMOTE, &gl->gl_flags))
2022 return 1;
2023 if (rgd->rd_flags & GFS2_RDF_PREFERRED)
2024 return 1;
2025 return 0;
2026}
2027
2028/**
2029 * gfs2_inplace_reserve - Reserve space in the filesystem
2030 * @ip: the inode to reserve space for
2031 * @ap: the allocation parameters
2032 *
2033 * We try our best to find an rgrp that has at least ap->target blocks
2034 * available. After a couple of passes (loops == 2), the prospects of finding
2035 * such an rgrp diminish. At this stage, we return the first rgrp that has
2036 * at least ap->min_target blocks available.
2037 *
2038 * Returns: 0 on success,
2039 * -ENOMEM if a suitable rgrp can't be found
2040 * errno otherwise
2041 */
2042
2043int gfs2_inplace_reserve(struct gfs2_inode *ip, struct gfs2_alloc_parms *ap)
2044{
2045 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2046 struct gfs2_rgrpd *begin = NULL;
2047 struct gfs2_blkreserv *rs = &ip->i_res;
2048 int error = 0, flags = LM_FLAG_NODE_SCOPE;
2049 bool rg_locked;
2050 u64 last_unlinked = NO_BLOCK;
2051 u32 target = ap->target;
2052 int loops = 0;
2053 u32 free_blocks, blocks_available, skip = 0;
2054
2055 BUG_ON(rs->rs_reserved);
2056
2057 if (sdp->sd_args.ar_rgrplvb)
2058 flags |= GL_SKIP;
2059 if (gfs2_assert_warn(sdp, target))
2060 return -EINVAL;
2061 if (gfs2_rs_active(rs)) {
2062 begin = rs->rs_rgd;
2063 } else if (rs->rs_rgd &&
2064 rgrp_contains_block(rs->rs_rgd, ip->i_goal)) {
2065 begin = rs->rs_rgd;
2066 } else {
2067 check_and_update_goal(ip);
2068 rs->rs_rgd = begin = gfs2_blk2rgrpd(sdp, ip->i_goal, 1);
2069 }
2070 if (S_ISDIR(ip->i_inode.i_mode) && (ap->aflags & GFS2_AF_ORLOV))
2071 skip = gfs2_orlov_skip(ip);
2072 if (rs->rs_rgd == NULL)
2073 return -EBADSLT;
2074
2075 while (loops < 3) {
2076 struct gfs2_rgrpd *rgd;
2077
2078 rg_locked = gfs2_glock_is_locked_by_me(rs->rs_rgd->rd_gl);
2079 if (rg_locked) {
2080 rgrp_lock_local(rs->rs_rgd);
2081 } else {
2082 if (skip && skip--)
2083 goto next_rgrp;
2084 if (!gfs2_rs_active(rs)) {
2085 if (loops == 0 &&
2086 !fast_to_acquire(rs->rs_rgd))
2087 goto next_rgrp;
2088 if ((loops < 2) &&
2089 gfs2_rgrp_used_recently(rs, 1000) &&
2090 gfs2_rgrp_congested(rs->rs_rgd, loops))
2091 goto next_rgrp;
2092 }
2093 error = gfs2_glock_nq_init(rs->rs_rgd->rd_gl,
2094 LM_ST_EXCLUSIVE, flags,
2095 &ip->i_rgd_gh);
2096 if (unlikely(error))
2097 return error;
2098 rgrp_lock_local(rs->rs_rgd);
2099 if (!gfs2_rs_active(rs) && (loops < 2) &&
2100 gfs2_rgrp_congested(rs->rs_rgd, loops))
2101 goto skip_rgrp;
2102 if (sdp->sd_args.ar_rgrplvb) {
2103 error = update_rgrp_lvb(rs->rs_rgd,
2104 &ip->i_rgd_gh);
2105 if (unlikely(error)) {
2106 rgrp_unlock_local(rs->rs_rgd);
2107 gfs2_glock_dq_uninit(&ip->i_rgd_gh);
2108 return error;
2109 }
2110 }
2111 }
2112
2113 /* Skip unusable resource groups */
2114 if ((rs->rs_rgd->rd_flags & (GFS2_RGF_NOALLOC |
2115 GFS2_RDF_ERROR)) ||
2116 (loops == 0 && target > rs->rs_rgd->rd_extfail_pt))
2117 goto skip_rgrp;
2118
2119 if (sdp->sd_args.ar_rgrplvb) {
2120 error = gfs2_instantiate(&ip->i_rgd_gh);
2121 if (error)
2122 goto skip_rgrp;
2123 }
2124
2125 /* Get a reservation if we don't already have one */
2126 if (!gfs2_rs_active(rs))
2127 rg_mblk_search(rs->rs_rgd, ip, ap);
2128
2129 /* Skip rgrps when we can't get a reservation on first pass */
2130 if (!gfs2_rs_active(rs) && (loops < 1))
2131 goto check_rgrp;
2132
2133 /* If rgrp has enough free space, use it */
2134 rgd = rs->rs_rgd;
2135 spin_lock(&rgd->rd_rsspin);
2136 free_blocks = rgd_free(rgd, rs);
2137 blocks_available = rgd->rd_free_clone - rgd->rd_reserved;
2138 if (free_blocks < target || blocks_available < target) {
2139 spin_unlock(&rgd->rd_rsspin);
2140 goto check_rgrp;
2141 }
2142 rs->rs_reserved = ap->target;
2143 if (rs->rs_reserved > blocks_available)
2144 rs->rs_reserved = blocks_available;
2145 rgd->rd_reserved += rs->rs_reserved;
2146 spin_unlock(&rgd->rd_rsspin);
2147 rgrp_unlock_local(rs->rs_rgd);
2148 return 0;
2149check_rgrp:
2150 /* Check for unlinked inodes which can be reclaimed */
2151 if (rs->rs_rgd->rd_flags & GFS2_RDF_CHECK)
2152 try_rgrp_unlink(rs->rs_rgd, &last_unlinked,
2153 ip->i_no_addr);
2154skip_rgrp:
2155 rgrp_unlock_local(rs->rs_rgd);
2156
2157 /* Drop reservation, if we couldn't use reserved rgrp */
2158 if (gfs2_rs_active(rs))
2159 gfs2_rs_deltree(rs);
2160
2161 /* Unlock rgrp if required */
2162 if (!rg_locked)
2163 gfs2_glock_dq_uninit(&ip->i_rgd_gh);
2164next_rgrp:
2165 /* Find the next rgrp, and continue looking */
2166 if (gfs2_select_rgrp(&rs->rs_rgd, begin))
2167 continue;
2168 if (skip)
2169 continue;
2170
2171 /* If we've scanned all the rgrps, but found no free blocks
2172 * then this checks for some less likely conditions before
2173 * trying again.
2174 */
2175 loops++;
2176 /* Check that fs hasn't grown if writing to rindex */
2177 if (ip == GFS2_I(sdp->sd_rindex) && !sdp->sd_rindex_uptodate) {
2178 error = gfs2_ri_update(ip);
2179 if (error)
2180 return error;
2181 }
2182 /* Flushing the log may release space */
2183 if (loops == 2) {
2184 if (ap->min_target)
2185 target = ap->min_target;
2186 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
2187 GFS2_LFC_INPLACE_RESERVE);
2188 }
2189 }
2190
2191 return -ENOSPC;
2192}
2193
2194/**
2195 * gfs2_inplace_release - release an inplace reservation
2196 * @ip: the inode the reservation was taken out on
2197 *
2198 * Release a reservation made by gfs2_inplace_reserve().
2199 */
2200
2201void gfs2_inplace_release(struct gfs2_inode *ip)
2202{
2203 struct gfs2_blkreserv *rs = &ip->i_res;
2204
2205 if (rs->rs_reserved) {
2206 struct gfs2_rgrpd *rgd = rs->rs_rgd;
2207
2208 spin_lock(&rgd->rd_rsspin);
2209 GLOCK_BUG_ON(rgd->rd_gl, rgd->rd_reserved < rs->rs_reserved);
2210 rgd->rd_reserved -= rs->rs_reserved;
2211 spin_unlock(&rgd->rd_rsspin);
2212 rs->rs_reserved = 0;
2213 }
2214 if (gfs2_holder_initialized(&ip->i_rgd_gh))
2215 gfs2_glock_dq_uninit(&ip->i_rgd_gh);
2216}
2217
2218/**
2219 * gfs2_alloc_extent - allocate an extent from a given bitmap
2220 * @rbm: the resource group information
2221 * @dinode: TRUE if the first block we allocate is for a dinode
2222 * @n: The extent length (value/result)
2223 *
2224 * Add the bitmap buffer to the transaction.
2225 * Set the found bits to @new_state to change block's allocation state.
2226 */
2227static void gfs2_alloc_extent(const struct gfs2_rbm *rbm, bool dinode,
2228 unsigned int *n)
2229{
2230 struct gfs2_rbm pos = { .rgd = rbm->rgd, };
2231 const unsigned int elen = *n;
2232 u64 block;
2233 int ret;
2234
2235 *n = 1;
2236 block = gfs2_rbm_to_block(rbm);
2237 gfs2_trans_add_meta(rbm->rgd->rd_gl, rbm_bi(rbm)->bi_bh);
2238 gfs2_setbit(rbm, true, dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED);
2239 block++;
2240 while (*n < elen) {
2241 ret = gfs2_rbm_from_block(&pos, block);
2242 if (ret || gfs2_testbit(&pos, true) != GFS2_BLKST_FREE)
2243 break;
2244 gfs2_trans_add_meta(pos.rgd->rd_gl, rbm_bi(&pos)->bi_bh);
2245 gfs2_setbit(&pos, true, GFS2_BLKST_USED);
2246 (*n)++;
2247 block++;
2248 }
2249}
2250
2251/**
2252 * rgblk_free - Change alloc state of given block(s)
2253 * @sdp: the filesystem
2254 * @rgd: the resource group the blocks are in
2255 * @bstart: the start of a run of blocks to free
2256 * @blen: the length of the block run (all must lie within ONE RG!)
2257 * @new_state: GFS2_BLKST_XXX the after-allocation block state
2258 */
2259
2260static void rgblk_free(struct gfs2_sbd *sdp, struct gfs2_rgrpd *rgd,
2261 u64 bstart, u32 blen, unsigned char new_state)
2262{
2263 struct gfs2_rbm rbm;
2264 struct gfs2_bitmap *bi, *bi_prev = NULL;
2265
2266 rbm.rgd = rgd;
2267 if (WARN_ON_ONCE(gfs2_rbm_from_block(&rbm, bstart)))
2268 return;
2269 while (blen--) {
2270 bi = rbm_bi(&rbm);
2271 if (bi != bi_prev) {
2272 if (!bi->bi_clone) {
2273 bi->bi_clone = kmalloc(bi->bi_bh->b_size,
2274 GFP_NOFS | __GFP_NOFAIL);
2275 memcpy(bi->bi_clone + bi->bi_offset,
2276 bi->bi_bh->b_data + bi->bi_offset,
2277 bi->bi_bytes);
2278 }
2279 gfs2_trans_add_meta(rbm.rgd->rd_gl, bi->bi_bh);
2280 bi_prev = bi;
2281 }
2282 gfs2_setbit(&rbm, false, new_state);
2283 gfs2_rbm_add(&rbm, 1);
2284 }
2285}
2286
2287/**
2288 * gfs2_rgrp_dump - print out an rgrp
2289 * @seq: The iterator
2290 * @rgd: The rgrp in question
2291 * @fs_id_buf: pointer to file system id (if requested)
2292 *
2293 */
2294
2295void gfs2_rgrp_dump(struct seq_file *seq, struct gfs2_rgrpd *rgd,
2296 const char *fs_id_buf)
2297{
2298 struct gfs2_blkreserv *trs;
2299 const struct rb_node *n;
2300
2301 spin_lock(&rgd->rd_rsspin);
2302 gfs2_print_dbg(seq, "%s R: n:%llu f:%02x b:%u/%u i:%u q:%u r:%u e:%u\n",
2303 fs_id_buf,
2304 (unsigned long long)rgd->rd_addr, rgd->rd_flags,
2305 rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes,
2306 rgd->rd_requested, rgd->rd_reserved, rgd->rd_extfail_pt);
2307 if (rgd->rd_sbd->sd_args.ar_rgrplvb && rgd->rd_rgl) {
2308 struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl;
2309
2310 gfs2_print_dbg(seq, "%s L: f:%02x b:%u i:%u\n", fs_id_buf,
2311 be32_to_cpu(rgl->rl_flags),
2312 be32_to_cpu(rgl->rl_free),
2313 be32_to_cpu(rgl->rl_dinodes));
2314 }
2315 for (n = rb_first(&rgd->rd_rstree); n; n = rb_next(&trs->rs_node)) {
2316 trs = rb_entry(n, struct gfs2_blkreserv, rs_node);
2317 dump_rs(seq, trs, fs_id_buf);
2318 }
2319 spin_unlock(&rgd->rd_rsspin);
2320}
2321
2322static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd)
2323{
2324 struct gfs2_sbd *sdp = rgd->rd_sbd;
2325 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2326
2327 fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n",
2328 (unsigned long long)rgd->rd_addr);
2329 fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n");
2330 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2331 gfs2_rgrp_dump(NULL, rgd, fs_id_buf);
2332 rgd->rd_flags |= GFS2_RDF_ERROR;
2333}
2334
2335/**
2336 * gfs2_adjust_reservation - Adjust (or remove) a reservation after allocation
2337 * @ip: The inode we have just allocated blocks for
2338 * @rbm: The start of the allocated blocks
2339 * @len: The extent length
2340 *
2341 * Adjusts a reservation after an allocation has taken place. If the
2342 * reservation does not match the allocation, or if it is now empty
2343 * then it is removed.
2344 */
2345
2346static void gfs2_adjust_reservation(struct gfs2_inode *ip,
2347 const struct gfs2_rbm *rbm, unsigned len)
2348{
2349 struct gfs2_blkreserv *rs = &ip->i_res;
2350 struct gfs2_rgrpd *rgd = rbm->rgd;
2351
2352 BUG_ON(rs->rs_reserved < len);
2353 rs->rs_reserved -= len;
2354 if (gfs2_rs_active(rs)) {
2355 u64 start = gfs2_rbm_to_block(rbm);
2356
2357 if (rs->rs_start == start) {
2358 unsigned int rlen;
2359
2360 rs->rs_start += len;
2361 rlen = min(rs->rs_requested, len);
2362 rs->rs_requested -= rlen;
2363 rgd->rd_requested -= rlen;
2364 trace_gfs2_rs(rs, TRACE_RS_CLAIM);
2365 if (rs->rs_start < rgd->rd_data0 + rgd->rd_data &&
2366 rs->rs_requested)
2367 return;
2368 /* We used up our block reservation, so we should
2369 reserve more blocks next time. */
2370 atomic_add(RGRP_RSRV_ADDBLKS, &ip->i_sizehint);
2371 }
2372 __rs_deltree(rs);
2373 }
2374}
2375
2376/**
2377 * gfs2_set_alloc_start - Set starting point for block allocation
2378 * @rbm: The rbm which will be set to the required location
2379 * @ip: The gfs2 inode
2380 * @dinode: Flag to say if allocation includes a new inode
2381 *
2382 * This sets the starting point from the reservation if one is active
2383 * otherwise it falls back to guessing a start point based on the
2384 * inode's goal block or the last allocation point in the rgrp.
2385 */
2386
2387static void gfs2_set_alloc_start(struct gfs2_rbm *rbm,
2388 const struct gfs2_inode *ip, bool dinode)
2389{
2390 u64 goal;
2391
2392 if (gfs2_rs_active(&ip->i_res)) {
2393 goal = ip->i_res.rs_start;
2394 } else {
2395 if (!dinode && rgrp_contains_block(rbm->rgd, ip->i_goal))
2396 goal = ip->i_goal;
2397 else
2398 goal = rbm->rgd->rd_last_alloc + rbm->rgd->rd_data0;
2399 }
2400 if (WARN_ON_ONCE(gfs2_rbm_from_block(rbm, goal))) {
2401 rbm->bii = 0;
2402 rbm->offset = 0;
2403 }
2404}
2405
2406/**
2407 * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode
2408 * @ip: the inode to allocate the block for
2409 * @bn: Used to return the starting block number
2410 * @nblocks: requested number of blocks/extent length (value/result)
2411 * @dinode: 1 if we're allocating a dinode block, else 0
2412 *
2413 * Returns: 0 or error
2414 */
2415
2416int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *nblocks,
2417 bool dinode)
2418{
2419 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2420 struct buffer_head *dibh;
2421 struct gfs2_rbm rbm = { .rgd = ip->i_res.rs_rgd, };
2422 u64 block; /* block, within the file system scope */
2423 u32 minext = 1;
2424 int error = -ENOSPC;
2425
2426 BUG_ON(ip->i_res.rs_reserved < *nblocks);
2427
2428 rgrp_lock_local(rbm.rgd);
2429 if (gfs2_rs_active(&ip->i_res)) {
2430 gfs2_set_alloc_start(&rbm, ip, dinode);
2431 error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, &minext, &ip->i_res, false);
2432 }
2433 if (error == -ENOSPC) {
2434 gfs2_set_alloc_start(&rbm, ip, dinode);
2435 error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, &minext, NULL, false);
2436 }
2437
2438 /* Since all blocks are reserved in advance, this shouldn't happen */
2439 if (error) {
2440 fs_warn(sdp, "inum=%llu error=%d, nblocks=%u, full=%d fail_pt=%d\n",
2441 (unsigned long long)ip->i_no_addr, error, *nblocks,
2442 test_bit(GBF_FULL, &rbm.rgd->rd_bits->bi_flags),
2443 rbm.rgd->rd_extfail_pt);
2444 goto rgrp_error;
2445 }
2446
2447 gfs2_alloc_extent(&rbm, dinode, nblocks);
2448 block = gfs2_rbm_to_block(&rbm);
2449 rbm.rgd->rd_last_alloc = block - rbm.rgd->rd_data0;
2450 if (!dinode) {
2451 ip->i_goal = block + *nblocks - 1;
2452 error = gfs2_meta_inode_buffer(ip, &dibh);
2453 if (error == 0) {
2454 struct gfs2_dinode *di =
2455 (struct gfs2_dinode *)dibh->b_data;
2456 gfs2_trans_add_meta(ip->i_gl, dibh);
2457 di->di_goal_meta = di->di_goal_data =
2458 cpu_to_be64(ip->i_goal);
2459 brelse(dibh);
2460 }
2461 }
2462 spin_lock(&rbm.rgd->rd_rsspin);
2463 gfs2_adjust_reservation(ip, &rbm, *nblocks);
2464 if (rbm.rgd->rd_free < *nblocks || rbm.rgd->rd_reserved < *nblocks) {
2465 fs_warn(sdp, "nblocks=%u\n", *nblocks);
2466 spin_unlock(&rbm.rgd->rd_rsspin);
2467 goto rgrp_error;
2468 }
2469 GLOCK_BUG_ON(rbm.rgd->rd_gl, rbm.rgd->rd_reserved < *nblocks);
2470 GLOCK_BUG_ON(rbm.rgd->rd_gl, rbm.rgd->rd_free_clone < *nblocks);
2471 GLOCK_BUG_ON(rbm.rgd->rd_gl, rbm.rgd->rd_free < *nblocks);
2472 rbm.rgd->rd_reserved -= *nblocks;
2473 rbm.rgd->rd_free_clone -= *nblocks;
2474 rbm.rgd->rd_free -= *nblocks;
2475 spin_unlock(&rbm.rgd->rd_rsspin);
2476 if (dinode) {
2477 u64 generation;
2478
2479 rbm.rgd->rd_dinodes++;
2480 generation = rbm.rgd->rd_igeneration++;
2481 if (generation == 0)
2482 generation = rbm.rgd->rd_igeneration++;
2483 ip->i_generation = generation;
2484 }
2485
2486 gfs2_trans_add_meta(rbm.rgd->rd_gl, rbm.rgd->rd_bits[0].bi_bh);
2487 gfs2_rgrp_out(rbm.rgd, rbm.rgd->rd_bits[0].bi_bh->b_data);
2488 rgrp_unlock_local(rbm.rgd);
2489
2490 gfs2_statfs_change(sdp, 0, -(s64)*nblocks, dinode ? 1 : 0);
2491 if (dinode)
2492 gfs2_trans_remove_revoke(sdp, block, *nblocks);
2493
2494 gfs2_quota_change(ip, *nblocks, ip->i_inode.i_uid, ip->i_inode.i_gid);
2495
2496 trace_gfs2_block_alloc(ip, rbm.rgd, block, *nblocks,
2497 dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED);
2498 *bn = block;
2499 return 0;
2500
2501rgrp_error:
2502 rgrp_unlock_local(rbm.rgd);
2503 gfs2_rgrp_error(rbm.rgd);
2504 return -EIO;
2505}
2506
2507/**
2508 * __gfs2_free_blocks - free a contiguous run of block(s)
2509 * @ip: the inode these blocks are being freed from
2510 * @rgd: the resource group the blocks are in
2511 * @bstart: first block of a run of contiguous blocks
2512 * @blen: the length of the block run
2513 * @meta: 1 if the blocks represent metadata
2514 *
2515 */
2516
2517void __gfs2_free_blocks(struct gfs2_inode *ip, struct gfs2_rgrpd *rgd,
2518 u64 bstart, u32 blen, int meta)
2519{
2520 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2521
2522 rgrp_lock_local(rgd);
2523 rgblk_free(sdp, rgd, bstart, blen, GFS2_BLKST_FREE);
2524 trace_gfs2_block_alloc(ip, rgd, bstart, blen, GFS2_BLKST_FREE);
2525 rgd->rd_free += blen;
2526 rgd->rd_flags &= ~GFS2_RGF_TRIMMED;
2527 gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh);
2528 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
2529 rgrp_unlock_local(rgd);
2530
2531 /* Directories keep their data in the metadata address space */
2532 if (meta || ip->i_depth || gfs2_is_jdata(ip))
2533 gfs2_journal_wipe(ip, bstart, blen);
2534}
2535
2536/**
2537 * gfs2_free_meta - free a contiguous run of data block(s)
2538 * @ip: the inode these blocks are being freed from
2539 * @rgd: the resource group the blocks are in
2540 * @bstart: first block of a run of contiguous blocks
2541 * @blen: the length of the block run
2542 *
2543 */
2544
2545void gfs2_free_meta(struct gfs2_inode *ip, struct gfs2_rgrpd *rgd,
2546 u64 bstart, u32 blen)
2547{
2548 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2549
2550 __gfs2_free_blocks(ip, rgd, bstart, blen, 1);
2551 gfs2_statfs_change(sdp, 0, +blen, 0);
2552 gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid);
2553}
2554
2555void gfs2_unlink_di(struct inode *inode)
2556{
2557 struct gfs2_inode *ip = GFS2_I(inode);
2558 struct gfs2_sbd *sdp = GFS2_SB(inode);
2559 struct gfs2_rgrpd *rgd;
2560 u64 blkno = ip->i_no_addr;
2561
2562 rgd = gfs2_blk2rgrpd(sdp, blkno, true);
2563 if (!rgd)
2564 return;
2565 rgrp_lock_local(rgd);
2566 rgblk_free(sdp, rgd, blkno, 1, GFS2_BLKST_UNLINKED);
2567 trace_gfs2_block_alloc(ip, rgd, blkno, 1, GFS2_BLKST_UNLINKED);
2568 gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh);
2569 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
2570 be32_add_cpu(&rgd->rd_rgl->rl_unlinked, 1);
2571 rgrp_unlock_local(rgd);
2572}
2573
2574void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip)
2575{
2576 struct gfs2_sbd *sdp = rgd->rd_sbd;
2577
2578 rgrp_lock_local(rgd);
2579 rgblk_free(sdp, rgd, ip->i_no_addr, 1, GFS2_BLKST_FREE);
2580 if (!rgd->rd_dinodes)
2581 gfs2_consist_rgrpd(rgd);
2582 rgd->rd_dinodes--;
2583 rgd->rd_free++;
2584
2585 gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh);
2586 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
2587 be32_add_cpu(&rgd->rd_rgl->rl_unlinked, -1);
2588 rgrp_unlock_local(rgd);
2589
2590 gfs2_statfs_change(sdp, 0, +1, -1);
2591 trace_gfs2_block_alloc(ip, rgd, ip->i_no_addr, 1, GFS2_BLKST_FREE);
2592 gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid);
2593 gfs2_journal_wipe(ip, ip->i_no_addr, 1);
2594}
2595
2596/**
2597 * gfs2_check_blk_type - Check the type of a block
2598 * @sdp: The superblock
2599 * @no_addr: The block number to check
2600 * @type: The block type we are looking for
2601 *
2602 * The inode glock of @no_addr must be held. The @type to check for is either
2603 * GFS2_BLKST_DINODE or GFS2_BLKST_UNLINKED; checking for type GFS2_BLKST_FREE
2604 * or GFS2_BLKST_USED would make no sense.
2605 *
2606 * Returns: 0 if the block type matches the expected type
2607 * -ESTALE if it doesn't match
2608 * or -ve errno if something went wrong while checking
2609 */
2610
2611int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type)
2612{
2613 struct gfs2_rgrpd *rgd;
2614 struct gfs2_holder rgd_gh;
2615 struct gfs2_rbm rbm;
2616 int error = -EINVAL;
2617
2618 rgd = gfs2_blk2rgrpd(sdp, no_addr, 1);
2619 if (!rgd)
2620 goto fail;
2621
2622 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh);
2623 if (error)
2624 goto fail;
2625
2626 rbm.rgd = rgd;
2627 error = gfs2_rbm_from_block(&rbm, no_addr);
2628 if (!WARN_ON_ONCE(error)) {
2629 /*
2630 * No need to take the local resource group lock here; the
2631 * inode glock of @no_addr provides the necessary
2632 * synchronization in case the block is an inode. (In case
2633 * the block is not an inode, the block type will not match
2634 * the @type we are looking for.)
2635 */
2636 if (gfs2_testbit(&rbm, false) != type)
2637 error = -ESTALE;
2638 }
2639
2640 gfs2_glock_dq_uninit(&rgd_gh);
2641
2642fail:
2643 return error;
2644}
2645
2646/**
2647 * gfs2_rlist_add - add a RG to a list of RGs
2648 * @ip: the inode
2649 * @rlist: the list of resource groups
2650 * @block: the block
2651 *
2652 * Figure out what RG a block belongs to and add that RG to the list
2653 *
2654 * FIXME: Don't use NOFAIL
2655 *
2656 */
2657
2658void gfs2_rlist_add(struct gfs2_inode *ip, struct gfs2_rgrp_list *rlist,
2659 u64 block)
2660{
2661 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2662 struct gfs2_rgrpd *rgd;
2663 struct gfs2_rgrpd **tmp;
2664 unsigned int new_space;
2665 unsigned int x;
2666
2667 if (gfs2_assert_warn(sdp, !rlist->rl_ghs))
2668 return;
2669
2670 /*
2671 * The resource group last accessed is kept in the last position.
2672 */
2673
2674 if (rlist->rl_rgrps) {
2675 rgd = rlist->rl_rgd[rlist->rl_rgrps - 1];
2676 if (rgrp_contains_block(rgd, block))
2677 return;
2678 rgd = gfs2_blk2rgrpd(sdp, block, 1);
2679 } else {
2680 rgd = ip->i_res.rs_rgd;
2681 if (!rgd || !rgrp_contains_block(rgd, block))
2682 rgd = gfs2_blk2rgrpd(sdp, block, 1);
2683 }
2684
2685 if (!rgd) {
2686 fs_err(sdp, "rlist_add: no rgrp for block %llu\n",
2687 (unsigned long long)block);
2688 return;
2689 }
2690
2691 for (x = 0; x < rlist->rl_rgrps; x++) {
2692 if (rlist->rl_rgd[x] == rgd) {
2693 swap(rlist->rl_rgd[x],
2694 rlist->rl_rgd[rlist->rl_rgrps - 1]);
2695 return;
2696 }
2697 }
2698
2699 if (rlist->rl_rgrps == rlist->rl_space) {
2700 new_space = rlist->rl_space + 10;
2701
2702 tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *),
2703 GFP_NOFS | __GFP_NOFAIL);
2704
2705 if (rlist->rl_rgd) {
2706 memcpy(tmp, rlist->rl_rgd,
2707 rlist->rl_space * sizeof(struct gfs2_rgrpd *));
2708 kfree(rlist->rl_rgd);
2709 }
2710
2711 rlist->rl_space = new_space;
2712 rlist->rl_rgd = tmp;
2713 }
2714
2715 rlist->rl_rgd[rlist->rl_rgrps++] = rgd;
2716}
2717
2718/**
2719 * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate
2720 * and initialize an array of glock holders for them
2721 * @rlist: the list of resource groups
2722 * @state: the state we're requesting
2723 * @flags: the modifier flags
2724 *
2725 * FIXME: Don't use NOFAIL
2726 *
2727 */
2728
2729void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist,
2730 unsigned int state, u16 flags)
2731{
2732 unsigned int x;
2733
2734 rlist->rl_ghs = kmalloc_array(rlist->rl_rgrps,
2735 sizeof(struct gfs2_holder),
2736 GFP_NOFS | __GFP_NOFAIL);
2737 for (x = 0; x < rlist->rl_rgrps; x++)
2738 gfs2_holder_init(rlist->rl_rgd[x]->rd_gl, state, flags,
2739 &rlist->rl_ghs[x]);
2740}
2741
2742/**
2743 * gfs2_rlist_free - free a resource group list
2744 * @rlist: the list of resource groups
2745 *
2746 */
2747
2748void gfs2_rlist_free(struct gfs2_rgrp_list *rlist)
2749{
2750 unsigned int x;
2751
2752 kfree(rlist->rl_rgd);
2753
2754 if (rlist->rl_ghs) {
2755 for (x = 0; x < rlist->rl_rgrps; x++)
2756 gfs2_holder_uninit(&rlist->rl_ghs[x]);
2757 kfree(rlist->rl_ghs);
2758 rlist->rl_ghs = NULL;
2759 }
2760}
2761
2762void rgrp_lock_local(struct gfs2_rgrpd *rgd)
2763{
2764 mutex_lock(&rgd->rd_mutex);
2765}
2766
2767void rgrp_unlock_local(struct gfs2_rgrpd *rgd)
2768{
2769 mutex_unlock(&rgd->rd_mutex);
2770}