xref: /openbmc/linux/fs/xfs/libxfs/xfs_trans_resv.c (revision 2cf1c348)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
4  * Copyright (C) 2010 Red Hat, Inc.
5  * All Rights Reserved.
6  */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_da_format.h"
15 #include "xfs_da_btree.h"
16 #include "xfs_inode.h"
17 #include "xfs_bmap_btree.h"
18 #include "xfs_quota.h"
19 #include "xfs_trans.h"
20 #include "xfs_qm.h"
21 #include "xfs_trans_space.h"
22 
23 #define _ALLOC	true
24 #define _FREE	false
25 
26 /*
27  * A buffer has a format structure overhead in the log in addition
28  * to the data, so we need to take this into account when reserving
29  * space in a transaction for a buffer.  Round the space required up
30  * to a multiple of 128 bytes so that we don't change the historical
31  * reservation that has been used for this overhead.
32  */
33 STATIC uint
34 xfs_buf_log_overhead(void)
35 {
36 	return round_up(sizeof(struct xlog_op_header) +
37 			sizeof(struct xfs_buf_log_format), 128);
38 }
39 
40 /*
41  * Calculate out transaction log reservation per item in bytes.
42  *
43  * The nbufs argument is used to indicate the number of items that
44  * will be changed in a transaction.  size is used to tell how many
45  * bytes should be reserved per item.
46  */
47 STATIC uint
48 xfs_calc_buf_res(
49 	uint		nbufs,
50 	uint		size)
51 {
52 	return nbufs * (size + xfs_buf_log_overhead());
53 }
54 
55 /*
56  * Per-extent log reservation for the btree changes involved in freeing or
57  * allocating an extent.  In classic XFS there were two trees that will be
58  * modified (bnobt + cntbt).  With rmap enabled, there are three trees
59  * (rmapbt).  With reflink, there are four trees (refcountbt).  The number of
60  * blocks reserved is based on the formula:
61  *
62  * num trees * ((2 blocks/level * max depth) - 1)
63  *
64  * Keep in mind that max depth is calculated separately for each type of tree.
65  */
66 uint
67 xfs_allocfree_log_count(
68 	struct xfs_mount *mp,
69 	uint		num_ops)
70 {
71 	uint		blocks;
72 
73 	blocks = num_ops * 2 * (2 * mp->m_alloc_maxlevels - 1);
74 	if (xfs_has_rmapbt(mp))
75 		blocks += num_ops * (2 * mp->m_rmap_maxlevels - 1);
76 	if (xfs_has_reflink(mp))
77 		blocks += num_ops * (2 * mp->m_refc_maxlevels - 1);
78 
79 	return blocks;
80 }
81 
82 /*
83  * Logging inodes is really tricksy. They are logged in memory format,
84  * which means that what we write into the log doesn't directly translate into
85  * the amount of space they use on disk.
86  *
87  * Case in point - btree format forks in memory format use more space than the
88  * on-disk format. In memory, the buffer contains a normal btree block header so
89  * the btree code can treat it as though it is just another generic buffer.
90  * However, when we write it to the inode fork, we don't write all of this
91  * header as it isn't needed. e.g. the root is only ever in the inode, so
92  * there's no need for sibling pointers which would waste 16 bytes of space.
93  *
94  * Hence when we have an inode with a maximally sized btree format fork, then
95  * amount of information we actually log is greater than the size of the inode
96  * on disk. Hence we need an inode reservation function that calculates all this
97  * correctly. So, we log:
98  *
99  * - 4 log op headers for object
100  *	- for the ilf, the inode core and 2 forks
101  * - inode log format object
102  * - the inode core
103  * - two inode forks containing bmap btree root blocks.
104  *	- the btree data contained by both forks will fit into the inode size,
105  *	  hence when combined with the inode core above, we have a total of the
106  *	  actual inode size.
107  *	- the BMBT headers need to be accounted separately, as they are
108  *	  additional to the records and pointers that fit inside the inode
109  *	  forks.
110  */
111 STATIC uint
112 xfs_calc_inode_res(
113 	struct xfs_mount	*mp,
114 	uint			ninodes)
115 {
116 	return ninodes *
117 		(4 * sizeof(struct xlog_op_header) +
118 		 sizeof(struct xfs_inode_log_format) +
119 		 mp->m_sb.sb_inodesize +
120 		 2 * XFS_BMBT_BLOCK_LEN(mp));
121 }
122 
123 /*
124  * Inode btree record insertion/removal modifies the inode btree and free space
125  * btrees (since the inobt does not use the agfl). This requires the following
126  * reservation:
127  *
128  * the inode btree: max depth * blocksize
129  * the allocation btrees: 2 trees * (max depth - 1) * block size
130  *
131  * The caller must account for SB and AG header modifications, etc.
132  */
133 STATIC uint
134 xfs_calc_inobt_res(
135 	struct xfs_mount	*mp)
136 {
137 	return xfs_calc_buf_res(M_IGEO(mp)->inobt_maxlevels,
138 			XFS_FSB_TO_B(mp, 1)) +
139 				xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1),
140 			XFS_FSB_TO_B(mp, 1));
141 }
142 
143 /*
144  * The free inode btree is a conditional feature. The behavior differs slightly
145  * from that of the traditional inode btree in that the finobt tracks records
146  * for inode chunks with at least one free inode. A record can be removed from
147  * the tree during individual inode allocation. Therefore the finobt
148  * reservation is unconditional for both the inode chunk allocation and
149  * individual inode allocation (modify) cases.
150  *
151  * Behavior aside, the reservation for finobt modification is equivalent to the
152  * traditional inobt: cover a full finobt shape change plus block allocation.
153  */
154 STATIC uint
155 xfs_calc_finobt_res(
156 	struct xfs_mount	*mp)
157 {
158 	if (!xfs_has_finobt(mp))
159 		return 0;
160 
161 	return xfs_calc_inobt_res(mp);
162 }
163 
164 /*
165  * Calculate the reservation required to allocate or free an inode chunk. This
166  * includes:
167  *
168  * the allocation btrees: 2 trees * (max depth - 1) * block size
169  * the inode chunk: m_ino_geo.ialloc_blks * N
170  *
171  * The size N of the inode chunk reservation depends on whether it is for
172  * allocation or free and which type of create transaction is in use. An inode
173  * chunk free always invalidates the buffers and only requires reservation for
174  * headers (N == 0). An inode chunk allocation requires a chunk sized
175  * reservation on v4 and older superblocks to initialize the chunk. No chunk
176  * reservation is required for allocation on v5 supers, which use ordered
177  * buffers to initialize.
178  */
179 STATIC uint
180 xfs_calc_inode_chunk_res(
181 	struct xfs_mount	*mp,
182 	bool			alloc)
183 {
184 	uint			res, size = 0;
185 
186 	res = xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1),
187 			       XFS_FSB_TO_B(mp, 1));
188 	if (alloc) {
189 		/* icreate tx uses ordered buffers */
190 		if (xfs_has_v3inodes(mp))
191 			return res;
192 		size = XFS_FSB_TO_B(mp, 1);
193 	}
194 
195 	res += xfs_calc_buf_res(M_IGEO(mp)->ialloc_blks, size);
196 	return res;
197 }
198 
199 /*
200  * Per-extent log reservation for the btree changes involved in freeing or
201  * allocating a realtime extent.  We have to be able to log as many rtbitmap
202  * blocks as needed to mark inuse MAXEXTLEN blocks' worth of realtime extents,
203  * as well as the realtime summary block.
204  */
205 static unsigned int
206 xfs_rtalloc_log_count(
207 	struct xfs_mount	*mp,
208 	unsigned int		num_ops)
209 {
210 	unsigned int		blksz = XFS_FSB_TO_B(mp, 1);
211 	unsigned int		rtbmp_bytes;
212 
213 	rtbmp_bytes = (MAXEXTLEN / mp->m_sb.sb_rextsize) / NBBY;
214 	return (howmany(rtbmp_bytes, blksz) + 1) * num_ops;
215 }
216 
217 /*
218  * Various log reservation values.
219  *
220  * These are based on the size of the file system block because that is what
221  * most transactions manipulate.  Each adds in an additional 128 bytes per
222  * item logged to try to account for the overhead of the transaction mechanism.
223  *
224  * Note:  Most of the reservations underestimate the number of allocation
225  * groups into which they could free extents in the xfs_defer_finish() call.
226  * This is because the number in the worst case is quite high and quite
227  * unusual.  In order to fix this we need to change xfs_defer_finish() to free
228  * extents in only a single AG at a time.  This will require changes to the
229  * EFI code as well, however, so that the EFI for the extents not freed is
230  * logged again in each transaction.  See SGI PV #261917.
231  *
232  * Reservation functions here avoid a huge stack in xfs_trans_init due to
233  * register overflow from temporaries in the calculations.
234  */
235 
236 
237 /*
238  * In a write transaction we can allocate a maximum of 2
239  * extents.  This gives (t1):
240  *    the inode getting the new extents: inode size
241  *    the inode's bmap btree: max depth * block size
242  *    the agfs of the ags from which the extents are allocated: 2 * sector
243  *    the superblock free block counter: sector size
244  *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
245  * Or, if we're writing to a realtime file (t2):
246  *    the inode getting the new extents: inode size
247  *    the inode's bmap btree: max depth * block size
248  *    the agfs of the ags from which the extents are allocated: 2 * sector
249  *    the superblock free block counter: sector size
250  *    the realtime bitmap: ((MAXEXTLEN / rtextsize) / NBBY) bytes
251  *    the realtime summary: 1 block
252  *    the allocation btrees: 2 trees * (2 * max depth - 1) * block size
253  * And the bmap_finish transaction can free bmap blocks in a join (t3):
254  *    the agfs of the ags containing the blocks: 2 * sector size
255  *    the agfls of the ags containing the blocks: 2 * sector size
256  *    the super block free block counter: sector size
257  *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
258  */
259 STATIC uint
260 xfs_calc_write_reservation(
261 	struct xfs_mount	*mp)
262 {
263 	unsigned int		t1, t2, t3;
264 	unsigned int		blksz = XFS_FSB_TO_B(mp, 1);
265 
266 	t1 = xfs_calc_inode_res(mp, 1) +
267 	     xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK), blksz) +
268 	     xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) +
269 	     xfs_calc_buf_res(xfs_allocfree_log_count(mp, 2), blksz);
270 
271 	if (xfs_has_realtime(mp)) {
272 		t2 = xfs_calc_inode_res(mp, 1) +
273 		     xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK),
274 				     blksz) +
275 		     xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) +
276 		     xfs_calc_buf_res(xfs_rtalloc_log_count(mp, 1), blksz) +
277 		     xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1), blksz);
278 	} else {
279 		t2 = 0;
280 	}
281 
282 	t3 = xfs_calc_buf_res(5, mp->m_sb.sb_sectsize) +
283 	     xfs_calc_buf_res(xfs_allocfree_log_count(mp, 2), blksz);
284 
285 	return XFS_DQUOT_LOGRES(mp) + max3(t1, t2, t3);
286 }
287 
288 /*
289  * In truncating a file we free up to two extents at once.  We can modify (t1):
290  *    the inode being truncated: inode size
291  *    the inode's bmap btree: (max depth + 1) * block size
292  * And the bmap_finish transaction can free the blocks and bmap blocks (t2):
293  *    the agf for each of the ags: 4 * sector size
294  *    the agfl for each of the ags: 4 * sector size
295  *    the super block to reflect the freed blocks: sector size
296  *    worst case split in allocation btrees per extent assuming 4 extents:
297  *		4 exts * 2 trees * (2 * max depth - 1) * block size
298  * Or, if it's a realtime file (t3):
299  *    the agf for each of the ags: 2 * sector size
300  *    the agfl for each of the ags: 2 * sector size
301  *    the super block to reflect the freed blocks: sector size
302  *    the realtime bitmap: 2 exts * ((MAXEXTLEN / rtextsize) / NBBY) bytes
303  *    the realtime summary: 2 exts * 1 block
304  *    worst case split in allocation btrees per extent assuming 2 extents:
305  *		2 exts * 2 trees * (2 * max depth - 1) * block size
306  */
307 STATIC uint
308 xfs_calc_itruncate_reservation(
309 	struct xfs_mount	*mp)
310 {
311 	unsigned int		t1, t2, t3;
312 	unsigned int		blksz = XFS_FSB_TO_B(mp, 1);
313 
314 	t1 = xfs_calc_inode_res(mp, 1) +
315 	     xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) + 1, blksz);
316 
317 	t2 = xfs_calc_buf_res(9, mp->m_sb.sb_sectsize) +
318 	     xfs_calc_buf_res(xfs_allocfree_log_count(mp, 4), blksz);
319 
320 	if (xfs_has_realtime(mp)) {
321 		t3 = xfs_calc_buf_res(5, mp->m_sb.sb_sectsize) +
322 		     xfs_calc_buf_res(xfs_rtalloc_log_count(mp, 2), blksz) +
323 		     xfs_calc_buf_res(xfs_allocfree_log_count(mp, 2), blksz);
324 	} else {
325 		t3 = 0;
326 	}
327 
328 	return XFS_DQUOT_LOGRES(mp) + max3(t1, t2, t3);
329 }
330 
331 /*
332  * In renaming a files we can modify:
333  *    the four inodes involved: 4 * inode size
334  *    the two directory btrees: 2 * (max depth + v2) * dir block size
335  *    the two directory bmap btrees: 2 * max depth * block size
336  * And the bmap_finish transaction can free dir and bmap blocks (two sets
337  *	of bmap blocks) giving:
338  *    the agf for the ags in which the blocks live: 3 * sector size
339  *    the agfl for the ags in which the blocks live: 3 * sector size
340  *    the superblock for the free block count: sector size
341  *    the allocation btrees: 3 exts * 2 trees * (2 * max depth - 1) * block size
342  */
343 STATIC uint
344 xfs_calc_rename_reservation(
345 	struct xfs_mount	*mp)
346 {
347 	return XFS_DQUOT_LOGRES(mp) +
348 		max((xfs_calc_inode_res(mp, 4) +
349 		     xfs_calc_buf_res(2 * XFS_DIROP_LOG_COUNT(mp),
350 				      XFS_FSB_TO_B(mp, 1))),
351 		    (xfs_calc_buf_res(7, mp->m_sb.sb_sectsize) +
352 		     xfs_calc_buf_res(xfs_allocfree_log_count(mp, 3),
353 				      XFS_FSB_TO_B(mp, 1))));
354 }
355 
356 /*
357  * For removing an inode from unlinked list at first, we can modify:
358  *    the agi hash list and counters: sector size
359  *    the on disk inode before ours in the agi hash list: inode cluster size
360  *    the on disk inode in the agi hash list: inode cluster size
361  */
362 STATIC uint
363 xfs_calc_iunlink_remove_reservation(
364 	struct xfs_mount        *mp)
365 {
366 	return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
367 	       2 * M_IGEO(mp)->inode_cluster_size;
368 }
369 
370 /*
371  * For creating a link to an inode:
372  *    the parent directory inode: inode size
373  *    the linked inode: inode size
374  *    the directory btree could split: (max depth + v2) * dir block size
375  *    the directory bmap btree could join or split: (max depth + v2) * blocksize
376  * And the bmap_finish transaction can free some bmap blocks giving:
377  *    the agf for the ag in which the blocks live: sector size
378  *    the agfl for the ag in which the blocks live: sector size
379  *    the superblock for the free block count: sector size
380  *    the allocation btrees: 2 trees * (2 * max depth - 1) * block size
381  */
382 STATIC uint
383 xfs_calc_link_reservation(
384 	struct xfs_mount	*mp)
385 {
386 	return XFS_DQUOT_LOGRES(mp) +
387 		xfs_calc_iunlink_remove_reservation(mp) +
388 		max((xfs_calc_inode_res(mp, 2) +
389 		     xfs_calc_buf_res(XFS_DIROP_LOG_COUNT(mp),
390 				      XFS_FSB_TO_B(mp, 1))),
391 		    (xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) +
392 		     xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1),
393 				      XFS_FSB_TO_B(mp, 1))));
394 }
395 
396 /*
397  * For adding an inode to unlinked list we can modify:
398  *    the agi hash list: sector size
399  *    the on disk inode: inode cluster size
400  */
401 STATIC uint
402 xfs_calc_iunlink_add_reservation(xfs_mount_t *mp)
403 {
404 	return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
405 			M_IGEO(mp)->inode_cluster_size;
406 }
407 
408 /*
409  * For removing a directory entry we can modify:
410  *    the parent directory inode: inode size
411  *    the removed inode: inode size
412  *    the directory btree could join: (max depth + v2) * dir block size
413  *    the directory bmap btree could join or split: (max depth + v2) * blocksize
414  * And the bmap_finish transaction can free the dir and bmap blocks giving:
415  *    the agf for the ag in which the blocks live: 2 * sector size
416  *    the agfl for the ag in which the blocks live: 2 * sector size
417  *    the superblock for the free block count: sector size
418  *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
419  */
420 STATIC uint
421 xfs_calc_remove_reservation(
422 	struct xfs_mount	*mp)
423 {
424 	return XFS_DQUOT_LOGRES(mp) +
425 		xfs_calc_iunlink_add_reservation(mp) +
426 		max((xfs_calc_inode_res(mp, 1) +
427 		     xfs_calc_buf_res(XFS_DIROP_LOG_COUNT(mp),
428 				      XFS_FSB_TO_B(mp, 1))),
429 		    (xfs_calc_buf_res(4, mp->m_sb.sb_sectsize) +
430 		     xfs_calc_buf_res(xfs_allocfree_log_count(mp, 2),
431 				      XFS_FSB_TO_B(mp, 1))));
432 }
433 
434 /*
435  * For create, break it in to the two cases that the transaction
436  * covers. We start with the modify case - allocation done by modification
437  * of the state of existing inodes - and the allocation case.
438  */
439 
440 /*
441  * For create we can modify:
442  *    the parent directory inode: inode size
443  *    the new inode: inode size
444  *    the inode btree entry: block size
445  *    the superblock for the nlink flag: sector size
446  *    the directory btree: (max depth + v2) * dir block size
447  *    the directory inode's bmap btree: (max depth + v2) * block size
448  *    the finobt (record modification and allocation btrees)
449  */
450 STATIC uint
451 xfs_calc_create_resv_modify(
452 	struct xfs_mount	*mp)
453 {
454 	return xfs_calc_inode_res(mp, 2) +
455 		xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
456 		(uint)XFS_FSB_TO_B(mp, 1) +
457 		xfs_calc_buf_res(XFS_DIROP_LOG_COUNT(mp), XFS_FSB_TO_B(mp, 1)) +
458 		xfs_calc_finobt_res(mp);
459 }
460 
461 /*
462  * For icreate we can allocate some inodes giving:
463  *    the agi and agf of the ag getting the new inodes: 2 * sectorsize
464  *    the superblock for the nlink flag: sector size
465  *    the inode chunk (allocation, optional init)
466  *    the inobt (record insertion)
467  *    the finobt (optional, record insertion)
468  */
469 STATIC uint
470 xfs_calc_icreate_resv_alloc(
471 	struct xfs_mount	*mp)
472 {
473 	return xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) +
474 		mp->m_sb.sb_sectsize +
475 		xfs_calc_inode_chunk_res(mp, _ALLOC) +
476 		xfs_calc_inobt_res(mp) +
477 		xfs_calc_finobt_res(mp);
478 }
479 
480 STATIC uint
481 xfs_calc_icreate_reservation(xfs_mount_t *mp)
482 {
483 	return XFS_DQUOT_LOGRES(mp) +
484 		max(xfs_calc_icreate_resv_alloc(mp),
485 		    xfs_calc_create_resv_modify(mp));
486 }
487 
488 STATIC uint
489 xfs_calc_create_tmpfile_reservation(
490 	struct xfs_mount        *mp)
491 {
492 	uint	res = XFS_DQUOT_LOGRES(mp);
493 
494 	res += xfs_calc_icreate_resv_alloc(mp);
495 	return res + xfs_calc_iunlink_add_reservation(mp);
496 }
497 
498 /*
499  * Making a new directory is the same as creating a new file.
500  */
501 STATIC uint
502 xfs_calc_mkdir_reservation(
503 	struct xfs_mount	*mp)
504 {
505 	return xfs_calc_icreate_reservation(mp);
506 }
507 
508 
509 /*
510  * Making a new symplink is the same as creating a new file, but
511  * with the added blocks for remote symlink data which can be up to 1kB in
512  * length (XFS_SYMLINK_MAXLEN).
513  */
514 STATIC uint
515 xfs_calc_symlink_reservation(
516 	struct xfs_mount	*mp)
517 {
518 	return xfs_calc_icreate_reservation(mp) +
519 	       xfs_calc_buf_res(1, XFS_SYMLINK_MAXLEN);
520 }
521 
522 /*
523  * In freeing an inode we can modify:
524  *    the inode being freed: inode size
525  *    the super block free inode counter, AGF and AGFL: sector size
526  *    the on disk inode (agi unlinked list removal)
527  *    the inode chunk (invalidated, headers only)
528  *    the inode btree
529  *    the finobt (record insertion, removal or modification)
530  *
531  * Note that the inode chunk res. includes an allocfree res. for freeing of the
532  * inode chunk. This is technically extraneous because the inode chunk free is
533  * deferred (it occurs after a transaction roll). Include the extra reservation
534  * anyways since we've had reports of ifree transaction overruns due to too many
535  * agfl fixups during inode chunk frees.
536  */
537 STATIC uint
538 xfs_calc_ifree_reservation(
539 	struct xfs_mount	*mp)
540 {
541 	return XFS_DQUOT_LOGRES(mp) +
542 		xfs_calc_inode_res(mp, 1) +
543 		xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) +
544 		xfs_calc_iunlink_remove_reservation(mp) +
545 		xfs_calc_inode_chunk_res(mp, _FREE) +
546 		xfs_calc_inobt_res(mp) +
547 		xfs_calc_finobt_res(mp);
548 }
549 
550 /*
551  * When only changing the inode we log the inode and possibly the superblock
552  * We also add a bit of slop for the transaction stuff.
553  */
554 STATIC uint
555 xfs_calc_ichange_reservation(
556 	struct xfs_mount	*mp)
557 {
558 	return XFS_DQUOT_LOGRES(mp) +
559 		xfs_calc_inode_res(mp, 1) +
560 		xfs_calc_buf_res(1, mp->m_sb.sb_sectsize);
561 
562 }
563 
564 /*
565  * Growing the data section of the filesystem.
566  *	superblock
567  *	agi and agf
568  *	allocation btrees
569  */
570 STATIC uint
571 xfs_calc_growdata_reservation(
572 	struct xfs_mount	*mp)
573 {
574 	return xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) +
575 		xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1),
576 				 XFS_FSB_TO_B(mp, 1));
577 }
578 
579 /*
580  * Growing the rt section of the filesystem.
581  * In the first set of transactions (ALLOC) we allocate space to the
582  * bitmap or summary files.
583  *	superblock: sector size
584  *	agf of the ag from which the extent is allocated: sector size
585  *	bmap btree for bitmap/summary inode: max depth * blocksize
586  *	bitmap/summary inode: inode size
587  *	allocation btrees for 1 block alloc: 2 * (2 * maxdepth - 1) * blocksize
588  */
589 STATIC uint
590 xfs_calc_growrtalloc_reservation(
591 	struct xfs_mount	*mp)
592 {
593 	return xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) +
594 		xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK),
595 				 XFS_FSB_TO_B(mp, 1)) +
596 		xfs_calc_inode_res(mp, 1) +
597 		xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1),
598 				 XFS_FSB_TO_B(mp, 1));
599 }
600 
601 /*
602  * Growing the rt section of the filesystem.
603  * In the second set of transactions (ZERO) we zero the new metadata blocks.
604  *	one bitmap/summary block: blocksize
605  */
606 STATIC uint
607 xfs_calc_growrtzero_reservation(
608 	struct xfs_mount	*mp)
609 {
610 	return xfs_calc_buf_res(1, mp->m_sb.sb_blocksize);
611 }
612 
613 /*
614  * Growing the rt section of the filesystem.
615  * In the third set of transactions (FREE) we update metadata without
616  * allocating any new blocks.
617  *	superblock: sector size
618  *	bitmap inode: inode size
619  *	summary inode: inode size
620  *	one bitmap block: blocksize
621  *	summary blocks: new summary size
622  */
623 STATIC uint
624 xfs_calc_growrtfree_reservation(
625 	struct xfs_mount	*mp)
626 {
627 	return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
628 		xfs_calc_inode_res(mp, 2) +
629 		xfs_calc_buf_res(1, mp->m_sb.sb_blocksize) +
630 		xfs_calc_buf_res(1, mp->m_rsumsize);
631 }
632 
633 /*
634  * Logging the inode modification timestamp on a synchronous write.
635  *	inode
636  */
637 STATIC uint
638 xfs_calc_swrite_reservation(
639 	struct xfs_mount	*mp)
640 {
641 	return xfs_calc_inode_res(mp, 1);
642 }
643 
644 /*
645  * Logging the inode mode bits when writing a setuid/setgid file
646  *	inode
647  */
648 STATIC uint
649 xfs_calc_writeid_reservation(
650 	struct xfs_mount	*mp)
651 {
652 	return xfs_calc_inode_res(mp, 1);
653 }
654 
655 /*
656  * Converting the inode from non-attributed to attributed.
657  *	the inode being converted: inode size
658  *	agf block and superblock (for block allocation)
659  *	the new block (directory sized)
660  *	bmap blocks for the new directory block
661  *	allocation btrees
662  */
663 STATIC uint
664 xfs_calc_addafork_reservation(
665 	struct xfs_mount	*mp)
666 {
667 	return XFS_DQUOT_LOGRES(mp) +
668 		xfs_calc_inode_res(mp, 1) +
669 		xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) +
670 		xfs_calc_buf_res(1, mp->m_dir_geo->blksize) +
671 		xfs_calc_buf_res(XFS_DAENTER_BMAP1B(mp, XFS_DATA_FORK) + 1,
672 				 XFS_FSB_TO_B(mp, 1)) +
673 		xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1),
674 				 XFS_FSB_TO_B(mp, 1));
675 }
676 
677 /*
678  * Removing the attribute fork of a file
679  *    the inode being truncated: inode size
680  *    the inode's bmap btree: max depth * block size
681  * And the bmap_finish transaction can free the blocks and bmap blocks:
682  *    the agf for each of the ags: 4 * sector size
683  *    the agfl for each of the ags: 4 * sector size
684  *    the super block to reflect the freed blocks: sector size
685  *    worst case split in allocation btrees per extent assuming 4 extents:
686  *		4 exts * 2 trees * (2 * max depth - 1) * block size
687  */
688 STATIC uint
689 xfs_calc_attrinval_reservation(
690 	struct xfs_mount	*mp)
691 {
692 	return max((xfs_calc_inode_res(mp, 1) +
693 		    xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK),
694 				     XFS_FSB_TO_B(mp, 1))),
695 		   (xfs_calc_buf_res(9, mp->m_sb.sb_sectsize) +
696 		    xfs_calc_buf_res(xfs_allocfree_log_count(mp, 4),
697 				     XFS_FSB_TO_B(mp, 1))));
698 }
699 
700 /*
701  * Setting an attribute at mount time.
702  *	the inode getting the attribute
703  *	the superblock for allocations
704  *	the agfs extents are allocated from
705  *	the attribute btree * max depth
706  *	the inode allocation btree
707  * Since attribute transaction space is dependent on the size of the attribute,
708  * the calculation is done partially at mount time and partially at runtime(see
709  * below).
710  */
711 STATIC uint
712 xfs_calc_attrsetm_reservation(
713 	struct xfs_mount	*mp)
714 {
715 	return XFS_DQUOT_LOGRES(mp) +
716 		xfs_calc_inode_res(mp, 1) +
717 		xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
718 		xfs_calc_buf_res(XFS_DA_NODE_MAXDEPTH, XFS_FSB_TO_B(mp, 1));
719 }
720 
721 /*
722  * Setting an attribute at runtime, transaction space unit per block.
723  * 	the superblock for allocations: sector size
724  *	the inode bmap btree could join or split: max depth * block size
725  * Since the runtime attribute transaction space is dependent on the total
726  * blocks needed for the 1st bmap, here we calculate out the space unit for
727  * one block so that the caller could figure out the total space according
728  * to the attibute extent length in blocks by:
729  *	ext * M_RES(mp)->tr_attrsetrt.tr_logres
730  */
731 STATIC uint
732 xfs_calc_attrsetrt_reservation(
733 	struct xfs_mount	*mp)
734 {
735 	return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
736 		xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK),
737 				 XFS_FSB_TO_B(mp, 1));
738 }
739 
740 /*
741  * Removing an attribute.
742  *    the inode: inode size
743  *    the attribute btree could join: max depth * block size
744  *    the inode bmap btree could join or split: max depth * block size
745  * And the bmap_finish transaction can free the attr blocks freed giving:
746  *    the agf for the ag in which the blocks live: 2 * sector size
747  *    the agfl for the ag in which the blocks live: 2 * sector size
748  *    the superblock for the free block count: sector size
749  *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
750  */
751 STATIC uint
752 xfs_calc_attrrm_reservation(
753 	struct xfs_mount	*mp)
754 {
755 	return XFS_DQUOT_LOGRES(mp) +
756 		max((xfs_calc_inode_res(mp, 1) +
757 		     xfs_calc_buf_res(XFS_DA_NODE_MAXDEPTH,
758 				      XFS_FSB_TO_B(mp, 1)) +
759 		     (uint)XFS_FSB_TO_B(mp,
760 					XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK)) +
761 		     xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK), 0)),
762 		    (xfs_calc_buf_res(5, mp->m_sb.sb_sectsize) +
763 		     xfs_calc_buf_res(xfs_allocfree_log_count(mp, 2),
764 				      XFS_FSB_TO_B(mp, 1))));
765 }
766 
767 /*
768  * Clearing a bad agino number in an agi hash bucket.
769  */
770 STATIC uint
771 xfs_calc_clear_agi_bucket_reservation(
772 	struct xfs_mount	*mp)
773 {
774 	return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize);
775 }
776 
777 /*
778  * Adjusting quota limits.
779  *    the disk quota buffer: sizeof(struct xfs_disk_dquot)
780  */
781 STATIC uint
782 xfs_calc_qm_setqlim_reservation(void)
783 {
784 	return xfs_calc_buf_res(1, sizeof(struct xfs_disk_dquot));
785 }
786 
787 /*
788  * Allocating quota on disk if needed.
789  *	the write transaction log space for quota file extent allocation
790  *	the unit of quota allocation: one system block size
791  */
792 STATIC uint
793 xfs_calc_qm_dqalloc_reservation(
794 	struct xfs_mount	*mp)
795 {
796 	return xfs_calc_write_reservation(mp) +
797 		xfs_calc_buf_res(1,
798 			XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB) - 1);
799 }
800 
801 /*
802  * Syncing the incore super block changes to disk.
803  *     the super block to reflect the changes: sector size
804  */
805 STATIC uint
806 xfs_calc_sb_reservation(
807 	struct xfs_mount	*mp)
808 {
809 	return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize);
810 }
811 
812 void
813 xfs_trans_resv_calc(
814 	struct xfs_mount	*mp,
815 	struct xfs_trans_resv	*resp)
816 {
817 	unsigned int		rmap_maxlevels = mp->m_rmap_maxlevels;
818 
819 	/*
820 	 * In the early days of rmap+reflink, we always set the rmap maxlevels
821 	 * to 9 even if the AG was small enough that it would never grow to
822 	 * that height.  Transaction reservation sizes influence the minimum
823 	 * log size calculation, which influences the size of the log that mkfs
824 	 * creates.  Use the old value here to ensure that newly formatted
825 	 * small filesystems will mount on older kernels.
826 	 */
827 	if (xfs_has_rmapbt(mp) && xfs_has_reflink(mp))
828 		mp->m_rmap_maxlevels = XFS_OLD_REFLINK_RMAP_MAXLEVELS;
829 
830 	/*
831 	 * The following transactions are logged in physical format and
832 	 * require a permanent reservation on space.
833 	 */
834 	resp->tr_write.tr_logres = xfs_calc_write_reservation(mp);
835 	if (xfs_has_reflink(mp))
836 		resp->tr_write.tr_logcount = XFS_WRITE_LOG_COUNT_REFLINK;
837 	else
838 		resp->tr_write.tr_logcount = XFS_WRITE_LOG_COUNT;
839 	resp->tr_write.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
840 
841 	resp->tr_itruncate.tr_logres = xfs_calc_itruncate_reservation(mp);
842 	if (xfs_has_reflink(mp))
843 		resp->tr_itruncate.tr_logcount =
844 				XFS_ITRUNCATE_LOG_COUNT_REFLINK;
845 	else
846 		resp->tr_itruncate.tr_logcount = XFS_ITRUNCATE_LOG_COUNT;
847 	resp->tr_itruncate.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
848 
849 	resp->tr_rename.tr_logres = xfs_calc_rename_reservation(mp);
850 	resp->tr_rename.tr_logcount = XFS_RENAME_LOG_COUNT;
851 	resp->tr_rename.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
852 
853 	resp->tr_link.tr_logres = xfs_calc_link_reservation(mp);
854 	resp->tr_link.tr_logcount = XFS_LINK_LOG_COUNT;
855 	resp->tr_link.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
856 
857 	resp->tr_remove.tr_logres = xfs_calc_remove_reservation(mp);
858 	resp->tr_remove.tr_logcount = XFS_REMOVE_LOG_COUNT;
859 	resp->tr_remove.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
860 
861 	resp->tr_symlink.tr_logres = xfs_calc_symlink_reservation(mp);
862 	resp->tr_symlink.tr_logcount = XFS_SYMLINK_LOG_COUNT;
863 	resp->tr_symlink.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
864 
865 	resp->tr_create.tr_logres = xfs_calc_icreate_reservation(mp);
866 	resp->tr_create.tr_logcount = XFS_CREATE_LOG_COUNT;
867 	resp->tr_create.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
868 
869 	resp->tr_create_tmpfile.tr_logres =
870 			xfs_calc_create_tmpfile_reservation(mp);
871 	resp->tr_create_tmpfile.tr_logcount = XFS_CREATE_TMPFILE_LOG_COUNT;
872 	resp->tr_create_tmpfile.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
873 
874 	resp->tr_mkdir.tr_logres = xfs_calc_mkdir_reservation(mp);
875 	resp->tr_mkdir.tr_logcount = XFS_MKDIR_LOG_COUNT;
876 	resp->tr_mkdir.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
877 
878 	resp->tr_ifree.tr_logres = xfs_calc_ifree_reservation(mp);
879 	resp->tr_ifree.tr_logcount = XFS_INACTIVE_LOG_COUNT;
880 	resp->tr_ifree.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
881 
882 	resp->tr_addafork.tr_logres = xfs_calc_addafork_reservation(mp);
883 	resp->tr_addafork.tr_logcount = XFS_ADDAFORK_LOG_COUNT;
884 	resp->tr_addafork.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
885 
886 	resp->tr_attrinval.tr_logres = xfs_calc_attrinval_reservation(mp);
887 	resp->tr_attrinval.tr_logcount = XFS_ATTRINVAL_LOG_COUNT;
888 	resp->tr_attrinval.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
889 
890 	resp->tr_attrsetm.tr_logres = xfs_calc_attrsetm_reservation(mp);
891 	resp->tr_attrsetm.tr_logcount = XFS_ATTRSET_LOG_COUNT;
892 	resp->tr_attrsetm.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
893 
894 	resp->tr_attrrm.tr_logres = xfs_calc_attrrm_reservation(mp);
895 	resp->tr_attrrm.tr_logcount = XFS_ATTRRM_LOG_COUNT;
896 	resp->tr_attrrm.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
897 
898 	resp->tr_growrtalloc.tr_logres = xfs_calc_growrtalloc_reservation(mp);
899 	resp->tr_growrtalloc.tr_logcount = XFS_DEFAULT_PERM_LOG_COUNT;
900 	resp->tr_growrtalloc.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
901 
902 	resp->tr_qm_dqalloc.tr_logres = xfs_calc_qm_dqalloc_reservation(mp);
903 	if (xfs_has_reflink(mp))
904 		resp->tr_qm_dqalloc.tr_logcount = XFS_WRITE_LOG_COUNT_REFLINK;
905 	else
906 		resp->tr_qm_dqalloc.tr_logcount = XFS_WRITE_LOG_COUNT;
907 	resp->tr_qm_dqalloc.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
908 
909 	/*
910 	 * The following transactions are logged in logical format with
911 	 * a default log count.
912 	 */
913 	resp->tr_qm_setqlim.tr_logres = xfs_calc_qm_setqlim_reservation();
914 	resp->tr_qm_setqlim.tr_logcount = XFS_DEFAULT_LOG_COUNT;
915 
916 	resp->tr_sb.tr_logres = xfs_calc_sb_reservation(mp);
917 	resp->tr_sb.tr_logcount = XFS_DEFAULT_LOG_COUNT;
918 
919 	/* growdata requires permanent res; it can free space to the last AG */
920 	resp->tr_growdata.tr_logres = xfs_calc_growdata_reservation(mp);
921 	resp->tr_growdata.tr_logcount = XFS_DEFAULT_PERM_LOG_COUNT;
922 	resp->tr_growdata.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
923 
924 	/* The following transaction are logged in logical format */
925 	resp->tr_ichange.tr_logres = xfs_calc_ichange_reservation(mp);
926 	resp->tr_fsyncts.tr_logres = xfs_calc_swrite_reservation(mp);
927 	resp->tr_writeid.tr_logres = xfs_calc_writeid_reservation(mp);
928 	resp->tr_attrsetrt.tr_logres = xfs_calc_attrsetrt_reservation(mp);
929 	resp->tr_clearagi.tr_logres = xfs_calc_clear_agi_bucket_reservation(mp);
930 	resp->tr_growrtzero.tr_logres = xfs_calc_growrtzero_reservation(mp);
931 	resp->tr_growrtfree.tr_logres = xfs_calc_growrtfree_reservation(mp);
932 
933 	/* Put everything back the way it was.  This goes at the end. */
934 	mp->m_rmap_maxlevels = rmap_maxlevels;
935 }
936