xref: /linux/fs/ext4/mballoc.c (revision 3a39d672e7f48b8d6b91a09afa4b55352773b4b5)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
4  * Written by Alex Tomas <alex@clusterfs.com>
5  */
6 
7 
8 /*
9  * mballoc.c contains the multiblocks allocation routines
10  */
11 
12 #include "ext4_jbd2.h"
13 #include "mballoc.h"
14 #include <linux/log2.h>
15 #include <linux/module.h>
16 #include <linux/slab.h>
17 #include <linux/nospec.h>
18 #include <linux/backing-dev.h>
19 #include <linux/freezer.h>
20 #include <trace/events/ext4.h>
21 #include <kunit/static_stub.h>
22 
23 /*
24  * MUSTDO:
25  *   - test ext4_ext_search_left() and ext4_ext_search_right()
26  *   - search for metadata in few groups
27  *
28  * TODO v4:
29  *   - normalization should take into account whether file is still open
30  *   - discard preallocations if no free space left (policy?)
31  *   - don't normalize tails
32  *   - quota
33  *   - reservation for superuser
34  *
35  * TODO v3:
36  *   - bitmap read-ahead (proposed by Oleg Drokin aka green)
37  *   - track min/max extents in each group for better group selection
38  *   - mb_mark_used() may allocate chunk right after splitting buddy
39  *   - tree of groups sorted by number of free blocks
40  *   - error handling
41  */
42 
43 /*
44  * The allocation request involve request for multiple number of blocks
45  * near to the goal(block) value specified.
46  *
47  * During initialization phase of the allocator we decide to use the
48  * group preallocation or inode preallocation depending on the size of
49  * the file. The size of the file could be the resulting file size we
50  * would have after allocation, or the current file size, which ever
51  * is larger. If the size is less than sbi->s_mb_stream_request we
52  * select to use the group preallocation. The default value of
53  * s_mb_stream_request is 16 blocks. This can also be tuned via
54  * /sys/fs/ext4/<partition>/mb_stream_req. The value is represented in
55  * terms of number of blocks.
56  *
57  * The main motivation for having small file use group preallocation is to
58  * ensure that we have small files closer together on the disk.
59  *
60  * First stage the allocator looks at the inode prealloc list,
61  * ext4_inode_info->i_prealloc_list, which contains list of prealloc
62  * spaces for this particular inode. The inode prealloc space is
63  * represented as:
64  *
65  * pa_lstart -> the logical start block for this prealloc space
66  * pa_pstart -> the physical start block for this prealloc space
67  * pa_len    -> length for this prealloc space (in clusters)
68  * pa_free   ->  free space available in this prealloc space (in clusters)
69  *
70  * The inode preallocation space is used looking at the _logical_ start
71  * block. If only the logical file block falls within the range of prealloc
72  * space we will consume the particular prealloc space. This makes sure that
73  * we have contiguous physical blocks representing the file blocks
74  *
75  * The important thing to be noted in case of inode prealloc space is that
76  * we don't modify the values associated to inode prealloc space except
77  * pa_free.
78  *
79  * If we are not able to find blocks in the inode prealloc space and if we
80  * have the group allocation flag set then we look at the locality group
81  * prealloc space. These are per CPU prealloc list represented as
82  *
83  * ext4_sb_info.s_locality_groups[smp_processor_id()]
84  *
85  * The reason for having a per cpu locality group is to reduce the contention
86  * between CPUs. It is possible to get scheduled at this point.
87  *
88  * The locality group prealloc space is used looking at whether we have
89  * enough free space (pa_free) within the prealloc space.
90  *
91  * If we can't allocate blocks via inode prealloc or/and locality group
92  * prealloc then we look at the buddy cache. The buddy cache is represented
93  * by ext4_sb_info.s_buddy_cache (struct inode) whose file offset gets
94  * mapped to the buddy and bitmap information regarding different
95  * groups. The buddy information is attached to buddy cache inode so that
96  * we can access them through the page cache. The information regarding
97  * each group is loaded via ext4_mb_load_buddy.  The information involve
98  * block bitmap and buddy information. The information are stored in the
99  * inode as:
100  *
101  *  {                        page                        }
102  *  [ group 0 bitmap][ group 0 buddy] [group 1][ group 1]...
103  *
104  *
105  * one block each for bitmap and buddy information.  So for each group we
106  * take up 2 blocks. A page can contain blocks_per_page (PAGE_SIZE /
107  * blocksize) blocks.  So it can have information regarding groups_per_page
108  * which is blocks_per_page/2
109  *
110  * The buddy cache inode is not stored on disk. The inode is thrown
111  * away when the filesystem is unmounted.
112  *
113  * We look for count number of blocks in the buddy cache. If we were able
114  * to locate that many free blocks we return with additional information
115  * regarding rest of the contiguous physical block available
116  *
117  * Before allocating blocks via buddy cache we normalize the request
118  * blocks. This ensure we ask for more blocks that we needed. The extra
119  * blocks that we get after allocation is added to the respective prealloc
120  * list. In case of inode preallocation we follow a list of heuristics
121  * based on file size. This can be found in ext4_mb_normalize_request. If
122  * we are doing a group prealloc we try to normalize the request to
123  * sbi->s_mb_group_prealloc.  The default value of s_mb_group_prealloc is
124  * dependent on the cluster size; for non-bigalloc file systems, it is
125  * 512 blocks. This can be tuned via
126  * /sys/fs/ext4/<partition>/mb_group_prealloc. The value is represented in
127  * terms of number of blocks. If we have mounted the file system with -O
128  * stripe=<value> option the group prealloc request is normalized to the
129  * smallest multiple of the stripe value (sbi->s_stripe) which is
130  * greater than the default mb_group_prealloc.
131  *
132  * If "mb_optimize_scan" mount option is set, we maintain in memory group info
133  * structures in two data structures:
134  *
135  * 1) Array of largest free order lists (sbi->s_mb_largest_free_orders)
136  *
137  *    Locking: sbi->s_mb_largest_free_orders_locks(array of rw locks)
138  *
139  *    This is an array of lists where the index in the array represents the
140  *    largest free order in the buddy bitmap of the participating group infos of
141  *    that list. So, there are exactly MB_NUM_ORDERS(sb) (which means total
142  *    number of buddy bitmap orders possible) number of lists. Group-infos are
143  *    placed in appropriate lists.
144  *
145  * 2) Average fragment size lists (sbi->s_mb_avg_fragment_size)
146  *
147  *    Locking: sbi->s_mb_avg_fragment_size_locks(array of rw locks)
148  *
149  *    This is an array of lists where in the i-th list there are groups with
150  *    average fragment size >= 2^i and < 2^(i+1). The average fragment size
151  *    is computed as ext4_group_info->bb_free / ext4_group_info->bb_fragments.
152  *    Note that we don't bother with a special list for completely empty groups
153  *    so we only have MB_NUM_ORDERS(sb) lists.
154  *
155  * When "mb_optimize_scan" mount option is set, mballoc consults the above data
156  * structures to decide the order in which groups are to be traversed for
157  * fulfilling an allocation request.
158  *
159  * At CR_POWER2_ALIGNED , we look for groups which have the largest_free_order
160  * >= the order of the request. We directly look at the largest free order list
161  * in the data structure (1) above where largest_free_order = order of the
162  * request. If that list is empty, we look at remaining list in the increasing
163  * order of largest_free_order. This allows us to perform CR_POWER2_ALIGNED
164  * lookup in O(1) time.
165  *
166  * At CR_GOAL_LEN_FAST, we only consider groups where
167  * average fragment size > request size. So, we lookup a group which has average
168  * fragment size just above or equal to request size using our average fragment
169  * size group lists (data structure 2) in O(1) time.
170  *
171  * At CR_BEST_AVAIL_LEN, we aim to optimize allocations which can't be satisfied
172  * in CR_GOAL_LEN_FAST. The fact that we couldn't find a group in
173  * CR_GOAL_LEN_FAST suggests that there is no BG that has avg
174  * fragment size > goal length. So before falling to the slower
175  * CR_GOAL_LEN_SLOW, in CR_BEST_AVAIL_LEN we proactively trim goal length and
176  * then use the same fragment lists as CR_GOAL_LEN_FAST to find a BG with a big
177  * enough average fragment size. This increases the chances of finding a
178  * suitable block group in O(1) time and results in faster allocation at the
179  * cost of reduced size of allocation.
180  *
181  * If "mb_optimize_scan" mount option is not set, mballoc traverses groups in
182  * linear order which requires O(N) search time for each CR_POWER2_ALIGNED and
183  * CR_GOAL_LEN_FAST phase.
184  *
185  * The regular allocator (using the buddy cache) supports a few tunables.
186  *
187  * /sys/fs/ext4/<partition>/mb_min_to_scan
188  * /sys/fs/ext4/<partition>/mb_max_to_scan
189  * /sys/fs/ext4/<partition>/mb_order2_req
190  * /sys/fs/ext4/<partition>/mb_linear_limit
191  *
192  * The regular allocator uses buddy scan only if the request len is power of
193  * 2 blocks and the order of allocation is >= sbi->s_mb_order2_reqs. The
194  * value of s_mb_order2_reqs can be tuned via
195  * /sys/fs/ext4/<partition>/mb_order2_req.  If the request len is equal to
196  * stripe size (sbi->s_stripe), we try to search for contiguous block in
197  * stripe size. This should result in better allocation on RAID setups. If
198  * not, we search in the specific group using bitmap for best extents. The
199  * tunable min_to_scan and max_to_scan control the behaviour here.
200  * min_to_scan indicate how long the mballoc __must__ look for a best
201  * extent and max_to_scan indicates how long the mballoc __can__ look for a
202  * best extent in the found extents. Searching for the blocks starts with
203  * the group specified as the goal value in allocation context via
204  * ac_g_ex. Each group is first checked based on the criteria whether it
205  * can be used for allocation. ext4_mb_good_group explains how the groups are
206  * checked.
207  *
208  * When "mb_optimize_scan" is turned on, as mentioned above, the groups may not
209  * get traversed linearly. That may result in subsequent allocations being not
210  * close to each other. And so, the underlying device may get filled up in a
211  * non-linear fashion. While that may not matter on non-rotational devices, for
212  * rotational devices that may result in higher seek times. "mb_linear_limit"
213  * tells mballoc how many groups mballoc should search linearly before
214  * performing consulting above data structures for more efficient lookups. For
215  * non rotational devices, this value defaults to 0 and for rotational devices
216  * this is set to MB_DEFAULT_LINEAR_LIMIT.
217  *
218  * Both the prealloc space are getting populated as above. So for the first
219  * request we will hit the buddy cache which will result in this prealloc
220  * space getting filled. The prealloc space is then later used for the
221  * subsequent request.
222  */
223 
224 /*
225  * mballoc operates on the following data:
226  *  - on-disk bitmap
227  *  - in-core buddy (actually includes buddy and bitmap)
228  *  - preallocation descriptors (PAs)
229  *
230  * there are two types of preallocations:
231  *  - inode
232  *    assiged to specific inode and can be used for this inode only.
233  *    it describes part of inode's space preallocated to specific
234  *    physical blocks. any block from that preallocated can be used
235  *    independent. the descriptor just tracks number of blocks left
236  *    unused. so, before taking some block from descriptor, one must
237  *    make sure corresponded logical block isn't allocated yet. this
238  *    also means that freeing any block within descriptor's range
239  *    must discard all preallocated blocks.
240  *  - locality group
241  *    assigned to specific locality group which does not translate to
242  *    permanent set of inodes: inode can join and leave group. space
243  *    from this type of preallocation can be used for any inode. thus
244  *    it's consumed from the beginning to the end.
245  *
246  * relation between them can be expressed as:
247  *    in-core buddy = on-disk bitmap + preallocation descriptors
248  *
249  * this mean blocks mballoc considers used are:
250  *  - allocated blocks (persistent)
251  *  - preallocated blocks (non-persistent)
252  *
253  * consistency in mballoc world means that at any time a block is either
254  * free or used in ALL structures. notice: "any time" should not be read
255  * literally -- time is discrete and delimited by locks.
256  *
257  *  to keep it simple, we don't use block numbers, instead we count number of
258  *  blocks: how many blocks marked used/free in on-disk bitmap, buddy and PA.
259  *
260  * all operations can be expressed as:
261  *  - init buddy:			buddy = on-disk + PAs
262  *  - new PA:				buddy += N; PA = N
263  *  - use inode PA:			on-disk += N; PA -= N
264  *  - discard inode PA			buddy -= on-disk - PA; PA = 0
265  *  - use locality group PA		on-disk += N; PA -= N
266  *  - discard locality group PA		buddy -= PA; PA = 0
267  *  note: 'buddy -= on-disk - PA' is used to show that on-disk bitmap
268  *        is used in real operation because we can't know actual used
269  *        bits from PA, only from on-disk bitmap
270  *
271  * if we follow this strict logic, then all operations above should be atomic.
272  * given some of them can block, we'd have to use something like semaphores
273  * killing performance on high-end SMP hardware. let's try to relax it using
274  * the following knowledge:
275  *  1) if buddy is referenced, it's already initialized
276  *  2) while block is used in buddy and the buddy is referenced,
277  *     nobody can re-allocate that block
278  *  3) we work on bitmaps and '+' actually means 'set bits'. if on-disk has
279  *     bit set and PA claims same block, it's OK. IOW, one can set bit in
280  *     on-disk bitmap if buddy has same bit set or/and PA covers corresponded
281  *     block
282  *
283  * so, now we're building a concurrency table:
284  *  - init buddy vs.
285  *    - new PA
286  *      blocks for PA are allocated in the buddy, buddy must be referenced
287  *      until PA is linked to allocation group to avoid concurrent buddy init
288  *    - use inode PA
289  *      we need to make sure that either on-disk bitmap or PA has uptodate data
290  *      given (3) we care that PA-=N operation doesn't interfere with init
291  *    - discard inode PA
292  *      the simplest way would be to have buddy initialized by the discard
293  *    - use locality group PA
294  *      again PA-=N must be serialized with init
295  *    - discard locality group PA
296  *      the simplest way would be to have buddy initialized by the discard
297  *  - new PA vs.
298  *    - use inode PA
299  *      i_data_sem serializes them
300  *    - discard inode PA
301  *      discard process must wait until PA isn't used by another process
302  *    - use locality group PA
303  *      some mutex should serialize them
304  *    - discard locality group PA
305  *      discard process must wait until PA isn't used by another process
306  *  - use inode PA
307  *    - use inode PA
308  *      i_data_sem or another mutex should serializes them
309  *    - discard inode PA
310  *      discard process must wait until PA isn't used by another process
311  *    - use locality group PA
312  *      nothing wrong here -- they're different PAs covering different blocks
313  *    - discard locality group PA
314  *      discard process must wait until PA isn't used by another process
315  *
316  * now we're ready to make few consequences:
317  *  - PA is referenced and while it is no discard is possible
318  *  - PA is referenced until block isn't marked in on-disk bitmap
319  *  - PA changes only after on-disk bitmap
320  *  - discard must not compete with init. either init is done before
321  *    any discard or they're serialized somehow
322  *  - buddy init as sum of on-disk bitmap and PAs is done atomically
323  *
324  * a special case when we've used PA to emptiness. no need to modify buddy
325  * in this case, but we should care about concurrent init
326  *
327  */
328 
329  /*
330  * Logic in few words:
331  *
332  *  - allocation:
333  *    load group
334  *    find blocks
335  *    mark bits in on-disk bitmap
336  *    release group
337  *
338  *  - use preallocation:
339  *    find proper PA (per-inode or group)
340  *    load group
341  *    mark bits in on-disk bitmap
342  *    release group
343  *    release PA
344  *
345  *  - free:
346  *    load group
347  *    mark bits in on-disk bitmap
348  *    release group
349  *
350  *  - discard preallocations in group:
351  *    mark PAs deleted
352  *    move them onto local list
353  *    load on-disk bitmap
354  *    load group
355  *    remove PA from object (inode or locality group)
356  *    mark free blocks in-core
357  *
358  *  - discard inode's preallocations:
359  */
360 
361 /*
362  * Locking rules
363  *
364  * Locks:
365  *  - bitlock on a group	(group)
366  *  - object (inode/locality)	(object)
367  *  - per-pa lock		(pa)
368  *  - cr_power2_aligned lists lock	(cr_power2_aligned)
369  *  - cr_goal_len_fast lists lock	(cr_goal_len_fast)
370  *
371  * Paths:
372  *  - new pa
373  *    object
374  *    group
375  *
376  *  - find and use pa:
377  *    pa
378  *
379  *  - release consumed pa:
380  *    pa
381  *    group
382  *    object
383  *
384  *  - generate in-core bitmap:
385  *    group
386  *        pa
387  *
388  *  - discard all for given object (inode, locality group):
389  *    object
390  *        pa
391  *    group
392  *
393  *  - discard all for given group:
394  *    group
395  *        pa
396  *    group
397  *        object
398  *
399  *  - allocation path (ext4_mb_regular_allocator)
400  *    group
401  *    cr_power2_aligned/cr_goal_len_fast
402  */
403 static struct kmem_cache *ext4_pspace_cachep;
404 static struct kmem_cache *ext4_ac_cachep;
405 static struct kmem_cache *ext4_free_data_cachep;
406 
407 /* We create slab caches for groupinfo data structures based on the
408  * superblock block size.  There will be one per mounted filesystem for
409  * each unique s_blocksize_bits */
410 #define NR_GRPINFO_CACHES 8
411 static struct kmem_cache *ext4_groupinfo_caches[NR_GRPINFO_CACHES];
412 
413 static const char * const ext4_groupinfo_slab_names[NR_GRPINFO_CACHES] = {
414 	"ext4_groupinfo_1k", "ext4_groupinfo_2k", "ext4_groupinfo_4k",
415 	"ext4_groupinfo_8k", "ext4_groupinfo_16k", "ext4_groupinfo_32k",
416 	"ext4_groupinfo_64k", "ext4_groupinfo_128k"
417 };
418 
419 static void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
420 					ext4_group_t group);
421 static void ext4_mb_new_preallocation(struct ext4_allocation_context *ac);
422 
423 static bool ext4_mb_good_group(struct ext4_allocation_context *ac,
424 			       ext4_group_t group, enum criteria cr);
425 
426 static int ext4_try_to_trim_range(struct super_block *sb,
427 		struct ext4_buddy *e4b, ext4_grpblk_t start,
428 		ext4_grpblk_t max, ext4_grpblk_t minblocks);
429 
430 /*
431  * The algorithm using this percpu seq counter goes below:
432  * 1. We sample the percpu discard_pa_seq counter before trying for block
433  *    allocation in ext4_mb_new_blocks().
434  * 2. We increment this percpu discard_pa_seq counter when we either allocate
435  *    or free these blocks i.e. while marking those blocks as used/free in
436  *    mb_mark_used()/mb_free_blocks().
437  * 3. We also increment this percpu seq counter when we successfully identify
438  *    that the bb_prealloc_list is not empty and hence proceed for discarding
439  *    of those PAs inside ext4_mb_discard_group_preallocations().
440  *
441  * Now to make sure that the regular fast path of block allocation is not
442  * affected, as a small optimization we only sample the percpu seq counter
443  * on that cpu. Only when the block allocation fails and when freed blocks
444  * found were 0, that is when we sample percpu seq counter for all cpus using
445  * below function ext4_get_discard_pa_seq_sum(). This happens after making
446  * sure that all the PAs on grp->bb_prealloc_list got freed or if it's empty.
447  */
448 static DEFINE_PER_CPU(u64, discard_pa_seq);
ext4_get_discard_pa_seq_sum(void)449 static inline u64 ext4_get_discard_pa_seq_sum(void)
450 {
451 	int __cpu;
452 	u64 __seq = 0;
453 
454 	for_each_possible_cpu(__cpu)
455 		__seq += per_cpu(discard_pa_seq, __cpu);
456 	return __seq;
457 }
458 
mb_correct_addr_and_bit(int * bit,void * addr)459 static inline void *mb_correct_addr_and_bit(int *bit, void *addr)
460 {
461 #if BITS_PER_LONG == 64
462 	*bit += ((unsigned long) addr & 7UL) << 3;
463 	addr = (void *) ((unsigned long) addr & ~7UL);
464 #elif BITS_PER_LONG == 32
465 	*bit += ((unsigned long) addr & 3UL) << 3;
466 	addr = (void *) ((unsigned long) addr & ~3UL);
467 #else
468 #error "how many bits you are?!"
469 #endif
470 	return addr;
471 }
472 
mb_test_bit(int bit,void * addr)473 static inline int mb_test_bit(int bit, void *addr)
474 {
475 	/*
476 	 * ext4_test_bit on architecture like powerpc
477 	 * needs unsigned long aligned address
478 	 */
479 	addr = mb_correct_addr_and_bit(&bit, addr);
480 	return ext4_test_bit(bit, addr);
481 }
482 
mb_set_bit(int bit,void * addr)483 static inline void mb_set_bit(int bit, void *addr)
484 {
485 	addr = mb_correct_addr_and_bit(&bit, addr);
486 	ext4_set_bit(bit, addr);
487 }
488 
mb_clear_bit(int bit,void * addr)489 static inline void mb_clear_bit(int bit, void *addr)
490 {
491 	addr = mb_correct_addr_and_bit(&bit, addr);
492 	ext4_clear_bit(bit, addr);
493 }
494 
mb_test_and_clear_bit(int bit,void * addr)495 static inline int mb_test_and_clear_bit(int bit, void *addr)
496 {
497 	addr = mb_correct_addr_and_bit(&bit, addr);
498 	return ext4_test_and_clear_bit(bit, addr);
499 }
500 
mb_find_next_zero_bit(void * addr,int max,int start)501 static inline int mb_find_next_zero_bit(void *addr, int max, int start)
502 {
503 	int fix = 0, ret, tmpmax;
504 	addr = mb_correct_addr_and_bit(&fix, addr);
505 	tmpmax = max + fix;
506 	start += fix;
507 
508 	ret = ext4_find_next_zero_bit(addr, tmpmax, start) - fix;
509 	if (ret > max)
510 		return max;
511 	return ret;
512 }
513 
mb_find_next_bit(void * addr,int max,int start)514 static inline int mb_find_next_bit(void *addr, int max, int start)
515 {
516 	int fix = 0, ret, tmpmax;
517 	addr = mb_correct_addr_and_bit(&fix, addr);
518 	tmpmax = max + fix;
519 	start += fix;
520 
521 	ret = ext4_find_next_bit(addr, tmpmax, start) - fix;
522 	if (ret > max)
523 		return max;
524 	return ret;
525 }
526 
mb_find_buddy(struct ext4_buddy * e4b,int order,int * max)527 static void *mb_find_buddy(struct ext4_buddy *e4b, int order, int *max)
528 {
529 	char *bb;
530 
531 	BUG_ON(e4b->bd_bitmap == e4b->bd_buddy);
532 	BUG_ON(max == NULL);
533 
534 	if (order > e4b->bd_blkbits + 1) {
535 		*max = 0;
536 		return NULL;
537 	}
538 
539 	/* at order 0 we see each particular block */
540 	if (order == 0) {
541 		*max = 1 << (e4b->bd_blkbits + 3);
542 		return e4b->bd_bitmap;
543 	}
544 
545 	bb = e4b->bd_buddy + EXT4_SB(e4b->bd_sb)->s_mb_offsets[order];
546 	*max = EXT4_SB(e4b->bd_sb)->s_mb_maxs[order];
547 
548 	return bb;
549 }
550 
551 #ifdef DOUBLE_CHECK
mb_free_blocks_double(struct inode * inode,struct ext4_buddy * e4b,int first,int count)552 static void mb_free_blocks_double(struct inode *inode, struct ext4_buddy *e4b,
553 			   int first, int count)
554 {
555 	int i;
556 	struct super_block *sb = e4b->bd_sb;
557 
558 	if (unlikely(e4b->bd_info->bb_bitmap == NULL))
559 		return;
560 	assert_spin_locked(ext4_group_lock_ptr(sb, e4b->bd_group));
561 	for (i = 0; i < count; i++) {
562 		if (!mb_test_bit(first + i, e4b->bd_info->bb_bitmap)) {
563 			ext4_fsblk_t blocknr;
564 
565 			blocknr = ext4_group_first_block_no(sb, e4b->bd_group);
566 			blocknr += EXT4_C2B(EXT4_SB(sb), first + i);
567 			ext4_mark_group_bitmap_corrupted(sb, e4b->bd_group,
568 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
569 			ext4_grp_locked_error(sb, e4b->bd_group,
570 					      inode ? inode->i_ino : 0,
571 					      blocknr,
572 					      "freeing block already freed "
573 					      "(bit %u)",
574 					      first + i);
575 		}
576 		mb_clear_bit(first + i, e4b->bd_info->bb_bitmap);
577 	}
578 }
579 
mb_mark_used_double(struct ext4_buddy * e4b,int first,int count)580 static void mb_mark_used_double(struct ext4_buddy *e4b, int first, int count)
581 {
582 	int i;
583 
584 	if (unlikely(e4b->bd_info->bb_bitmap == NULL))
585 		return;
586 	assert_spin_locked(ext4_group_lock_ptr(e4b->bd_sb, e4b->bd_group));
587 	for (i = 0; i < count; i++) {
588 		BUG_ON(mb_test_bit(first + i, e4b->bd_info->bb_bitmap));
589 		mb_set_bit(first + i, e4b->bd_info->bb_bitmap);
590 	}
591 }
592 
mb_cmp_bitmaps(struct ext4_buddy * e4b,void * bitmap)593 static void mb_cmp_bitmaps(struct ext4_buddy *e4b, void *bitmap)
594 {
595 	if (unlikely(e4b->bd_info->bb_bitmap == NULL))
596 		return;
597 	if (memcmp(e4b->bd_info->bb_bitmap, bitmap, e4b->bd_sb->s_blocksize)) {
598 		unsigned char *b1, *b2;
599 		int i;
600 		b1 = (unsigned char *) e4b->bd_info->bb_bitmap;
601 		b2 = (unsigned char *) bitmap;
602 		for (i = 0; i < e4b->bd_sb->s_blocksize; i++) {
603 			if (b1[i] != b2[i]) {
604 				ext4_msg(e4b->bd_sb, KERN_ERR,
605 					 "corruption in group %u "
606 					 "at byte %u(%u): %x in copy != %x "
607 					 "on disk/prealloc",
608 					 e4b->bd_group, i, i * 8, b1[i], b2[i]);
609 				BUG();
610 			}
611 		}
612 	}
613 }
614 
mb_group_bb_bitmap_alloc(struct super_block * sb,struct ext4_group_info * grp,ext4_group_t group)615 static void mb_group_bb_bitmap_alloc(struct super_block *sb,
616 			struct ext4_group_info *grp, ext4_group_t group)
617 {
618 	struct buffer_head *bh;
619 
620 	grp->bb_bitmap = kmalloc(sb->s_blocksize, GFP_NOFS);
621 	if (!grp->bb_bitmap)
622 		return;
623 
624 	bh = ext4_read_block_bitmap(sb, group);
625 	if (IS_ERR_OR_NULL(bh)) {
626 		kfree(grp->bb_bitmap);
627 		grp->bb_bitmap = NULL;
628 		return;
629 	}
630 
631 	memcpy(grp->bb_bitmap, bh->b_data, sb->s_blocksize);
632 	put_bh(bh);
633 }
634 
mb_group_bb_bitmap_free(struct ext4_group_info * grp)635 static void mb_group_bb_bitmap_free(struct ext4_group_info *grp)
636 {
637 	kfree(grp->bb_bitmap);
638 }
639 
640 #else
mb_free_blocks_double(struct inode * inode,struct ext4_buddy * e4b,int first,int count)641 static inline void mb_free_blocks_double(struct inode *inode,
642 				struct ext4_buddy *e4b, int first, int count)
643 {
644 	return;
645 }
mb_mark_used_double(struct ext4_buddy * e4b,int first,int count)646 static inline void mb_mark_used_double(struct ext4_buddy *e4b,
647 						int first, int count)
648 {
649 	return;
650 }
mb_cmp_bitmaps(struct ext4_buddy * e4b,void * bitmap)651 static inline void mb_cmp_bitmaps(struct ext4_buddy *e4b, void *bitmap)
652 {
653 	return;
654 }
655 
mb_group_bb_bitmap_alloc(struct super_block * sb,struct ext4_group_info * grp,ext4_group_t group)656 static inline void mb_group_bb_bitmap_alloc(struct super_block *sb,
657 			struct ext4_group_info *grp, ext4_group_t group)
658 {
659 	return;
660 }
661 
mb_group_bb_bitmap_free(struct ext4_group_info * grp)662 static inline void mb_group_bb_bitmap_free(struct ext4_group_info *grp)
663 {
664 	return;
665 }
666 #endif
667 
668 #ifdef AGGRESSIVE_CHECK
669 
670 #define MB_CHECK_ASSERT(assert)						\
671 do {									\
672 	if (!(assert)) {						\
673 		printk(KERN_EMERG					\
674 			"Assertion failure in %s() at %s:%d: \"%s\"\n",	\
675 			function, file, line, # assert);		\
676 		BUG();							\
677 	}								\
678 } while (0)
679 
__mb_check_buddy(struct ext4_buddy * e4b,char * file,const char * function,int line)680 static void __mb_check_buddy(struct ext4_buddy *e4b, char *file,
681 				const char *function, int line)
682 {
683 	struct super_block *sb = e4b->bd_sb;
684 	int order = e4b->bd_blkbits + 1;
685 	int max;
686 	int max2;
687 	int i;
688 	int j;
689 	int k;
690 	int count;
691 	struct ext4_group_info *grp;
692 	int fragments = 0;
693 	int fstart;
694 	struct list_head *cur;
695 	void *buddy;
696 	void *buddy2;
697 
698 	if (e4b->bd_info->bb_check_counter++ % 10)
699 		return;
700 
701 	while (order > 1) {
702 		buddy = mb_find_buddy(e4b, order, &max);
703 		MB_CHECK_ASSERT(buddy);
704 		buddy2 = mb_find_buddy(e4b, order - 1, &max2);
705 		MB_CHECK_ASSERT(buddy2);
706 		MB_CHECK_ASSERT(buddy != buddy2);
707 		MB_CHECK_ASSERT(max * 2 == max2);
708 
709 		count = 0;
710 		for (i = 0; i < max; i++) {
711 
712 			if (mb_test_bit(i, buddy)) {
713 				/* only single bit in buddy2 may be 0 */
714 				if (!mb_test_bit(i << 1, buddy2)) {
715 					MB_CHECK_ASSERT(
716 						mb_test_bit((i<<1)+1, buddy2));
717 				}
718 				continue;
719 			}
720 
721 			/* both bits in buddy2 must be 1 */
722 			MB_CHECK_ASSERT(mb_test_bit(i << 1, buddy2));
723 			MB_CHECK_ASSERT(mb_test_bit((i << 1) + 1, buddy2));
724 
725 			for (j = 0; j < (1 << order); j++) {
726 				k = (i * (1 << order)) + j;
727 				MB_CHECK_ASSERT(
728 					!mb_test_bit(k, e4b->bd_bitmap));
729 			}
730 			count++;
731 		}
732 		MB_CHECK_ASSERT(e4b->bd_info->bb_counters[order] == count);
733 		order--;
734 	}
735 
736 	fstart = -1;
737 	buddy = mb_find_buddy(e4b, 0, &max);
738 	for (i = 0; i < max; i++) {
739 		if (!mb_test_bit(i, buddy)) {
740 			MB_CHECK_ASSERT(i >= e4b->bd_info->bb_first_free);
741 			if (fstart == -1) {
742 				fragments++;
743 				fstart = i;
744 			}
745 			continue;
746 		}
747 		fstart = -1;
748 		/* check used bits only */
749 		for (j = 0; j < e4b->bd_blkbits + 1; j++) {
750 			buddy2 = mb_find_buddy(e4b, j, &max2);
751 			k = i >> j;
752 			MB_CHECK_ASSERT(k < max2);
753 			MB_CHECK_ASSERT(mb_test_bit(k, buddy2));
754 		}
755 	}
756 	MB_CHECK_ASSERT(!EXT4_MB_GRP_NEED_INIT(e4b->bd_info));
757 	MB_CHECK_ASSERT(e4b->bd_info->bb_fragments == fragments);
758 
759 	grp = ext4_get_group_info(sb, e4b->bd_group);
760 	if (!grp)
761 		return;
762 	list_for_each(cur, &grp->bb_prealloc_list) {
763 		ext4_group_t groupnr;
764 		struct ext4_prealloc_space *pa;
765 		pa = list_entry(cur, struct ext4_prealloc_space, pa_group_list);
766 		ext4_get_group_no_and_offset(sb, pa->pa_pstart, &groupnr, &k);
767 		MB_CHECK_ASSERT(groupnr == e4b->bd_group);
768 		for (i = 0; i < pa->pa_len; i++)
769 			MB_CHECK_ASSERT(mb_test_bit(k + i, buddy));
770 	}
771 }
772 #undef MB_CHECK_ASSERT
773 #define mb_check_buddy(e4b) __mb_check_buddy(e4b,	\
774 					__FILE__, __func__, __LINE__)
775 #else
776 #define mb_check_buddy(e4b)
777 #endif
778 
779 /*
780  * Divide blocks started from @first with length @len into
781  * smaller chunks with power of 2 blocks.
782  * Clear the bits in bitmap which the blocks of the chunk(s) covered,
783  * then increase bb_counters[] for corresponded chunk size.
784  */
ext4_mb_mark_free_simple(struct super_block * sb,void * buddy,ext4_grpblk_t first,ext4_grpblk_t len,struct ext4_group_info * grp)785 static void ext4_mb_mark_free_simple(struct super_block *sb,
786 				void *buddy, ext4_grpblk_t first, ext4_grpblk_t len,
787 					struct ext4_group_info *grp)
788 {
789 	struct ext4_sb_info *sbi = EXT4_SB(sb);
790 	ext4_grpblk_t min;
791 	ext4_grpblk_t max;
792 	ext4_grpblk_t chunk;
793 	unsigned int border;
794 
795 	BUG_ON(len > EXT4_CLUSTERS_PER_GROUP(sb));
796 
797 	border = 2 << sb->s_blocksize_bits;
798 
799 	while (len > 0) {
800 		/* find how many blocks can be covered since this position */
801 		max = ffs(first | border) - 1;
802 
803 		/* find how many blocks of power 2 we need to mark */
804 		min = fls(len) - 1;
805 
806 		if (max < min)
807 			min = max;
808 		chunk = 1 << min;
809 
810 		/* mark multiblock chunks only */
811 		grp->bb_counters[min]++;
812 		if (min > 0)
813 			mb_clear_bit(first >> min,
814 				     buddy + sbi->s_mb_offsets[min]);
815 
816 		len -= chunk;
817 		first += chunk;
818 	}
819 }
820 
mb_avg_fragment_size_order(struct super_block * sb,ext4_grpblk_t len)821 static int mb_avg_fragment_size_order(struct super_block *sb, ext4_grpblk_t len)
822 {
823 	int order;
824 
825 	/*
826 	 * We don't bother with a special lists groups with only 1 block free
827 	 * extents and for completely empty groups.
828 	 */
829 	order = fls(len) - 2;
830 	if (order < 0)
831 		return 0;
832 	if (order == MB_NUM_ORDERS(sb))
833 		order--;
834 	if (WARN_ON_ONCE(order > MB_NUM_ORDERS(sb)))
835 		order = MB_NUM_ORDERS(sb) - 1;
836 	return order;
837 }
838 
839 /* Move group to appropriate avg_fragment_size list */
840 static void
mb_update_avg_fragment_size(struct super_block * sb,struct ext4_group_info * grp)841 mb_update_avg_fragment_size(struct super_block *sb, struct ext4_group_info *grp)
842 {
843 	struct ext4_sb_info *sbi = EXT4_SB(sb);
844 	int new_order;
845 
846 	if (!test_opt2(sb, MB_OPTIMIZE_SCAN) || grp->bb_fragments == 0)
847 		return;
848 
849 	new_order = mb_avg_fragment_size_order(sb,
850 					grp->bb_free / grp->bb_fragments);
851 	if (new_order == grp->bb_avg_fragment_size_order)
852 		return;
853 
854 	if (grp->bb_avg_fragment_size_order != -1) {
855 		write_lock(&sbi->s_mb_avg_fragment_size_locks[
856 					grp->bb_avg_fragment_size_order]);
857 		list_del(&grp->bb_avg_fragment_size_node);
858 		write_unlock(&sbi->s_mb_avg_fragment_size_locks[
859 					grp->bb_avg_fragment_size_order]);
860 	}
861 	grp->bb_avg_fragment_size_order = new_order;
862 	write_lock(&sbi->s_mb_avg_fragment_size_locks[
863 					grp->bb_avg_fragment_size_order]);
864 	list_add_tail(&grp->bb_avg_fragment_size_node,
865 		&sbi->s_mb_avg_fragment_size[grp->bb_avg_fragment_size_order]);
866 	write_unlock(&sbi->s_mb_avg_fragment_size_locks[
867 					grp->bb_avg_fragment_size_order]);
868 }
869 
870 /*
871  * Choose next group by traversing largest_free_order lists. Updates *new_cr if
872  * cr level needs an update.
873  */
ext4_mb_choose_next_group_p2_aligned(struct ext4_allocation_context * ac,enum criteria * new_cr,ext4_group_t * group)874 static void ext4_mb_choose_next_group_p2_aligned(struct ext4_allocation_context *ac,
875 			enum criteria *new_cr, ext4_group_t *group)
876 {
877 	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
878 	struct ext4_group_info *iter;
879 	int i;
880 
881 	if (ac->ac_status == AC_STATUS_FOUND)
882 		return;
883 
884 	if (unlikely(sbi->s_mb_stats && ac->ac_flags & EXT4_MB_CR_POWER2_ALIGNED_OPTIMIZED))
885 		atomic_inc(&sbi->s_bal_p2_aligned_bad_suggestions);
886 
887 	for (i = ac->ac_2order; i < MB_NUM_ORDERS(ac->ac_sb); i++) {
888 		if (list_empty(&sbi->s_mb_largest_free_orders[i]))
889 			continue;
890 		read_lock(&sbi->s_mb_largest_free_orders_locks[i]);
891 		if (list_empty(&sbi->s_mb_largest_free_orders[i])) {
892 			read_unlock(&sbi->s_mb_largest_free_orders_locks[i]);
893 			continue;
894 		}
895 		list_for_each_entry(iter, &sbi->s_mb_largest_free_orders[i],
896 				    bb_largest_free_order_node) {
897 			if (sbi->s_mb_stats)
898 				atomic64_inc(&sbi->s_bal_cX_groups_considered[CR_POWER2_ALIGNED]);
899 			if (likely(ext4_mb_good_group(ac, iter->bb_group, CR_POWER2_ALIGNED))) {
900 				*group = iter->bb_group;
901 				ac->ac_flags |= EXT4_MB_CR_POWER2_ALIGNED_OPTIMIZED;
902 				read_unlock(&sbi->s_mb_largest_free_orders_locks[i]);
903 				return;
904 			}
905 		}
906 		read_unlock(&sbi->s_mb_largest_free_orders_locks[i]);
907 	}
908 
909 	/* Increment cr and search again if no group is found */
910 	*new_cr = CR_GOAL_LEN_FAST;
911 }
912 
913 /*
914  * Find a suitable group of given order from the average fragments list.
915  */
916 static struct ext4_group_info *
ext4_mb_find_good_group_avg_frag_lists(struct ext4_allocation_context * ac,int order)917 ext4_mb_find_good_group_avg_frag_lists(struct ext4_allocation_context *ac, int order)
918 {
919 	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
920 	struct list_head *frag_list = &sbi->s_mb_avg_fragment_size[order];
921 	rwlock_t *frag_list_lock = &sbi->s_mb_avg_fragment_size_locks[order];
922 	struct ext4_group_info *grp = NULL, *iter;
923 	enum criteria cr = ac->ac_criteria;
924 
925 	if (list_empty(frag_list))
926 		return NULL;
927 	read_lock(frag_list_lock);
928 	if (list_empty(frag_list)) {
929 		read_unlock(frag_list_lock);
930 		return NULL;
931 	}
932 	list_for_each_entry(iter, frag_list, bb_avg_fragment_size_node) {
933 		if (sbi->s_mb_stats)
934 			atomic64_inc(&sbi->s_bal_cX_groups_considered[cr]);
935 		if (likely(ext4_mb_good_group(ac, iter->bb_group, cr))) {
936 			grp = iter;
937 			break;
938 		}
939 	}
940 	read_unlock(frag_list_lock);
941 	return grp;
942 }
943 
944 /*
945  * Choose next group by traversing average fragment size list of suitable
946  * order. Updates *new_cr if cr level needs an update.
947  */
ext4_mb_choose_next_group_goal_fast(struct ext4_allocation_context * ac,enum criteria * new_cr,ext4_group_t * group)948 static void ext4_mb_choose_next_group_goal_fast(struct ext4_allocation_context *ac,
949 		enum criteria *new_cr, ext4_group_t *group)
950 {
951 	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
952 	struct ext4_group_info *grp = NULL;
953 	int i;
954 
955 	if (unlikely(ac->ac_flags & EXT4_MB_CR_GOAL_LEN_FAST_OPTIMIZED)) {
956 		if (sbi->s_mb_stats)
957 			atomic_inc(&sbi->s_bal_goal_fast_bad_suggestions);
958 	}
959 
960 	for (i = mb_avg_fragment_size_order(ac->ac_sb, ac->ac_g_ex.fe_len);
961 	     i < MB_NUM_ORDERS(ac->ac_sb); i++) {
962 		grp = ext4_mb_find_good_group_avg_frag_lists(ac, i);
963 		if (grp) {
964 			*group = grp->bb_group;
965 			ac->ac_flags |= EXT4_MB_CR_GOAL_LEN_FAST_OPTIMIZED;
966 			return;
967 		}
968 	}
969 
970 	/*
971 	 * CR_BEST_AVAIL_LEN works based on the concept that we have
972 	 * a larger normalized goal len request which can be trimmed to
973 	 * a smaller goal len such that it can still satisfy original
974 	 * request len. However, allocation request for non-regular
975 	 * files never gets normalized.
976 	 * See function ext4_mb_normalize_request() (EXT4_MB_HINT_DATA).
977 	 */
978 	if (ac->ac_flags & EXT4_MB_HINT_DATA)
979 		*new_cr = CR_BEST_AVAIL_LEN;
980 	else
981 		*new_cr = CR_GOAL_LEN_SLOW;
982 }
983 
984 /*
985  * We couldn't find a group in CR_GOAL_LEN_FAST so try to find the highest free fragment
986  * order we have and proactively trim the goal request length to that order to
987  * find a suitable group faster.
988  *
989  * This optimizes allocation speed at the cost of slightly reduced
990  * preallocations. However, we make sure that we don't trim the request too
991  * much and fall to CR_GOAL_LEN_SLOW in that case.
992  */
ext4_mb_choose_next_group_best_avail(struct ext4_allocation_context * ac,enum criteria * new_cr,ext4_group_t * group)993 static void ext4_mb_choose_next_group_best_avail(struct ext4_allocation_context *ac,
994 		enum criteria *new_cr, ext4_group_t *group)
995 {
996 	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
997 	struct ext4_group_info *grp = NULL;
998 	int i, order, min_order;
999 	unsigned long num_stripe_clusters = 0;
1000 
1001 	if (unlikely(ac->ac_flags & EXT4_MB_CR_BEST_AVAIL_LEN_OPTIMIZED)) {
1002 		if (sbi->s_mb_stats)
1003 			atomic_inc(&sbi->s_bal_best_avail_bad_suggestions);
1004 	}
1005 
1006 	/*
1007 	 * mb_avg_fragment_size_order() returns order in a way that makes
1008 	 * retrieving back the length using (1 << order) inaccurate. Hence, use
1009 	 * fls() instead since we need to know the actual length while modifying
1010 	 * goal length.
1011 	 */
1012 	order = fls(ac->ac_g_ex.fe_len) - 1;
1013 	if (WARN_ON_ONCE(order - 1 > MB_NUM_ORDERS(ac->ac_sb)))
1014 		order = MB_NUM_ORDERS(ac->ac_sb);
1015 	min_order = order - sbi->s_mb_best_avail_max_trim_order;
1016 	if (min_order < 0)
1017 		min_order = 0;
1018 
1019 	if (sbi->s_stripe > 0) {
1020 		/*
1021 		 * We are assuming that stripe size is always a multiple of
1022 		 * cluster ratio otherwise __ext4_fill_super exists early.
1023 		 */
1024 		num_stripe_clusters = EXT4_NUM_B2C(sbi, sbi->s_stripe);
1025 		if (1 << min_order < num_stripe_clusters)
1026 			/*
1027 			 * We consider 1 order less because later we round
1028 			 * up the goal len to num_stripe_clusters
1029 			 */
1030 			min_order = fls(num_stripe_clusters) - 1;
1031 	}
1032 
1033 	if (1 << min_order < ac->ac_o_ex.fe_len)
1034 		min_order = fls(ac->ac_o_ex.fe_len);
1035 
1036 	for (i = order; i >= min_order; i--) {
1037 		int frag_order;
1038 		/*
1039 		 * Scale down goal len to make sure we find something
1040 		 * in the free fragments list. Basically, reduce
1041 		 * preallocations.
1042 		 */
1043 		ac->ac_g_ex.fe_len = 1 << i;
1044 
1045 		if (num_stripe_clusters > 0) {
1046 			/*
1047 			 * Try to round up the adjusted goal length to
1048 			 * stripe size (in cluster units) multiple for
1049 			 * efficiency.
1050 			 */
1051 			ac->ac_g_ex.fe_len = roundup(ac->ac_g_ex.fe_len,
1052 						     num_stripe_clusters);
1053 		}
1054 
1055 		frag_order = mb_avg_fragment_size_order(ac->ac_sb,
1056 							ac->ac_g_ex.fe_len);
1057 
1058 		grp = ext4_mb_find_good_group_avg_frag_lists(ac, frag_order);
1059 		if (grp) {
1060 			*group = grp->bb_group;
1061 			ac->ac_flags |= EXT4_MB_CR_BEST_AVAIL_LEN_OPTIMIZED;
1062 			return;
1063 		}
1064 	}
1065 
1066 	/* Reset goal length to original goal length before falling into CR_GOAL_LEN_SLOW */
1067 	ac->ac_g_ex.fe_len = ac->ac_orig_goal_len;
1068 	*new_cr = CR_GOAL_LEN_SLOW;
1069 }
1070 
should_optimize_scan(struct ext4_allocation_context * ac)1071 static inline int should_optimize_scan(struct ext4_allocation_context *ac)
1072 {
1073 	if (unlikely(!test_opt2(ac->ac_sb, MB_OPTIMIZE_SCAN)))
1074 		return 0;
1075 	if (ac->ac_criteria >= CR_GOAL_LEN_SLOW)
1076 		return 0;
1077 	if (!ext4_test_inode_flag(ac->ac_inode, EXT4_INODE_EXTENTS))
1078 		return 0;
1079 	return 1;
1080 }
1081 
1082 /*
1083  * Return next linear group for allocation.
1084  */
1085 static ext4_group_t
next_linear_group(ext4_group_t group,ext4_group_t ngroups)1086 next_linear_group(ext4_group_t group, ext4_group_t ngroups)
1087 {
1088 	/*
1089 	 * Artificially restricted ngroups for non-extent
1090 	 * files makes group > ngroups possible on first loop.
1091 	 */
1092 	return group + 1 >= ngroups ? 0 : group + 1;
1093 }
1094 
1095 /*
1096  * ext4_mb_choose_next_group: choose next group for allocation.
1097  *
1098  * @ac        Allocation Context
1099  * @new_cr    This is an output parameter. If the there is no good group
1100  *            available at current CR level, this field is updated to indicate
1101  *            the new cr level that should be used.
1102  * @group     This is an input / output parameter. As an input it indicates the
1103  *            next group that the allocator intends to use for allocation. As
1104  *            output, this field indicates the next group that should be used as
1105  *            determined by the optimization functions.
1106  * @ngroups   Total number of groups
1107  */
ext4_mb_choose_next_group(struct ext4_allocation_context * ac,enum criteria * new_cr,ext4_group_t * group,ext4_group_t ngroups)1108 static void ext4_mb_choose_next_group(struct ext4_allocation_context *ac,
1109 		enum criteria *new_cr, ext4_group_t *group, ext4_group_t ngroups)
1110 {
1111 	*new_cr = ac->ac_criteria;
1112 
1113 	if (!should_optimize_scan(ac)) {
1114 		*group = next_linear_group(*group, ngroups);
1115 		return;
1116 	}
1117 
1118 	/*
1119 	 * Optimized scanning can return non adjacent groups which can cause
1120 	 * seek overhead for rotational disks. So try few linear groups before
1121 	 * trying optimized scan.
1122 	 */
1123 	if (ac->ac_groups_linear_remaining) {
1124 		*group = next_linear_group(*group, ngroups);
1125 		ac->ac_groups_linear_remaining--;
1126 		return;
1127 	}
1128 
1129 	if (*new_cr == CR_POWER2_ALIGNED) {
1130 		ext4_mb_choose_next_group_p2_aligned(ac, new_cr, group);
1131 	} else if (*new_cr == CR_GOAL_LEN_FAST) {
1132 		ext4_mb_choose_next_group_goal_fast(ac, new_cr, group);
1133 	} else if (*new_cr == CR_BEST_AVAIL_LEN) {
1134 		ext4_mb_choose_next_group_best_avail(ac, new_cr, group);
1135 	} else {
1136 		/*
1137 		 * TODO: For CR_GOAL_LEN_SLOW, we can arrange groups in an
1138 		 * rb tree sorted by bb_free. But until that happens, we should
1139 		 * never come here.
1140 		 */
1141 		WARN_ON(1);
1142 	}
1143 }
1144 
1145 /*
1146  * Cache the order of the largest free extent we have available in this block
1147  * group.
1148  */
1149 static void
mb_set_largest_free_order(struct super_block * sb,struct ext4_group_info * grp)1150 mb_set_largest_free_order(struct super_block *sb, struct ext4_group_info *grp)
1151 {
1152 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1153 	int i;
1154 
1155 	for (i = MB_NUM_ORDERS(sb) - 1; i >= 0; i--)
1156 		if (grp->bb_counters[i] > 0)
1157 			break;
1158 	/* No need to move between order lists? */
1159 	if (!test_opt2(sb, MB_OPTIMIZE_SCAN) ||
1160 	    i == grp->bb_largest_free_order) {
1161 		grp->bb_largest_free_order = i;
1162 		return;
1163 	}
1164 
1165 	if (grp->bb_largest_free_order >= 0) {
1166 		write_lock(&sbi->s_mb_largest_free_orders_locks[
1167 					      grp->bb_largest_free_order]);
1168 		list_del_init(&grp->bb_largest_free_order_node);
1169 		write_unlock(&sbi->s_mb_largest_free_orders_locks[
1170 					      grp->bb_largest_free_order]);
1171 	}
1172 	grp->bb_largest_free_order = i;
1173 	if (grp->bb_largest_free_order >= 0 && grp->bb_free) {
1174 		write_lock(&sbi->s_mb_largest_free_orders_locks[
1175 					      grp->bb_largest_free_order]);
1176 		list_add_tail(&grp->bb_largest_free_order_node,
1177 		      &sbi->s_mb_largest_free_orders[grp->bb_largest_free_order]);
1178 		write_unlock(&sbi->s_mb_largest_free_orders_locks[
1179 					      grp->bb_largest_free_order]);
1180 	}
1181 }
1182 
1183 static noinline_for_stack
ext4_mb_generate_buddy(struct super_block * sb,void * buddy,void * bitmap,ext4_group_t group,struct ext4_group_info * grp)1184 void ext4_mb_generate_buddy(struct super_block *sb,
1185 			    void *buddy, void *bitmap, ext4_group_t group,
1186 			    struct ext4_group_info *grp)
1187 {
1188 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1189 	ext4_grpblk_t max = EXT4_CLUSTERS_PER_GROUP(sb);
1190 	ext4_grpblk_t i = 0;
1191 	ext4_grpblk_t first;
1192 	ext4_grpblk_t len;
1193 	unsigned free = 0;
1194 	unsigned fragments = 0;
1195 	unsigned long long period = get_cycles();
1196 
1197 	/* initialize buddy from bitmap which is aggregation
1198 	 * of on-disk bitmap and preallocations */
1199 	i = mb_find_next_zero_bit(bitmap, max, 0);
1200 	grp->bb_first_free = i;
1201 	while (i < max) {
1202 		fragments++;
1203 		first = i;
1204 		i = mb_find_next_bit(bitmap, max, i);
1205 		len = i - first;
1206 		free += len;
1207 		if (len > 1)
1208 			ext4_mb_mark_free_simple(sb, buddy, first, len, grp);
1209 		else
1210 			grp->bb_counters[0]++;
1211 		if (i < max)
1212 			i = mb_find_next_zero_bit(bitmap, max, i);
1213 	}
1214 	grp->bb_fragments = fragments;
1215 
1216 	if (free != grp->bb_free) {
1217 		ext4_grp_locked_error(sb, group, 0, 0,
1218 				      "block bitmap and bg descriptor "
1219 				      "inconsistent: %u vs %u free clusters",
1220 				      free, grp->bb_free);
1221 		/*
1222 		 * If we intend to continue, we consider group descriptor
1223 		 * corrupt and update bb_free using bitmap value
1224 		 */
1225 		grp->bb_free = free;
1226 		ext4_mark_group_bitmap_corrupted(sb, group,
1227 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
1228 	}
1229 	mb_set_largest_free_order(sb, grp);
1230 	mb_update_avg_fragment_size(sb, grp);
1231 
1232 	clear_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &(grp->bb_state));
1233 
1234 	period = get_cycles() - period;
1235 	atomic_inc(&sbi->s_mb_buddies_generated);
1236 	atomic64_add(period, &sbi->s_mb_generation_time);
1237 }
1238 
mb_regenerate_buddy(struct ext4_buddy * e4b)1239 static void mb_regenerate_buddy(struct ext4_buddy *e4b)
1240 {
1241 	int count;
1242 	int order = 1;
1243 	void *buddy;
1244 
1245 	while ((buddy = mb_find_buddy(e4b, order++, &count)))
1246 		mb_set_bits(buddy, 0, count);
1247 
1248 	e4b->bd_info->bb_fragments = 0;
1249 	memset(e4b->bd_info->bb_counters, 0,
1250 		sizeof(*e4b->bd_info->bb_counters) *
1251 		(e4b->bd_sb->s_blocksize_bits + 2));
1252 
1253 	ext4_mb_generate_buddy(e4b->bd_sb, e4b->bd_buddy,
1254 		e4b->bd_bitmap, e4b->bd_group, e4b->bd_info);
1255 }
1256 
1257 /* The buddy information is attached the buddy cache inode
1258  * for convenience. The information regarding each group
1259  * is loaded via ext4_mb_load_buddy. The information involve
1260  * block bitmap and buddy information. The information are
1261  * stored in the inode as
1262  *
1263  * {                        page                        }
1264  * [ group 0 bitmap][ group 0 buddy] [group 1][ group 1]...
1265  *
1266  *
1267  * one block each for bitmap and buddy information.
1268  * So for each group we take up 2 blocks. A page can
1269  * contain blocks_per_page (PAGE_SIZE / blocksize)  blocks.
1270  * So it can have information regarding groups_per_page which
1271  * is blocks_per_page/2
1272  *
1273  * Locking note:  This routine takes the block group lock of all groups
1274  * for this page; do not hold this lock when calling this routine!
1275  */
1276 
ext4_mb_init_cache(struct folio * folio,char * incore,gfp_t gfp)1277 static int ext4_mb_init_cache(struct folio *folio, char *incore, gfp_t gfp)
1278 {
1279 	ext4_group_t ngroups;
1280 	unsigned int blocksize;
1281 	int blocks_per_page;
1282 	int groups_per_page;
1283 	int err = 0;
1284 	int i;
1285 	ext4_group_t first_group, group;
1286 	int first_block;
1287 	struct super_block *sb;
1288 	struct buffer_head *bhs;
1289 	struct buffer_head **bh = NULL;
1290 	struct inode *inode;
1291 	char *data;
1292 	char *bitmap;
1293 	struct ext4_group_info *grinfo;
1294 
1295 	inode = folio->mapping->host;
1296 	sb = inode->i_sb;
1297 	ngroups = ext4_get_groups_count(sb);
1298 	blocksize = i_blocksize(inode);
1299 	blocks_per_page = PAGE_SIZE / blocksize;
1300 
1301 	mb_debug(sb, "init folio %lu\n", folio->index);
1302 
1303 	groups_per_page = blocks_per_page >> 1;
1304 	if (groups_per_page == 0)
1305 		groups_per_page = 1;
1306 
1307 	/* allocate buffer_heads to read bitmaps */
1308 	if (groups_per_page > 1) {
1309 		i = sizeof(struct buffer_head *) * groups_per_page;
1310 		bh = kzalloc(i, gfp);
1311 		if (bh == NULL)
1312 			return -ENOMEM;
1313 	} else
1314 		bh = &bhs;
1315 
1316 	first_group = folio->index * blocks_per_page / 2;
1317 
1318 	/* read all groups the folio covers into the cache */
1319 	for (i = 0, group = first_group; i < groups_per_page; i++, group++) {
1320 		if (group >= ngroups)
1321 			break;
1322 
1323 		grinfo = ext4_get_group_info(sb, group);
1324 		if (!grinfo)
1325 			continue;
1326 		/*
1327 		 * If page is uptodate then we came here after online resize
1328 		 * which added some new uninitialized group info structs, so
1329 		 * we must skip all initialized uptodate buddies on the folio,
1330 		 * which may be currently in use by an allocating task.
1331 		 */
1332 		if (folio_test_uptodate(folio) &&
1333 				!EXT4_MB_GRP_NEED_INIT(grinfo)) {
1334 			bh[i] = NULL;
1335 			continue;
1336 		}
1337 		bh[i] = ext4_read_block_bitmap_nowait(sb, group, false);
1338 		if (IS_ERR(bh[i])) {
1339 			err = PTR_ERR(bh[i]);
1340 			bh[i] = NULL;
1341 			goto out;
1342 		}
1343 		mb_debug(sb, "read bitmap for group %u\n", group);
1344 	}
1345 
1346 	/* wait for I/O completion */
1347 	for (i = 0, group = first_group; i < groups_per_page; i++, group++) {
1348 		int err2;
1349 
1350 		if (!bh[i])
1351 			continue;
1352 		err2 = ext4_wait_block_bitmap(sb, group, bh[i]);
1353 		if (!err)
1354 			err = err2;
1355 	}
1356 
1357 	first_block = folio->index * blocks_per_page;
1358 	for (i = 0; i < blocks_per_page; i++) {
1359 		group = (first_block + i) >> 1;
1360 		if (group >= ngroups)
1361 			break;
1362 
1363 		if (!bh[group - first_group])
1364 			/* skip initialized uptodate buddy */
1365 			continue;
1366 
1367 		if (!buffer_verified(bh[group - first_group]))
1368 			/* Skip faulty bitmaps */
1369 			continue;
1370 		err = 0;
1371 
1372 		/*
1373 		 * data carry information regarding this
1374 		 * particular group in the format specified
1375 		 * above
1376 		 *
1377 		 */
1378 		data = folio_address(folio) + (i * blocksize);
1379 		bitmap = bh[group - first_group]->b_data;
1380 
1381 		/*
1382 		 * We place the buddy block and bitmap block
1383 		 * close together
1384 		 */
1385 		grinfo = ext4_get_group_info(sb, group);
1386 		if (!grinfo) {
1387 			err = -EFSCORRUPTED;
1388 		        goto out;
1389 		}
1390 		if ((first_block + i) & 1) {
1391 			/* this is block of buddy */
1392 			BUG_ON(incore == NULL);
1393 			mb_debug(sb, "put buddy for group %u in folio %lu/%x\n",
1394 				group, folio->index, i * blocksize);
1395 			trace_ext4_mb_buddy_bitmap_load(sb, group);
1396 			grinfo->bb_fragments = 0;
1397 			memset(grinfo->bb_counters, 0,
1398 			       sizeof(*grinfo->bb_counters) *
1399 			       (MB_NUM_ORDERS(sb)));
1400 			/*
1401 			 * incore got set to the group block bitmap below
1402 			 */
1403 			ext4_lock_group(sb, group);
1404 			/* init the buddy */
1405 			memset(data, 0xff, blocksize);
1406 			ext4_mb_generate_buddy(sb, data, incore, group, grinfo);
1407 			ext4_unlock_group(sb, group);
1408 			incore = NULL;
1409 		} else {
1410 			/* this is block of bitmap */
1411 			BUG_ON(incore != NULL);
1412 			mb_debug(sb, "put bitmap for group %u in folio %lu/%x\n",
1413 				group, folio->index, i * blocksize);
1414 			trace_ext4_mb_bitmap_load(sb, group);
1415 
1416 			/* see comments in ext4_mb_put_pa() */
1417 			ext4_lock_group(sb, group);
1418 			memcpy(data, bitmap, blocksize);
1419 
1420 			/* mark all preallocated blks used in in-core bitmap */
1421 			ext4_mb_generate_from_pa(sb, data, group);
1422 			WARN_ON_ONCE(!RB_EMPTY_ROOT(&grinfo->bb_free_root));
1423 			ext4_unlock_group(sb, group);
1424 
1425 			/* set incore so that the buddy information can be
1426 			 * generated using this
1427 			 */
1428 			incore = data;
1429 		}
1430 	}
1431 	folio_mark_uptodate(folio);
1432 
1433 out:
1434 	if (bh) {
1435 		for (i = 0; i < groups_per_page; i++)
1436 			brelse(bh[i]);
1437 		if (bh != &bhs)
1438 			kfree(bh);
1439 	}
1440 	return err;
1441 }
1442 
1443 /*
1444  * Lock the buddy and bitmap pages. This make sure other parallel init_group
1445  * on the same buddy page doesn't happen whild holding the buddy page lock.
1446  * Return locked buddy and bitmap pages on e4b struct. If buddy and bitmap
1447  * are on the same page e4b->bd_buddy_folio is NULL and return value is 0.
1448  */
ext4_mb_get_buddy_page_lock(struct super_block * sb,ext4_group_t group,struct ext4_buddy * e4b,gfp_t gfp)1449 static int ext4_mb_get_buddy_page_lock(struct super_block *sb,
1450 		ext4_group_t group, struct ext4_buddy *e4b, gfp_t gfp)
1451 {
1452 	struct inode *inode = EXT4_SB(sb)->s_buddy_cache;
1453 	int block, pnum, poff;
1454 	int blocks_per_page;
1455 	struct folio *folio;
1456 
1457 	e4b->bd_buddy_folio = NULL;
1458 	e4b->bd_bitmap_folio = NULL;
1459 
1460 	blocks_per_page = PAGE_SIZE / sb->s_blocksize;
1461 	/*
1462 	 * the buddy cache inode stores the block bitmap
1463 	 * and buddy information in consecutive blocks.
1464 	 * So for each group we need two blocks.
1465 	 */
1466 	block = group * 2;
1467 	pnum = block / blocks_per_page;
1468 	poff = block % blocks_per_page;
1469 	folio = __filemap_get_folio(inode->i_mapping, pnum,
1470 			FGP_LOCK | FGP_ACCESSED | FGP_CREAT, gfp);
1471 	if (IS_ERR(folio))
1472 		return PTR_ERR(folio);
1473 	BUG_ON(folio->mapping != inode->i_mapping);
1474 	e4b->bd_bitmap_folio = folio;
1475 	e4b->bd_bitmap = folio_address(folio) + (poff * sb->s_blocksize);
1476 
1477 	if (blocks_per_page >= 2) {
1478 		/* buddy and bitmap are on the same page */
1479 		return 0;
1480 	}
1481 
1482 	/* blocks_per_page == 1, hence we need another page for the buddy */
1483 	folio = __filemap_get_folio(inode->i_mapping, block + 1,
1484 			FGP_LOCK | FGP_ACCESSED | FGP_CREAT, gfp);
1485 	if (IS_ERR(folio))
1486 		return PTR_ERR(folio);
1487 	BUG_ON(folio->mapping != inode->i_mapping);
1488 	e4b->bd_buddy_folio = folio;
1489 	return 0;
1490 }
1491 
ext4_mb_put_buddy_page_lock(struct ext4_buddy * e4b)1492 static void ext4_mb_put_buddy_page_lock(struct ext4_buddy *e4b)
1493 {
1494 	if (e4b->bd_bitmap_folio) {
1495 		folio_unlock(e4b->bd_bitmap_folio);
1496 		folio_put(e4b->bd_bitmap_folio);
1497 	}
1498 	if (e4b->bd_buddy_folio) {
1499 		folio_unlock(e4b->bd_buddy_folio);
1500 		folio_put(e4b->bd_buddy_folio);
1501 	}
1502 }
1503 
1504 /*
1505  * Locking note:  This routine calls ext4_mb_init_cache(), which takes the
1506  * block group lock of all groups for this page; do not hold the BG lock when
1507  * calling this routine!
1508  */
1509 static noinline_for_stack
ext4_mb_init_group(struct super_block * sb,ext4_group_t group,gfp_t gfp)1510 int ext4_mb_init_group(struct super_block *sb, ext4_group_t group, gfp_t gfp)
1511 {
1512 
1513 	struct ext4_group_info *this_grp;
1514 	struct ext4_buddy e4b;
1515 	struct folio *folio;
1516 	int ret = 0;
1517 
1518 	might_sleep();
1519 	mb_debug(sb, "init group %u\n", group);
1520 	this_grp = ext4_get_group_info(sb, group);
1521 	if (!this_grp)
1522 		return -EFSCORRUPTED;
1523 
1524 	/*
1525 	 * This ensures that we don't reinit the buddy cache
1526 	 * page which map to the group from which we are already
1527 	 * allocating. If we are looking at the buddy cache we would
1528 	 * have taken a reference using ext4_mb_load_buddy and that
1529 	 * would have pinned buddy page to page cache.
1530 	 * The call to ext4_mb_get_buddy_page_lock will mark the
1531 	 * page accessed.
1532 	 */
1533 	ret = ext4_mb_get_buddy_page_lock(sb, group, &e4b, gfp);
1534 	if (ret || !EXT4_MB_GRP_NEED_INIT(this_grp)) {
1535 		/*
1536 		 * somebody initialized the group
1537 		 * return without doing anything
1538 		 */
1539 		goto err;
1540 	}
1541 
1542 	folio = e4b.bd_bitmap_folio;
1543 	ret = ext4_mb_init_cache(folio, NULL, gfp);
1544 	if (ret)
1545 		goto err;
1546 	if (!folio_test_uptodate(folio)) {
1547 		ret = -EIO;
1548 		goto err;
1549 	}
1550 
1551 	if (e4b.bd_buddy_folio == NULL) {
1552 		/*
1553 		 * If both the bitmap and buddy are in
1554 		 * the same page we don't need to force
1555 		 * init the buddy
1556 		 */
1557 		ret = 0;
1558 		goto err;
1559 	}
1560 	/* init buddy cache */
1561 	folio = e4b.bd_buddy_folio;
1562 	ret = ext4_mb_init_cache(folio, e4b.bd_bitmap, gfp);
1563 	if (ret)
1564 		goto err;
1565 	if (!folio_test_uptodate(folio)) {
1566 		ret = -EIO;
1567 		goto err;
1568 	}
1569 err:
1570 	ext4_mb_put_buddy_page_lock(&e4b);
1571 	return ret;
1572 }
1573 
1574 /*
1575  * Locking note:  This routine calls ext4_mb_init_cache(), which takes the
1576  * block group lock of all groups for this page; do not hold the BG lock when
1577  * calling this routine!
1578  */
1579 static noinline_for_stack int
ext4_mb_load_buddy_gfp(struct super_block * sb,ext4_group_t group,struct ext4_buddy * e4b,gfp_t gfp)1580 ext4_mb_load_buddy_gfp(struct super_block *sb, ext4_group_t group,
1581 		       struct ext4_buddy *e4b, gfp_t gfp)
1582 {
1583 	int blocks_per_page;
1584 	int block;
1585 	int pnum;
1586 	int poff;
1587 	struct folio *folio;
1588 	int ret;
1589 	struct ext4_group_info *grp;
1590 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1591 	struct inode *inode = sbi->s_buddy_cache;
1592 
1593 	might_sleep();
1594 	mb_debug(sb, "load group %u\n", group);
1595 
1596 	blocks_per_page = PAGE_SIZE / sb->s_blocksize;
1597 	grp = ext4_get_group_info(sb, group);
1598 	if (!grp)
1599 		return -EFSCORRUPTED;
1600 
1601 	e4b->bd_blkbits = sb->s_blocksize_bits;
1602 	e4b->bd_info = grp;
1603 	e4b->bd_sb = sb;
1604 	e4b->bd_group = group;
1605 	e4b->bd_buddy_folio = NULL;
1606 	e4b->bd_bitmap_folio = NULL;
1607 
1608 	if (unlikely(EXT4_MB_GRP_NEED_INIT(grp))) {
1609 		/*
1610 		 * we need full data about the group
1611 		 * to make a good selection
1612 		 */
1613 		ret = ext4_mb_init_group(sb, group, gfp);
1614 		if (ret)
1615 			return ret;
1616 	}
1617 
1618 	/*
1619 	 * the buddy cache inode stores the block bitmap
1620 	 * and buddy information in consecutive blocks.
1621 	 * So for each group we need two blocks.
1622 	 */
1623 	block = group * 2;
1624 	pnum = block / blocks_per_page;
1625 	poff = block % blocks_per_page;
1626 
1627 	/* Avoid locking the folio in the fast path ... */
1628 	folio = __filemap_get_folio(inode->i_mapping, pnum, FGP_ACCESSED, 0);
1629 	if (IS_ERR(folio) || !folio_test_uptodate(folio)) {
1630 		if (!IS_ERR(folio))
1631 			/*
1632 			 * drop the folio reference and try
1633 			 * to get the folio with lock. If we
1634 			 * are not uptodate that implies
1635 			 * somebody just created the folio but
1636 			 * is yet to initialize it. So
1637 			 * wait for it to initialize.
1638 			 */
1639 			folio_put(folio);
1640 		folio = __filemap_get_folio(inode->i_mapping, pnum,
1641 				FGP_LOCK | FGP_ACCESSED | FGP_CREAT, gfp);
1642 		if (!IS_ERR(folio)) {
1643 			if (WARN_RATELIMIT(folio->mapping != inode->i_mapping,
1644 	"ext4: bitmap's mapping != inode->i_mapping\n")) {
1645 				/* should never happen */
1646 				folio_unlock(folio);
1647 				ret = -EINVAL;
1648 				goto err;
1649 			}
1650 			if (!folio_test_uptodate(folio)) {
1651 				ret = ext4_mb_init_cache(folio, NULL, gfp);
1652 				if (ret) {
1653 					folio_unlock(folio);
1654 					goto err;
1655 				}
1656 				mb_cmp_bitmaps(e4b, folio_address(folio) +
1657 					       (poff * sb->s_blocksize));
1658 			}
1659 			folio_unlock(folio);
1660 		}
1661 	}
1662 	if (IS_ERR(folio)) {
1663 		ret = PTR_ERR(folio);
1664 		goto err;
1665 	}
1666 	if (!folio_test_uptodate(folio)) {
1667 		ret = -EIO;
1668 		goto err;
1669 	}
1670 
1671 	/* Folios marked accessed already */
1672 	e4b->bd_bitmap_folio = folio;
1673 	e4b->bd_bitmap = folio_address(folio) + (poff * sb->s_blocksize);
1674 
1675 	block++;
1676 	pnum = block / blocks_per_page;
1677 	poff = block % blocks_per_page;
1678 
1679 	folio = __filemap_get_folio(inode->i_mapping, pnum, FGP_ACCESSED, 0);
1680 	if (IS_ERR(folio) || !folio_test_uptodate(folio)) {
1681 		if (!IS_ERR(folio))
1682 			folio_put(folio);
1683 		folio = __filemap_get_folio(inode->i_mapping, pnum,
1684 				FGP_LOCK | FGP_ACCESSED | FGP_CREAT, gfp);
1685 		if (!IS_ERR(folio)) {
1686 			if (WARN_RATELIMIT(folio->mapping != inode->i_mapping,
1687 	"ext4: buddy bitmap's mapping != inode->i_mapping\n")) {
1688 				/* should never happen */
1689 				folio_unlock(folio);
1690 				ret = -EINVAL;
1691 				goto err;
1692 			}
1693 			if (!folio_test_uptodate(folio)) {
1694 				ret = ext4_mb_init_cache(folio, e4b->bd_bitmap,
1695 							 gfp);
1696 				if (ret) {
1697 					folio_unlock(folio);
1698 					goto err;
1699 				}
1700 			}
1701 			folio_unlock(folio);
1702 		}
1703 	}
1704 	if (IS_ERR(folio)) {
1705 		ret = PTR_ERR(folio);
1706 		goto err;
1707 	}
1708 	if (!folio_test_uptodate(folio)) {
1709 		ret = -EIO;
1710 		goto err;
1711 	}
1712 
1713 	/* Folios marked accessed already */
1714 	e4b->bd_buddy_folio = folio;
1715 	e4b->bd_buddy = folio_address(folio) + (poff * sb->s_blocksize);
1716 
1717 	return 0;
1718 
1719 err:
1720 	if (!IS_ERR_OR_NULL(folio))
1721 		folio_put(folio);
1722 	if (e4b->bd_bitmap_folio)
1723 		folio_put(e4b->bd_bitmap_folio);
1724 
1725 	e4b->bd_buddy = NULL;
1726 	e4b->bd_bitmap = NULL;
1727 	return ret;
1728 }
1729 
ext4_mb_load_buddy(struct super_block * sb,ext4_group_t group,struct ext4_buddy * e4b)1730 static int ext4_mb_load_buddy(struct super_block *sb, ext4_group_t group,
1731 			      struct ext4_buddy *e4b)
1732 {
1733 	return ext4_mb_load_buddy_gfp(sb, group, e4b, GFP_NOFS);
1734 }
1735 
ext4_mb_unload_buddy(struct ext4_buddy * e4b)1736 static void ext4_mb_unload_buddy(struct ext4_buddy *e4b)
1737 {
1738 	if (e4b->bd_bitmap_folio)
1739 		folio_put(e4b->bd_bitmap_folio);
1740 	if (e4b->bd_buddy_folio)
1741 		folio_put(e4b->bd_buddy_folio);
1742 }
1743 
1744 
mb_find_order_for_block(struct ext4_buddy * e4b,int block)1745 static int mb_find_order_for_block(struct ext4_buddy *e4b, int block)
1746 {
1747 	int order = 1, max;
1748 	void *bb;
1749 
1750 	BUG_ON(e4b->bd_bitmap == e4b->bd_buddy);
1751 	BUG_ON(block >= (1 << (e4b->bd_blkbits + 3)));
1752 
1753 	while (order <= e4b->bd_blkbits + 1) {
1754 		bb = mb_find_buddy(e4b, order, &max);
1755 		if (!mb_test_bit(block >> order, bb)) {
1756 			/* this block is part of buddy of order 'order' */
1757 			return order;
1758 		}
1759 		order++;
1760 	}
1761 	return 0;
1762 }
1763 
mb_clear_bits(void * bm,int cur,int len)1764 static void mb_clear_bits(void *bm, int cur, int len)
1765 {
1766 	__u32 *addr;
1767 
1768 	len = cur + len;
1769 	while (cur < len) {
1770 		if ((cur & 31) == 0 && (len - cur) >= 32) {
1771 			/* fast path: clear whole word at once */
1772 			addr = bm + (cur >> 3);
1773 			*addr = 0;
1774 			cur += 32;
1775 			continue;
1776 		}
1777 		mb_clear_bit(cur, bm);
1778 		cur++;
1779 	}
1780 }
1781 
1782 /* clear bits in given range
1783  * will return first found zero bit if any, -1 otherwise
1784  */
mb_test_and_clear_bits(void * bm,int cur,int len)1785 static int mb_test_and_clear_bits(void *bm, int cur, int len)
1786 {
1787 	__u32 *addr;
1788 	int zero_bit = -1;
1789 
1790 	len = cur + len;
1791 	while (cur < len) {
1792 		if ((cur & 31) == 0 && (len - cur) >= 32) {
1793 			/* fast path: clear whole word at once */
1794 			addr = bm + (cur >> 3);
1795 			if (*addr != (__u32)(-1) && zero_bit == -1)
1796 				zero_bit = cur + mb_find_next_zero_bit(addr, 32, 0);
1797 			*addr = 0;
1798 			cur += 32;
1799 			continue;
1800 		}
1801 		if (!mb_test_and_clear_bit(cur, bm) && zero_bit == -1)
1802 			zero_bit = cur;
1803 		cur++;
1804 	}
1805 
1806 	return zero_bit;
1807 }
1808 
mb_set_bits(void * bm,int cur,int len)1809 void mb_set_bits(void *bm, int cur, int len)
1810 {
1811 	__u32 *addr;
1812 
1813 	len = cur + len;
1814 	while (cur < len) {
1815 		if ((cur & 31) == 0 && (len - cur) >= 32) {
1816 			/* fast path: set whole word at once */
1817 			addr = bm + (cur >> 3);
1818 			*addr = 0xffffffff;
1819 			cur += 32;
1820 			continue;
1821 		}
1822 		mb_set_bit(cur, bm);
1823 		cur++;
1824 	}
1825 }
1826 
mb_buddy_adjust_border(int * bit,void * bitmap,int side)1827 static inline int mb_buddy_adjust_border(int* bit, void* bitmap, int side)
1828 {
1829 	if (mb_test_bit(*bit + side, bitmap)) {
1830 		mb_clear_bit(*bit, bitmap);
1831 		(*bit) -= side;
1832 		return 1;
1833 	}
1834 	else {
1835 		(*bit) += side;
1836 		mb_set_bit(*bit, bitmap);
1837 		return -1;
1838 	}
1839 }
1840 
mb_buddy_mark_free(struct ext4_buddy * e4b,int first,int last)1841 static void mb_buddy_mark_free(struct ext4_buddy *e4b, int first, int last)
1842 {
1843 	int max;
1844 	int order = 1;
1845 	void *buddy = mb_find_buddy(e4b, order, &max);
1846 
1847 	while (buddy) {
1848 		void *buddy2;
1849 
1850 		/* Bits in range [first; last] are known to be set since
1851 		 * corresponding blocks were allocated. Bits in range
1852 		 * (first; last) will stay set because they form buddies on
1853 		 * upper layer. We just deal with borders if they don't
1854 		 * align with upper layer and then go up.
1855 		 * Releasing entire group is all about clearing
1856 		 * single bit of highest order buddy.
1857 		 */
1858 
1859 		/* Example:
1860 		 * ---------------------------------
1861 		 * |   1   |   1   |   1   |   1   |
1862 		 * ---------------------------------
1863 		 * | 0 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
1864 		 * ---------------------------------
1865 		 *   0   1   2   3   4   5   6   7
1866 		 *      \_____________________/
1867 		 *
1868 		 * Neither [1] nor [6] is aligned to above layer.
1869 		 * Left neighbour [0] is free, so mark it busy,
1870 		 * decrease bb_counters and extend range to
1871 		 * [0; 6]
1872 		 * Right neighbour [7] is busy. It can't be coaleasced with [6], so
1873 		 * mark [6] free, increase bb_counters and shrink range to
1874 		 * [0; 5].
1875 		 * Then shift range to [0; 2], go up and do the same.
1876 		 */
1877 
1878 
1879 		if (first & 1)
1880 			e4b->bd_info->bb_counters[order] += mb_buddy_adjust_border(&first, buddy, -1);
1881 		if (!(last & 1))
1882 			e4b->bd_info->bb_counters[order] += mb_buddy_adjust_border(&last, buddy, 1);
1883 		if (first > last)
1884 			break;
1885 		order++;
1886 
1887 		buddy2 = mb_find_buddy(e4b, order, &max);
1888 		if (!buddy2) {
1889 			mb_clear_bits(buddy, first, last - first + 1);
1890 			e4b->bd_info->bb_counters[order - 1] += last - first + 1;
1891 			break;
1892 		}
1893 		first >>= 1;
1894 		last >>= 1;
1895 		buddy = buddy2;
1896 	}
1897 }
1898 
mb_free_blocks(struct inode * inode,struct ext4_buddy * e4b,int first,int count)1899 static void mb_free_blocks(struct inode *inode, struct ext4_buddy *e4b,
1900 			   int first, int count)
1901 {
1902 	int left_is_free = 0;
1903 	int right_is_free = 0;
1904 	int block;
1905 	int last = first + count - 1;
1906 	struct super_block *sb = e4b->bd_sb;
1907 
1908 	if (WARN_ON(count == 0))
1909 		return;
1910 	BUG_ON(last >= (sb->s_blocksize << 3));
1911 	assert_spin_locked(ext4_group_lock_ptr(sb, e4b->bd_group));
1912 	/* Don't bother if the block group is corrupt. */
1913 	if (unlikely(EXT4_MB_GRP_BBITMAP_CORRUPT(e4b->bd_info)))
1914 		return;
1915 
1916 	mb_check_buddy(e4b);
1917 	mb_free_blocks_double(inode, e4b, first, count);
1918 
1919 	/* access memory sequentially: check left neighbour,
1920 	 * clear range and then check right neighbour
1921 	 */
1922 	if (first != 0)
1923 		left_is_free = !mb_test_bit(first - 1, e4b->bd_bitmap);
1924 	block = mb_test_and_clear_bits(e4b->bd_bitmap, first, count);
1925 	if (last + 1 < EXT4_SB(sb)->s_mb_maxs[0])
1926 		right_is_free = !mb_test_bit(last + 1, e4b->bd_bitmap);
1927 
1928 	if (unlikely(block != -1)) {
1929 		struct ext4_sb_info *sbi = EXT4_SB(sb);
1930 		ext4_fsblk_t blocknr;
1931 
1932 		/*
1933 		 * Fastcommit replay can free already freed blocks which
1934 		 * corrupts allocation info. Regenerate it.
1935 		 */
1936 		if (sbi->s_mount_state & EXT4_FC_REPLAY) {
1937 			mb_regenerate_buddy(e4b);
1938 			goto check;
1939 		}
1940 
1941 		blocknr = ext4_group_first_block_no(sb, e4b->bd_group);
1942 		blocknr += EXT4_C2B(sbi, block);
1943 		ext4_mark_group_bitmap_corrupted(sb, e4b->bd_group,
1944 				EXT4_GROUP_INFO_BBITMAP_CORRUPT);
1945 		ext4_grp_locked_error(sb, e4b->bd_group,
1946 				      inode ? inode->i_ino : 0, blocknr,
1947 				      "freeing already freed block (bit %u); block bitmap corrupt.",
1948 				      block);
1949 		return;
1950 	}
1951 
1952 	this_cpu_inc(discard_pa_seq);
1953 	e4b->bd_info->bb_free += count;
1954 	if (first < e4b->bd_info->bb_first_free)
1955 		e4b->bd_info->bb_first_free = first;
1956 
1957 	/* let's maintain fragments counter */
1958 	if (left_is_free && right_is_free)
1959 		e4b->bd_info->bb_fragments--;
1960 	else if (!left_is_free && !right_is_free)
1961 		e4b->bd_info->bb_fragments++;
1962 
1963 	/* buddy[0] == bd_bitmap is a special case, so handle
1964 	 * it right away and let mb_buddy_mark_free stay free of
1965 	 * zero order checks.
1966 	 * Check if neighbours are to be coaleasced,
1967 	 * adjust bitmap bb_counters and borders appropriately.
1968 	 */
1969 	if (first & 1) {
1970 		first += !left_is_free;
1971 		e4b->bd_info->bb_counters[0] += left_is_free ? -1 : 1;
1972 	}
1973 	if (!(last & 1)) {
1974 		last -= !right_is_free;
1975 		e4b->bd_info->bb_counters[0] += right_is_free ? -1 : 1;
1976 	}
1977 
1978 	if (first <= last)
1979 		mb_buddy_mark_free(e4b, first >> 1, last >> 1);
1980 
1981 	mb_set_largest_free_order(sb, e4b->bd_info);
1982 	mb_update_avg_fragment_size(sb, e4b->bd_info);
1983 check:
1984 	mb_check_buddy(e4b);
1985 }
1986 
mb_find_extent(struct ext4_buddy * e4b,int block,int needed,struct ext4_free_extent * ex)1987 static int mb_find_extent(struct ext4_buddy *e4b, int block,
1988 				int needed, struct ext4_free_extent *ex)
1989 {
1990 	int max, order, next;
1991 	void *buddy;
1992 
1993 	assert_spin_locked(ext4_group_lock_ptr(e4b->bd_sb, e4b->bd_group));
1994 	BUG_ON(ex == NULL);
1995 
1996 	buddy = mb_find_buddy(e4b, 0, &max);
1997 	BUG_ON(buddy == NULL);
1998 	BUG_ON(block >= max);
1999 	if (mb_test_bit(block, buddy)) {
2000 		ex->fe_len = 0;
2001 		ex->fe_start = 0;
2002 		ex->fe_group = 0;
2003 		return 0;
2004 	}
2005 
2006 	/* find actual order */
2007 	order = mb_find_order_for_block(e4b, block);
2008 
2009 	ex->fe_len = (1 << order) - (block & ((1 << order) - 1));
2010 	ex->fe_start = block;
2011 	ex->fe_group = e4b->bd_group;
2012 
2013 	block = block >> order;
2014 
2015 	while (needed > ex->fe_len &&
2016 	       mb_find_buddy(e4b, order, &max)) {
2017 
2018 		if (block + 1 >= max)
2019 			break;
2020 
2021 		next = (block + 1) * (1 << order);
2022 		if (mb_test_bit(next, e4b->bd_bitmap))
2023 			break;
2024 
2025 		order = mb_find_order_for_block(e4b, next);
2026 
2027 		block = next >> order;
2028 		ex->fe_len += 1 << order;
2029 	}
2030 
2031 	if (ex->fe_start + ex->fe_len > EXT4_CLUSTERS_PER_GROUP(e4b->bd_sb)) {
2032 		/* Should never happen! (but apparently sometimes does?!?) */
2033 		WARN_ON(1);
2034 		ext4_grp_locked_error(e4b->bd_sb, e4b->bd_group, 0, 0,
2035 			"corruption or bug in mb_find_extent "
2036 			"block=%d, order=%d needed=%d ex=%u/%d/%d@%u",
2037 			block, order, needed, ex->fe_group, ex->fe_start,
2038 			ex->fe_len, ex->fe_logical);
2039 		ex->fe_len = 0;
2040 		ex->fe_start = 0;
2041 		ex->fe_group = 0;
2042 	}
2043 	return ex->fe_len;
2044 }
2045 
mb_mark_used(struct ext4_buddy * e4b,struct ext4_free_extent * ex)2046 static int mb_mark_used(struct ext4_buddy *e4b, struct ext4_free_extent *ex)
2047 {
2048 	int ord;
2049 	int mlen = 0;
2050 	int max = 0;
2051 	int start = ex->fe_start;
2052 	int len = ex->fe_len;
2053 	unsigned ret = 0;
2054 	int len0 = len;
2055 	void *buddy;
2056 	int ord_start, ord_end;
2057 
2058 	BUG_ON(start + len > (e4b->bd_sb->s_blocksize << 3));
2059 	BUG_ON(e4b->bd_group != ex->fe_group);
2060 	assert_spin_locked(ext4_group_lock_ptr(e4b->bd_sb, e4b->bd_group));
2061 	mb_check_buddy(e4b);
2062 	mb_mark_used_double(e4b, start, len);
2063 
2064 	this_cpu_inc(discard_pa_seq);
2065 	e4b->bd_info->bb_free -= len;
2066 	if (e4b->bd_info->bb_first_free == start)
2067 		e4b->bd_info->bb_first_free += len;
2068 
2069 	/* let's maintain fragments counter */
2070 	if (start != 0)
2071 		mlen = !mb_test_bit(start - 1, e4b->bd_bitmap);
2072 	if (start + len < EXT4_SB(e4b->bd_sb)->s_mb_maxs[0])
2073 		max = !mb_test_bit(start + len, e4b->bd_bitmap);
2074 	if (mlen && max)
2075 		e4b->bd_info->bb_fragments++;
2076 	else if (!mlen && !max)
2077 		e4b->bd_info->bb_fragments--;
2078 
2079 	/* let's maintain buddy itself */
2080 	while (len) {
2081 		ord = mb_find_order_for_block(e4b, start);
2082 
2083 		if (((start >> ord) << ord) == start && len >= (1 << ord)) {
2084 			/* the whole chunk may be allocated at once! */
2085 			mlen = 1 << ord;
2086 			buddy = mb_find_buddy(e4b, ord, &max);
2087 			BUG_ON((start >> ord) >= max);
2088 			mb_set_bit(start >> ord, buddy);
2089 			e4b->bd_info->bb_counters[ord]--;
2090 			start += mlen;
2091 			len -= mlen;
2092 			BUG_ON(len < 0);
2093 			continue;
2094 		}
2095 
2096 		/* store for history */
2097 		if (ret == 0)
2098 			ret = len | (ord << 16);
2099 
2100 		BUG_ON(ord <= 0);
2101 		buddy = mb_find_buddy(e4b, ord, &max);
2102 		mb_set_bit(start >> ord, buddy);
2103 		e4b->bd_info->bb_counters[ord]--;
2104 
2105 		ord_start = (start >> ord) << ord;
2106 		ord_end = ord_start + (1 << ord);
2107 		/* first chunk */
2108 		if (start > ord_start)
2109 			ext4_mb_mark_free_simple(e4b->bd_sb, e4b->bd_buddy,
2110 						 ord_start, start - ord_start,
2111 						 e4b->bd_info);
2112 
2113 		/* last chunk */
2114 		if (start + len < ord_end) {
2115 			ext4_mb_mark_free_simple(e4b->bd_sb, e4b->bd_buddy,
2116 						 start + len,
2117 						 ord_end - (start + len),
2118 						 e4b->bd_info);
2119 			break;
2120 		}
2121 		len = start + len - ord_end;
2122 		start = ord_end;
2123 	}
2124 	mb_set_largest_free_order(e4b->bd_sb, e4b->bd_info);
2125 
2126 	mb_update_avg_fragment_size(e4b->bd_sb, e4b->bd_info);
2127 	mb_set_bits(e4b->bd_bitmap, ex->fe_start, len0);
2128 	mb_check_buddy(e4b);
2129 
2130 	return ret;
2131 }
2132 
2133 /*
2134  * Must be called under group lock!
2135  */
ext4_mb_use_best_found(struct ext4_allocation_context * ac,struct ext4_buddy * e4b)2136 static void ext4_mb_use_best_found(struct ext4_allocation_context *ac,
2137 					struct ext4_buddy *e4b)
2138 {
2139 	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
2140 	int ret;
2141 
2142 	BUG_ON(ac->ac_b_ex.fe_group != e4b->bd_group);
2143 	BUG_ON(ac->ac_status == AC_STATUS_FOUND);
2144 
2145 	ac->ac_b_ex.fe_len = min(ac->ac_b_ex.fe_len, ac->ac_g_ex.fe_len);
2146 	ac->ac_b_ex.fe_logical = ac->ac_g_ex.fe_logical;
2147 	ret = mb_mark_used(e4b, &ac->ac_b_ex);
2148 
2149 	/* preallocation can change ac_b_ex, thus we store actually
2150 	 * allocated blocks for history */
2151 	ac->ac_f_ex = ac->ac_b_ex;
2152 
2153 	ac->ac_status = AC_STATUS_FOUND;
2154 	ac->ac_tail = ret & 0xffff;
2155 	ac->ac_buddy = ret >> 16;
2156 
2157 	/*
2158 	 * take the page reference. We want the page to be pinned
2159 	 * so that we don't get a ext4_mb_init_cache_call for this
2160 	 * group until we update the bitmap. That would mean we
2161 	 * double allocate blocks. The reference is dropped
2162 	 * in ext4_mb_release_context
2163 	 */
2164 	ac->ac_bitmap_folio = e4b->bd_bitmap_folio;
2165 	folio_get(ac->ac_bitmap_folio);
2166 	ac->ac_buddy_folio = e4b->bd_buddy_folio;
2167 	folio_get(ac->ac_buddy_folio);
2168 	/* store last allocated for subsequent stream allocation */
2169 	if (ac->ac_flags & EXT4_MB_STREAM_ALLOC) {
2170 		spin_lock(&sbi->s_md_lock);
2171 		sbi->s_mb_last_group = ac->ac_f_ex.fe_group;
2172 		sbi->s_mb_last_start = ac->ac_f_ex.fe_start;
2173 		spin_unlock(&sbi->s_md_lock);
2174 	}
2175 	/*
2176 	 * As we've just preallocated more space than
2177 	 * user requested originally, we store allocated
2178 	 * space in a special descriptor.
2179 	 */
2180 	if (ac->ac_o_ex.fe_len < ac->ac_b_ex.fe_len)
2181 		ext4_mb_new_preallocation(ac);
2182 
2183 }
2184 
ext4_mb_check_limits(struct ext4_allocation_context * ac,struct ext4_buddy * e4b,int finish_group)2185 static void ext4_mb_check_limits(struct ext4_allocation_context *ac,
2186 					struct ext4_buddy *e4b,
2187 					int finish_group)
2188 {
2189 	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
2190 	struct ext4_free_extent *bex = &ac->ac_b_ex;
2191 	struct ext4_free_extent *gex = &ac->ac_g_ex;
2192 
2193 	if (ac->ac_status == AC_STATUS_FOUND)
2194 		return;
2195 	/*
2196 	 * We don't want to scan for a whole year
2197 	 */
2198 	if (ac->ac_found > sbi->s_mb_max_to_scan &&
2199 			!(ac->ac_flags & EXT4_MB_HINT_FIRST)) {
2200 		ac->ac_status = AC_STATUS_BREAK;
2201 		return;
2202 	}
2203 
2204 	/*
2205 	 * Haven't found good chunk so far, let's continue
2206 	 */
2207 	if (bex->fe_len < gex->fe_len)
2208 		return;
2209 
2210 	if (finish_group || ac->ac_found > sbi->s_mb_min_to_scan)
2211 		ext4_mb_use_best_found(ac, e4b);
2212 }
2213 
2214 /*
2215  * The routine checks whether found extent is good enough. If it is,
2216  * then the extent gets marked used and flag is set to the context
2217  * to stop scanning. Otherwise, the extent is compared with the
2218  * previous found extent and if new one is better, then it's stored
2219  * in the context. Later, the best found extent will be used, if
2220  * mballoc can't find good enough extent.
2221  *
2222  * The algorithm used is roughly as follows:
2223  *
2224  * * If free extent found is exactly as big as goal, then
2225  *   stop the scan and use it immediately
2226  *
2227  * * If free extent found is smaller than goal, then keep retrying
2228  *   upto a max of sbi->s_mb_max_to_scan times (default 200). After
2229  *   that stop scanning and use whatever we have.
2230  *
2231  * * If free extent found is bigger than goal, then keep retrying
2232  *   upto a max of sbi->s_mb_min_to_scan times (default 10) before
2233  *   stopping the scan and using the extent.
2234  *
2235  *
2236  * FIXME: real allocation policy is to be designed yet!
2237  */
ext4_mb_measure_extent(struct ext4_allocation_context * ac,struct ext4_free_extent * ex,struct ext4_buddy * e4b)2238 static void ext4_mb_measure_extent(struct ext4_allocation_context *ac,
2239 					struct ext4_free_extent *ex,
2240 					struct ext4_buddy *e4b)
2241 {
2242 	struct ext4_free_extent *bex = &ac->ac_b_ex;
2243 	struct ext4_free_extent *gex = &ac->ac_g_ex;
2244 
2245 	BUG_ON(ex->fe_len <= 0);
2246 	BUG_ON(ex->fe_len > EXT4_CLUSTERS_PER_GROUP(ac->ac_sb));
2247 	BUG_ON(ex->fe_start >= EXT4_CLUSTERS_PER_GROUP(ac->ac_sb));
2248 	BUG_ON(ac->ac_status != AC_STATUS_CONTINUE);
2249 
2250 	ac->ac_found++;
2251 	ac->ac_cX_found[ac->ac_criteria]++;
2252 
2253 	/*
2254 	 * The special case - take what you catch first
2255 	 */
2256 	if (unlikely(ac->ac_flags & EXT4_MB_HINT_FIRST)) {
2257 		*bex = *ex;
2258 		ext4_mb_use_best_found(ac, e4b);
2259 		return;
2260 	}
2261 
2262 	/*
2263 	 * Let's check whether the chuck is good enough
2264 	 */
2265 	if (ex->fe_len == gex->fe_len) {
2266 		*bex = *ex;
2267 		ext4_mb_use_best_found(ac, e4b);
2268 		return;
2269 	}
2270 
2271 	/*
2272 	 * If this is first found extent, just store it in the context
2273 	 */
2274 	if (bex->fe_len == 0) {
2275 		*bex = *ex;
2276 		return;
2277 	}
2278 
2279 	/*
2280 	 * If new found extent is better, store it in the context
2281 	 */
2282 	if (bex->fe_len < gex->fe_len) {
2283 		/* if the request isn't satisfied, any found extent
2284 		 * larger than previous best one is better */
2285 		if (ex->fe_len > bex->fe_len)
2286 			*bex = *ex;
2287 	} else if (ex->fe_len > gex->fe_len) {
2288 		/* if the request is satisfied, then we try to find
2289 		 * an extent that still satisfy the request, but is
2290 		 * smaller than previous one */
2291 		if (ex->fe_len < bex->fe_len)
2292 			*bex = *ex;
2293 	}
2294 
2295 	ext4_mb_check_limits(ac, e4b, 0);
2296 }
2297 
2298 static noinline_for_stack
ext4_mb_try_best_found(struct ext4_allocation_context * ac,struct ext4_buddy * e4b)2299 void ext4_mb_try_best_found(struct ext4_allocation_context *ac,
2300 					struct ext4_buddy *e4b)
2301 {
2302 	struct ext4_free_extent ex = ac->ac_b_ex;
2303 	ext4_group_t group = ex.fe_group;
2304 	int max;
2305 	int err;
2306 
2307 	BUG_ON(ex.fe_len <= 0);
2308 	err = ext4_mb_load_buddy(ac->ac_sb, group, e4b);
2309 	if (err)
2310 		return;
2311 
2312 	ext4_lock_group(ac->ac_sb, group);
2313 	if (unlikely(EXT4_MB_GRP_BBITMAP_CORRUPT(e4b->bd_info)))
2314 		goto out;
2315 
2316 	max = mb_find_extent(e4b, ex.fe_start, ex.fe_len, &ex);
2317 
2318 	if (max > 0) {
2319 		ac->ac_b_ex = ex;
2320 		ext4_mb_use_best_found(ac, e4b);
2321 	}
2322 
2323 out:
2324 	ext4_unlock_group(ac->ac_sb, group);
2325 	ext4_mb_unload_buddy(e4b);
2326 }
2327 
2328 static noinline_for_stack
ext4_mb_find_by_goal(struct ext4_allocation_context * ac,struct ext4_buddy * e4b)2329 int ext4_mb_find_by_goal(struct ext4_allocation_context *ac,
2330 				struct ext4_buddy *e4b)
2331 {
2332 	ext4_group_t group = ac->ac_g_ex.fe_group;
2333 	int max;
2334 	int err;
2335 	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
2336 	struct ext4_group_info *grp = ext4_get_group_info(ac->ac_sb, group);
2337 	struct ext4_free_extent ex;
2338 
2339 	if (!grp)
2340 		return -EFSCORRUPTED;
2341 	if (!(ac->ac_flags & (EXT4_MB_HINT_TRY_GOAL | EXT4_MB_HINT_GOAL_ONLY)))
2342 		return 0;
2343 	if (grp->bb_free == 0)
2344 		return 0;
2345 
2346 	err = ext4_mb_load_buddy(ac->ac_sb, group, e4b);
2347 	if (err)
2348 		return err;
2349 
2350 	ext4_lock_group(ac->ac_sb, group);
2351 	if (unlikely(EXT4_MB_GRP_BBITMAP_CORRUPT(e4b->bd_info)))
2352 		goto out;
2353 
2354 	max = mb_find_extent(e4b, ac->ac_g_ex.fe_start,
2355 			     ac->ac_g_ex.fe_len, &ex);
2356 	ex.fe_logical = 0xDEADFA11; /* debug value */
2357 
2358 	if (max >= ac->ac_g_ex.fe_len &&
2359 	    ac->ac_g_ex.fe_len == EXT4_NUM_B2C(sbi, sbi->s_stripe)) {
2360 		ext4_fsblk_t start;
2361 
2362 		start = ext4_grp_offs_to_block(ac->ac_sb, &ex);
2363 		/* use do_div to get remainder (would be 64-bit modulo) */
2364 		if (do_div(start, sbi->s_stripe) == 0) {
2365 			ac->ac_found++;
2366 			ac->ac_b_ex = ex;
2367 			ext4_mb_use_best_found(ac, e4b);
2368 		}
2369 	} else if (max >= ac->ac_g_ex.fe_len) {
2370 		BUG_ON(ex.fe_len <= 0);
2371 		BUG_ON(ex.fe_group != ac->ac_g_ex.fe_group);
2372 		BUG_ON(ex.fe_start != ac->ac_g_ex.fe_start);
2373 		ac->ac_found++;
2374 		ac->ac_b_ex = ex;
2375 		ext4_mb_use_best_found(ac, e4b);
2376 	} else if (max > 0 && (ac->ac_flags & EXT4_MB_HINT_MERGE)) {
2377 		/* Sometimes, caller may want to merge even small
2378 		 * number of blocks to an existing extent */
2379 		BUG_ON(ex.fe_len <= 0);
2380 		BUG_ON(ex.fe_group != ac->ac_g_ex.fe_group);
2381 		BUG_ON(ex.fe_start != ac->ac_g_ex.fe_start);
2382 		ac->ac_found++;
2383 		ac->ac_b_ex = ex;
2384 		ext4_mb_use_best_found(ac, e4b);
2385 	}
2386 out:
2387 	ext4_unlock_group(ac->ac_sb, group);
2388 	ext4_mb_unload_buddy(e4b);
2389 
2390 	return 0;
2391 }
2392 
2393 /*
2394  * The routine scans buddy structures (not bitmap!) from given order
2395  * to max order and tries to find big enough chunk to satisfy the req
2396  */
2397 static noinline_for_stack
ext4_mb_simple_scan_group(struct ext4_allocation_context * ac,struct ext4_buddy * e4b)2398 void ext4_mb_simple_scan_group(struct ext4_allocation_context *ac,
2399 					struct ext4_buddy *e4b)
2400 {
2401 	struct super_block *sb = ac->ac_sb;
2402 	struct ext4_group_info *grp = e4b->bd_info;
2403 	void *buddy;
2404 	int i;
2405 	int k;
2406 	int max;
2407 
2408 	BUG_ON(ac->ac_2order <= 0);
2409 	for (i = ac->ac_2order; i < MB_NUM_ORDERS(sb); i++) {
2410 		if (grp->bb_counters[i] == 0)
2411 			continue;
2412 
2413 		buddy = mb_find_buddy(e4b, i, &max);
2414 		if (WARN_RATELIMIT(buddy == NULL,
2415 			 "ext4: mb_simple_scan_group: mb_find_buddy failed, (%d)\n", i))
2416 			continue;
2417 
2418 		k = mb_find_next_zero_bit(buddy, max, 0);
2419 		if (k >= max) {
2420 			ext4_mark_group_bitmap_corrupted(ac->ac_sb,
2421 					e4b->bd_group,
2422 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
2423 			ext4_grp_locked_error(ac->ac_sb, e4b->bd_group, 0, 0,
2424 				"%d free clusters of order %d. But found 0",
2425 				grp->bb_counters[i], i);
2426 			break;
2427 		}
2428 		ac->ac_found++;
2429 		ac->ac_cX_found[ac->ac_criteria]++;
2430 
2431 		ac->ac_b_ex.fe_len = 1 << i;
2432 		ac->ac_b_ex.fe_start = k << i;
2433 		ac->ac_b_ex.fe_group = e4b->bd_group;
2434 
2435 		ext4_mb_use_best_found(ac, e4b);
2436 
2437 		BUG_ON(ac->ac_f_ex.fe_len != ac->ac_g_ex.fe_len);
2438 
2439 		if (EXT4_SB(sb)->s_mb_stats)
2440 			atomic_inc(&EXT4_SB(sb)->s_bal_2orders);
2441 
2442 		break;
2443 	}
2444 }
2445 
2446 /*
2447  * The routine scans the group and measures all found extents.
2448  * In order to optimize scanning, caller must pass number of
2449  * free blocks in the group, so the routine can know upper limit.
2450  */
2451 static noinline_for_stack
ext4_mb_complex_scan_group(struct ext4_allocation_context * ac,struct ext4_buddy * e4b)2452 void ext4_mb_complex_scan_group(struct ext4_allocation_context *ac,
2453 					struct ext4_buddy *e4b)
2454 {
2455 	struct super_block *sb = ac->ac_sb;
2456 	void *bitmap = e4b->bd_bitmap;
2457 	struct ext4_free_extent ex;
2458 	int i, j, freelen;
2459 	int free;
2460 
2461 	free = e4b->bd_info->bb_free;
2462 	if (WARN_ON(free <= 0))
2463 		return;
2464 
2465 	i = e4b->bd_info->bb_first_free;
2466 
2467 	while (free && ac->ac_status == AC_STATUS_CONTINUE) {
2468 		i = mb_find_next_zero_bit(bitmap,
2469 						EXT4_CLUSTERS_PER_GROUP(sb), i);
2470 		if (i >= EXT4_CLUSTERS_PER_GROUP(sb)) {
2471 			/*
2472 			 * IF we have corrupt bitmap, we won't find any
2473 			 * free blocks even though group info says we
2474 			 * have free blocks
2475 			 */
2476 			ext4_mark_group_bitmap_corrupted(sb, e4b->bd_group,
2477 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
2478 			ext4_grp_locked_error(sb, e4b->bd_group, 0, 0,
2479 					"%d free clusters as per "
2480 					"group info. But bitmap says 0",
2481 					free);
2482 			break;
2483 		}
2484 
2485 		if (!ext4_mb_cr_expensive(ac->ac_criteria)) {
2486 			/*
2487 			 * In CR_GOAL_LEN_FAST and CR_BEST_AVAIL_LEN, we are
2488 			 * sure that this group will have a large enough
2489 			 * continuous free extent, so skip over the smaller free
2490 			 * extents
2491 			 */
2492 			j = mb_find_next_bit(bitmap,
2493 						EXT4_CLUSTERS_PER_GROUP(sb), i);
2494 			freelen = j - i;
2495 
2496 			if (freelen < ac->ac_g_ex.fe_len) {
2497 				i = j;
2498 				free -= freelen;
2499 				continue;
2500 			}
2501 		}
2502 
2503 		mb_find_extent(e4b, i, ac->ac_g_ex.fe_len, &ex);
2504 		if (WARN_ON(ex.fe_len <= 0))
2505 			break;
2506 		if (free < ex.fe_len) {
2507 			ext4_mark_group_bitmap_corrupted(sb, e4b->bd_group,
2508 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
2509 			ext4_grp_locked_error(sb, e4b->bd_group, 0, 0,
2510 					"%d free clusters as per "
2511 					"group info. But got %d blocks",
2512 					free, ex.fe_len);
2513 			/*
2514 			 * The number of free blocks differs. This mostly
2515 			 * indicate that the bitmap is corrupt. So exit
2516 			 * without claiming the space.
2517 			 */
2518 			break;
2519 		}
2520 		ex.fe_logical = 0xDEADC0DE; /* debug value */
2521 		ext4_mb_measure_extent(ac, &ex, e4b);
2522 
2523 		i += ex.fe_len;
2524 		free -= ex.fe_len;
2525 	}
2526 
2527 	ext4_mb_check_limits(ac, e4b, 1);
2528 }
2529 
2530 /*
2531  * This is a special case for storages like raid5
2532  * we try to find stripe-aligned chunks for stripe-size-multiple requests
2533  */
2534 static noinline_for_stack
ext4_mb_scan_aligned(struct ext4_allocation_context * ac,struct ext4_buddy * e4b)2535 void ext4_mb_scan_aligned(struct ext4_allocation_context *ac,
2536 				 struct ext4_buddy *e4b)
2537 {
2538 	struct super_block *sb = ac->ac_sb;
2539 	struct ext4_sb_info *sbi = EXT4_SB(sb);
2540 	void *bitmap = e4b->bd_bitmap;
2541 	struct ext4_free_extent ex;
2542 	ext4_fsblk_t first_group_block;
2543 	ext4_fsblk_t a;
2544 	ext4_grpblk_t i, stripe;
2545 	int max;
2546 
2547 	BUG_ON(sbi->s_stripe == 0);
2548 
2549 	/* find first stripe-aligned block in group */
2550 	first_group_block = ext4_group_first_block_no(sb, e4b->bd_group);
2551 
2552 	a = first_group_block + sbi->s_stripe - 1;
2553 	do_div(a, sbi->s_stripe);
2554 	i = (a * sbi->s_stripe) - first_group_block;
2555 
2556 	stripe = EXT4_NUM_B2C(sbi, sbi->s_stripe);
2557 	i = EXT4_B2C(sbi, i);
2558 	while (i < EXT4_CLUSTERS_PER_GROUP(sb)) {
2559 		if (!mb_test_bit(i, bitmap)) {
2560 			max = mb_find_extent(e4b, i, stripe, &ex);
2561 			if (max >= stripe) {
2562 				ac->ac_found++;
2563 				ac->ac_cX_found[ac->ac_criteria]++;
2564 				ex.fe_logical = 0xDEADF00D; /* debug value */
2565 				ac->ac_b_ex = ex;
2566 				ext4_mb_use_best_found(ac, e4b);
2567 				break;
2568 			}
2569 		}
2570 		i += stripe;
2571 	}
2572 }
2573 
2574 /*
2575  * This is also called BEFORE we load the buddy bitmap.
2576  * Returns either 1 or 0 indicating that the group is either suitable
2577  * for the allocation or not.
2578  */
ext4_mb_good_group(struct ext4_allocation_context * ac,ext4_group_t group,enum criteria cr)2579 static bool ext4_mb_good_group(struct ext4_allocation_context *ac,
2580 				ext4_group_t group, enum criteria cr)
2581 {
2582 	ext4_grpblk_t free, fragments;
2583 	int flex_size = ext4_flex_bg_size(EXT4_SB(ac->ac_sb));
2584 	struct ext4_group_info *grp = ext4_get_group_info(ac->ac_sb, group);
2585 
2586 	BUG_ON(cr < CR_POWER2_ALIGNED || cr >= EXT4_MB_NUM_CRS);
2587 
2588 	if (unlikely(!grp || EXT4_MB_GRP_BBITMAP_CORRUPT(grp)))
2589 		return false;
2590 
2591 	free = grp->bb_free;
2592 	if (free == 0)
2593 		return false;
2594 
2595 	fragments = grp->bb_fragments;
2596 	if (fragments == 0)
2597 		return false;
2598 
2599 	switch (cr) {
2600 	case CR_POWER2_ALIGNED:
2601 		BUG_ON(ac->ac_2order == 0);
2602 
2603 		/* Avoid using the first bg of a flexgroup for data files */
2604 		if ((ac->ac_flags & EXT4_MB_HINT_DATA) &&
2605 		    (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) &&
2606 		    ((group % flex_size) == 0))
2607 			return false;
2608 
2609 		if (free < ac->ac_g_ex.fe_len)
2610 			return false;
2611 
2612 		if (ac->ac_2order >= MB_NUM_ORDERS(ac->ac_sb))
2613 			return true;
2614 
2615 		if (grp->bb_largest_free_order < ac->ac_2order)
2616 			return false;
2617 
2618 		return true;
2619 	case CR_GOAL_LEN_FAST:
2620 	case CR_BEST_AVAIL_LEN:
2621 		if ((free / fragments) >= ac->ac_g_ex.fe_len)
2622 			return true;
2623 		break;
2624 	case CR_GOAL_LEN_SLOW:
2625 		if (free >= ac->ac_g_ex.fe_len)
2626 			return true;
2627 		break;
2628 	case CR_ANY_FREE:
2629 		return true;
2630 	default:
2631 		BUG();
2632 	}
2633 
2634 	return false;
2635 }
2636 
2637 /*
2638  * This could return negative error code if something goes wrong
2639  * during ext4_mb_init_group(). This should not be called with
2640  * ext4_lock_group() held.
2641  *
2642  * Note: because we are conditionally operating with the group lock in
2643  * the EXT4_MB_STRICT_CHECK case, we need to fake out sparse in this
2644  * function using __acquire and __release.  This means we need to be
2645  * super careful before messing with the error path handling via "goto
2646  * out"!
2647  */
ext4_mb_good_group_nolock(struct ext4_allocation_context * ac,ext4_group_t group,enum criteria cr)2648 static int ext4_mb_good_group_nolock(struct ext4_allocation_context *ac,
2649 				     ext4_group_t group, enum criteria cr)
2650 {
2651 	struct ext4_group_info *grp = ext4_get_group_info(ac->ac_sb, group);
2652 	struct super_block *sb = ac->ac_sb;
2653 	struct ext4_sb_info *sbi = EXT4_SB(sb);
2654 	bool should_lock = ac->ac_flags & EXT4_MB_STRICT_CHECK;
2655 	ext4_grpblk_t free;
2656 	int ret = 0;
2657 
2658 	if (!grp)
2659 		return -EFSCORRUPTED;
2660 	if (sbi->s_mb_stats)
2661 		atomic64_inc(&sbi->s_bal_cX_groups_considered[ac->ac_criteria]);
2662 	if (should_lock) {
2663 		ext4_lock_group(sb, group);
2664 		__release(ext4_group_lock_ptr(sb, group));
2665 	}
2666 	free = grp->bb_free;
2667 	if (free == 0)
2668 		goto out;
2669 	/*
2670 	 * In all criterias except CR_ANY_FREE we try to avoid groups that
2671 	 * can't possibly satisfy the full goal request due to insufficient
2672 	 * free blocks.
2673 	 */
2674 	if (cr < CR_ANY_FREE && free < ac->ac_g_ex.fe_len)
2675 		goto out;
2676 	if (unlikely(EXT4_MB_GRP_BBITMAP_CORRUPT(grp)))
2677 		goto out;
2678 	if (should_lock) {
2679 		__acquire(ext4_group_lock_ptr(sb, group));
2680 		ext4_unlock_group(sb, group);
2681 	}
2682 
2683 	/* We only do this if the grp has never been initialized */
2684 	if (unlikely(EXT4_MB_GRP_NEED_INIT(grp))) {
2685 		struct ext4_group_desc *gdp =
2686 			ext4_get_group_desc(sb, group, NULL);
2687 		int ret;
2688 
2689 		/*
2690 		 * CR_POWER2_ALIGNED/CR_GOAL_LEN_FAST is a very optimistic
2691 		 * search to find large good chunks almost for free. If buddy
2692 		 * data is not ready, then this optimization makes no sense. But
2693 		 * we never skip the first block group in a flex_bg, since this
2694 		 * gets used for metadata block allocation, and we want to make
2695 		 * sure we locate metadata blocks in the first block group in
2696 		 * the flex_bg if possible.
2697 		 */
2698 		if (!ext4_mb_cr_expensive(cr) &&
2699 		    (!sbi->s_log_groups_per_flex ||
2700 		     ((group & ((1 << sbi->s_log_groups_per_flex) - 1)) != 0)) &&
2701 		    !(ext4_has_group_desc_csum(sb) &&
2702 		      (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))))
2703 			return 0;
2704 		ret = ext4_mb_init_group(sb, group, GFP_NOFS);
2705 		if (ret)
2706 			return ret;
2707 	}
2708 
2709 	if (should_lock) {
2710 		ext4_lock_group(sb, group);
2711 		__release(ext4_group_lock_ptr(sb, group));
2712 	}
2713 	ret = ext4_mb_good_group(ac, group, cr);
2714 out:
2715 	if (should_lock) {
2716 		__acquire(ext4_group_lock_ptr(sb, group));
2717 		ext4_unlock_group(sb, group);
2718 	}
2719 	return ret;
2720 }
2721 
2722 /*
2723  * Start prefetching @nr block bitmaps starting at @group.
2724  * Return the next group which needs to be prefetched.
2725  */
ext4_mb_prefetch(struct super_block * sb,ext4_group_t group,unsigned int nr,int * cnt)2726 ext4_group_t ext4_mb_prefetch(struct super_block *sb, ext4_group_t group,
2727 			      unsigned int nr, int *cnt)
2728 {
2729 	ext4_group_t ngroups = ext4_get_groups_count(sb);
2730 	struct buffer_head *bh;
2731 	struct blk_plug plug;
2732 
2733 	blk_start_plug(&plug);
2734 	while (nr-- > 0) {
2735 		struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group,
2736 								  NULL);
2737 		struct ext4_group_info *grp = ext4_get_group_info(sb, group);
2738 
2739 		/*
2740 		 * Prefetch block groups with free blocks; but don't
2741 		 * bother if it is marked uninitialized on disk, since
2742 		 * it won't require I/O to read.  Also only try to
2743 		 * prefetch once, so we avoid getblk() call, which can
2744 		 * be expensive.
2745 		 */
2746 		if (gdp && grp && !EXT4_MB_GRP_TEST_AND_SET_READ(grp) &&
2747 		    EXT4_MB_GRP_NEED_INIT(grp) &&
2748 		    ext4_free_group_clusters(sb, gdp) > 0 ) {
2749 			bh = ext4_read_block_bitmap_nowait(sb, group, true);
2750 			if (bh && !IS_ERR(bh)) {
2751 				if (!buffer_uptodate(bh) && cnt)
2752 					(*cnt)++;
2753 				brelse(bh);
2754 			}
2755 		}
2756 		if (++group >= ngroups)
2757 			group = 0;
2758 	}
2759 	blk_finish_plug(&plug);
2760 	return group;
2761 }
2762 
2763 /*
2764  * Prefetching reads the block bitmap into the buffer cache; but we
2765  * need to make sure that the buddy bitmap in the page cache has been
2766  * initialized.  Note that ext4_mb_init_group() will block if the I/O
2767  * is not yet completed, or indeed if it was not initiated by
2768  * ext4_mb_prefetch did not start the I/O.
2769  *
2770  * TODO: We should actually kick off the buddy bitmap setup in a work
2771  * queue when the buffer I/O is completed, so that we don't block
2772  * waiting for the block allocation bitmap read to finish when
2773  * ext4_mb_prefetch_fini is called from ext4_mb_regular_allocator().
2774  */
ext4_mb_prefetch_fini(struct super_block * sb,ext4_group_t group,unsigned int nr)2775 void ext4_mb_prefetch_fini(struct super_block *sb, ext4_group_t group,
2776 			   unsigned int nr)
2777 {
2778 	struct ext4_group_desc *gdp;
2779 	struct ext4_group_info *grp;
2780 
2781 	while (nr-- > 0) {
2782 		if (!group)
2783 			group = ext4_get_groups_count(sb);
2784 		group--;
2785 		gdp = ext4_get_group_desc(sb, group, NULL);
2786 		grp = ext4_get_group_info(sb, group);
2787 
2788 		if (grp && gdp && EXT4_MB_GRP_NEED_INIT(grp) &&
2789 		    ext4_free_group_clusters(sb, gdp) > 0) {
2790 			if (ext4_mb_init_group(sb, group, GFP_NOFS))
2791 				break;
2792 		}
2793 	}
2794 }
2795 
2796 static noinline_for_stack int
ext4_mb_regular_allocator(struct ext4_allocation_context * ac)2797 ext4_mb_regular_allocator(struct ext4_allocation_context *ac)
2798 {
2799 	ext4_group_t prefetch_grp = 0, ngroups, group, i;
2800 	enum criteria new_cr, cr = CR_GOAL_LEN_FAST;
2801 	int err = 0, first_err = 0;
2802 	unsigned int nr = 0, prefetch_ios = 0;
2803 	struct ext4_sb_info *sbi;
2804 	struct super_block *sb;
2805 	struct ext4_buddy e4b;
2806 	int lost;
2807 
2808 	sb = ac->ac_sb;
2809 	sbi = EXT4_SB(sb);
2810 	ngroups = ext4_get_groups_count(sb);
2811 	/* non-extent files are limited to low blocks/groups */
2812 	if (!(ext4_test_inode_flag(ac->ac_inode, EXT4_INODE_EXTENTS)))
2813 		ngroups = sbi->s_blockfile_groups;
2814 
2815 	BUG_ON(ac->ac_status == AC_STATUS_FOUND);
2816 
2817 	/* first, try the goal */
2818 	err = ext4_mb_find_by_goal(ac, &e4b);
2819 	if (err || ac->ac_status == AC_STATUS_FOUND)
2820 		goto out;
2821 
2822 	if (unlikely(ac->ac_flags & EXT4_MB_HINT_GOAL_ONLY))
2823 		goto out;
2824 
2825 	/*
2826 	 * ac->ac_2order is set only if the fe_len is a power of 2
2827 	 * if ac->ac_2order is set we also set criteria to CR_POWER2_ALIGNED
2828 	 * so that we try exact allocation using buddy.
2829 	 */
2830 	i = fls(ac->ac_g_ex.fe_len);
2831 	ac->ac_2order = 0;
2832 	/*
2833 	 * We search using buddy data only if the order of the request
2834 	 * is greater than equal to the sbi_s_mb_order2_reqs
2835 	 * You can tune it via /sys/fs/ext4/<partition>/mb_order2_req
2836 	 * We also support searching for power-of-two requests only for
2837 	 * requests upto maximum buddy size we have constructed.
2838 	 */
2839 	if (i >= sbi->s_mb_order2_reqs && i <= MB_NUM_ORDERS(sb)) {
2840 		if (is_power_of_2(ac->ac_g_ex.fe_len))
2841 			ac->ac_2order = array_index_nospec(i - 1,
2842 							   MB_NUM_ORDERS(sb));
2843 	}
2844 
2845 	/* if stream allocation is enabled, use global goal */
2846 	if (ac->ac_flags & EXT4_MB_STREAM_ALLOC) {
2847 		/* TBD: may be hot point */
2848 		spin_lock(&sbi->s_md_lock);
2849 		ac->ac_g_ex.fe_group = sbi->s_mb_last_group;
2850 		ac->ac_g_ex.fe_start = sbi->s_mb_last_start;
2851 		spin_unlock(&sbi->s_md_lock);
2852 	}
2853 
2854 	/*
2855 	 * Let's just scan groups to find more-less suitable blocks We
2856 	 * start with CR_GOAL_LEN_FAST, unless it is power of 2
2857 	 * aligned, in which case let's do that faster approach first.
2858 	 */
2859 	if (ac->ac_2order)
2860 		cr = CR_POWER2_ALIGNED;
2861 repeat:
2862 	for (; cr < EXT4_MB_NUM_CRS && ac->ac_status == AC_STATUS_CONTINUE; cr++) {
2863 		ac->ac_criteria = cr;
2864 		/*
2865 		 * searching for the right group start
2866 		 * from the goal value specified
2867 		 */
2868 		group = ac->ac_g_ex.fe_group;
2869 		ac->ac_groups_linear_remaining = sbi->s_mb_max_linear_groups;
2870 		prefetch_grp = group;
2871 		nr = 0;
2872 
2873 		for (i = 0, new_cr = cr; i < ngroups; i++,
2874 		     ext4_mb_choose_next_group(ac, &new_cr, &group, ngroups)) {
2875 			int ret = 0;
2876 
2877 			cond_resched();
2878 			if (new_cr != cr) {
2879 				cr = new_cr;
2880 				goto repeat;
2881 			}
2882 
2883 			/*
2884 			 * Batch reads of the block allocation bitmaps
2885 			 * to get multiple READs in flight; limit
2886 			 * prefetching at inexpensive CR, otherwise mballoc
2887 			 * can spend a lot of time loading imperfect groups
2888 			 */
2889 			if ((prefetch_grp == group) &&
2890 			    (ext4_mb_cr_expensive(cr) ||
2891 			     prefetch_ios < sbi->s_mb_prefetch_limit)) {
2892 				nr = sbi->s_mb_prefetch;
2893 				if (ext4_has_feature_flex_bg(sb)) {
2894 					nr = 1 << sbi->s_log_groups_per_flex;
2895 					nr -= group & (nr - 1);
2896 					nr = min(nr, sbi->s_mb_prefetch);
2897 				}
2898 				prefetch_grp = ext4_mb_prefetch(sb, group,
2899 							nr, &prefetch_ios);
2900 			}
2901 
2902 			/* This now checks without needing the buddy page */
2903 			ret = ext4_mb_good_group_nolock(ac, group, cr);
2904 			if (ret <= 0) {
2905 				if (!first_err)
2906 					first_err = ret;
2907 				continue;
2908 			}
2909 
2910 			err = ext4_mb_load_buddy(sb, group, &e4b);
2911 			if (err)
2912 				goto out;
2913 
2914 			ext4_lock_group(sb, group);
2915 
2916 			/*
2917 			 * We need to check again after locking the
2918 			 * block group
2919 			 */
2920 			ret = ext4_mb_good_group(ac, group, cr);
2921 			if (ret == 0) {
2922 				ext4_unlock_group(sb, group);
2923 				ext4_mb_unload_buddy(&e4b);
2924 				continue;
2925 			}
2926 
2927 			ac->ac_groups_scanned++;
2928 			if (cr == CR_POWER2_ALIGNED)
2929 				ext4_mb_simple_scan_group(ac, &e4b);
2930 			else {
2931 				bool is_stripe_aligned =
2932 					(sbi->s_stripe >=
2933 					 sbi->s_cluster_ratio) &&
2934 					!(ac->ac_g_ex.fe_len %
2935 					  EXT4_NUM_B2C(sbi, sbi->s_stripe));
2936 
2937 				if ((cr == CR_GOAL_LEN_FAST ||
2938 				     cr == CR_BEST_AVAIL_LEN) &&
2939 				    is_stripe_aligned)
2940 					ext4_mb_scan_aligned(ac, &e4b);
2941 
2942 				if (ac->ac_status == AC_STATUS_CONTINUE)
2943 					ext4_mb_complex_scan_group(ac, &e4b);
2944 			}
2945 
2946 			ext4_unlock_group(sb, group);
2947 			ext4_mb_unload_buddy(&e4b);
2948 
2949 			if (ac->ac_status != AC_STATUS_CONTINUE)
2950 				break;
2951 		}
2952 		/* Processed all groups and haven't found blocks */
2953 		if (sbi->s_mb_stats && i == ngroups)
2954 			atomic64_inc(&sbi->s_bal_cX_failed[cr]);
2955 
2956 		if (i == ngroups && ac->ac_criteria == CR_BEST_AVAIL_LEN)
2957 			/* Reset goal length to original goal length before
2958 			 * falling into CR_GOAL_LEN_SLOW */
2959 			ac->ac_g_ex.fe_len = ac->ac_orig_goal_len;
2960 	}
2961 
2962 	if (ac->ac_b_ex.fe_len > 0 && ac->ac_status != AC_STATUS_FOUND &&
2963 	    !(ac->ac_flags & EXT4_MB_HINT_FIRST)) {
2964 		/*
2965 		 * We've been searching too long. Let's try to allocate
2966 		 * the best chunk we've found so far
2967 		 */
2968 		ext4_mb_try_best_found(ac, &e4b);
2969 		if (ac->ac_status != AC_STATUS_FOUND) {
2970 			/*
2971 			 * Someone more lucky has already allocated it.
2972 			 * The only thing we can do is just take first
2973 			 * found block(s)
2974 			 */
2975 			lost = atomic_inc_return(&sbi->s_mb_lost_chunks);
2976 			mb_debug(sb, "lost chunk, group: %u, start: %d, len: %d, lost: %d\n",
2977 				 ac->ac_b_ex.fe_group, ac->ac_b_ex.fe_start,
2978 				 ac->ac_b_ex.fe_len, lost);
2979 
2980 			ac->ac_b_ex.fe_group = 0;
2981 			ac->ac_b_ex.fe_start = 0;
2982 			ac->ac_b_ex.fe_len = 0;
2983 			ac->ac_status = AC_STATUS_CONTINUE;
2984 			ac->ac_flags |= EXT4_MB_HINT_FIRST;
2985 			cr = CR_ANY_FREE;
2986 			goto repeat;
2987 		}
2988 	}
2989 
2990 	if (sbi->s_mb_stats && ac->ac_status == AC_STATUS_FOUND)
2991 		atomic64_inc(&sbi->s_bal_cX_hits[ac->ac_criteria]);
2992 out:
2993 	if (!err && ac->ac_status != AC_STATUS_FOUND && first_err)
2994 		err = first_err;
2995 
2996 	mb_debug(sb, "Best len %d, origin len %d, ac_status %u, ac_flags 0x%x, cr %d ret %d\n",
2997 		 ac->ac_b_ex.fe_len, ac->ac_o_ex.fe_len, ac->ac_status,
2998 		 ac->ac_flags, cr, err);
2999 
3000 	if (nr)
3001 		ext4_mb_prefetch_fini(sb, prefetch_grp, nr);
3002 
3003 	return err;
3004 }
3005 
ext4_mb_seq_groups_start(struct seq_file * seq,loff_t * pos)3006 static void *ext4_mb_seq_groups_start(struct seq_file *seq, loff_t *pos)
3007 {
3008 	struct super_block *sb = pde_data(file_inode(seq->file));
3009 	ext4_group_t group;
3010 
3011 	if (*pos < 0 || *pos >= ext4_get_groups_count(sb))
3012 		return NULL;
3013 	group = *pos + 1;
3014 	return (void *) ((unsigned long) group);
3015 }
3016 
ext4_mb_seq_groups_next(struct seq_file * seq,void * v,loff_t * pos)3017 static void *ext4_mb_seq_groups_next(struct seq_file *seq, void *v, loff_t *pos)
3018 {
3019 	struct super_block *sb = pde_data(file_inode(seq->file));
3020 	ext4_group_t group;
3021 
3022 	++*pos;
3023 	if (*pos < 0 || *pos >= ext4_get_groups_count(sb))
3024 		return NULL;
3025 	group = *pos + 1;
3026 	return (void *) ((unsigned long) group);
3027 }
3028 
ext4_mb_seq_groups_show(struct seq_file * seq,void * v)3029 static int ext4_mb_seq_groups_show(struct seq_file *seq, void *v)
3030 {
3031 	struct super_block *sb = pde_data(file_inode(seq->file));
3032 	ext4_group_t group = (ext4_group_t) ((unsigned long) v);
3033 	int i, err;
3034 	char nbuf[16];
3035 	struct ext4_buddy e4b;
3036 	struct ext4_group_info *grinfo;
3037 	unsigned char blocksize_bits = min_t(unsigned char,
3038 					     sb->s_blocksize_bits,
3039 					     EXT4_MAX_BLOCK_LOG_SIZE);
3040 	struct sg {
3041 		struct ext4_group_info info;
3042 		ext4_grpblk_t counters[EXT4_MAX_BLOCK_LOG_SIZE + 2];
3043 	} sg;
3044 
3045 	group--;
3046 	if (group == 0)
3047 		seq_puts(seq, "#group: free  frags first ["
3048 			      " 2^0   2^1   2^2   2^3   2^4   2^5   2^6  "
3049 			      " 2^7   2^8   2^9   2^10  2^11  2^12  2^13  ]\n");
3050 
3051 	i = (blocksize_bits + 2) * sizeof(sg.info.bb_counters[0]) +
3052 		sizeof(struct ext4_group_info);
3053 
3054 	grinfo = ext4_get_group_info(sb, group);
3055 	if (!grinfo)
3056 		return 0;
3057 	/* Load the group info in memory only if not already loaded. */
3058 	if (unlikely(EXT4_MB_GRP_NEED_INIT(grinfo))) {
3059 		err = ext4_mb_load_buddy(sb, group, &e4b);
3060 		if (err) {
3061 			seq_printf(seq, "#%-5u: %s\n", group, ext4_decode_error(NULL, err, nbuf));
3062 			return 0;
3063 		}
3064 		ext4_mb_unload_buddy(&e4b);
3065 	}
3066 
3067 	/*
3068 	 * We care only about free space counters in the group info and
3069 	 * these are safe to access even after the buddy has been unloaded
3070 	 */
3071 	memcpy(&sg, grinfo, i);
3072 	seq_printf(seq, "#%-5u: %-5u %-5u %-5u [", group, sg.info.bb_free,
3073 			sg.info.bb_fragments, sg.info.bb_first_free);
3074 	for (i = 0; i <= 13; i++)
3075 		seq_printf(seq, " %-5u", i <= blocksize_bits + 1 ?
3076 				sg.info.bb_counters[i] : 0);
3077 	seq_puts(seq, " ]");
3078 	if (EXT4_MB_GRP_BBITMAP_CORRUPT(&sg.info))
3079 		seq_puts(seq, " Block bitmap corrupted!");
3080 	seq_putc(seq, '\n');
3081 	return 0;
3082 }
3083 
ext4_mb_seq_groups_stop(struct seq_file * seq,void * v)3084 static void ext4_mb_seq_groups_stop(struct seq_file *seq, void *v)
3085 {
3086 }
3087 
3088 const struct seq_operations ext4_mb_seq_groups_ops = {
3089 	.start  = ext4_mb_seq_groups_start,
3090 	.next   = ext4_mb_seq_groups_next,
3091 	.stop   = ext4_mb_seq_groups_stop,
3092 	.show   = ext4_mb_seq_groups_show,
3093 };
3094 
ext4_seq_mb_stats_show(struct seq_file * seq,void * offset)3095 int ext4_seq_mb_stats_show(struct seq_file *seq, void *offset)
3096 {
3097 	struct super_block *sb = seq->private;
3098 	struct ext4_sb_info *sbi = EXT4_SB(sb);
3099 
3100 	seq_puts(seq, "mballoc:\n");
3101 	if (!sbi->s_mb_stats) {
3102 		seq_puts(seq, "\tmb stats collection turned off.\n");
3103 		seq_puts(
3104 			seq,
3105 			"\tTo enable, please write \"1\" to sysfs file mb_stats.\n");
3106 		return 0;
3107 	}
3108 	seq_printf(seq, "\treqs: %u\n", atomic_read(&sbi->s_bal_reqs));
3109 	seq_printf(seq, "\tsuccess: %u\n", atomic_read(&sbi->s_bal_success));
3110 
3111 	seq_printf(seq, "\tgroups_scanned: %u\n",
3112 		   atomic_read(&sbi->s_bal_groups_scanned));
3113 
3114 	/* CR_POWER2_ALIGNED stats */
3115 	seq_puts(seq, "\tcr_p2_aligned_stats:\n");
3116 	seq_printf(seq, "\t\thits: %llu\n",
3117 		   atomic64_read(&sbi->s_bal_cX_hits[CR_POWER2_ALIGNED]));
3118 	seq_printf(
3119 		seq, "\t\tgroups_considered: %llu\n",
3120 		atomic64_read(
3121 			&sbi->s_bal_cX_groups_considered[CR_POWER2_ALIGNED]));
3122 	seq_printf(seq, "\t\textents_scanned: %u\n",
3123 		   atomic_read(&sbi->s_bal_cX_ex_scanned[CR_POWER2_ALIGNED]));
3124 	seq_printf(seq, "\t\tuseless_loops: %llu\n",
3125 		   atomic64_read(&sbi->s_bal_cX_failed[CR_POWER2_ALIGNED]));
3126 	seq_printf(seq, "\t\tbad_suggestions: %u\n",
3127 		   atomic_read(&sbi->s_bal_p2_aligned_bad_suggestions));
3128 
3129 	/* CR_GOAL_LEN_FAST stats */
3130 	seq_puts(seq, "\tcr_goal_fast_stats:\n");
3131 	seq_printf(seq, "\t\thits: %llu\n",
3132 		   atomic64_read(&sbi->s_bal_cX_hits[CR_GOAL_LEN_FAST]));
3133 	seq_printf(seq, "\t\tgroups_considered: %llu\n",
3134 		   atomic64_read(
3135 			   &sbi->s_bal_cX_groups_considered[CR_GOAL_LEN_FAST]));
3136 	seq_printf(seq, "\t\textents_scanned: %u\n",
3137 		   atomic_read(&sbi->s_bal_cX_ex_scanned[CR_GOAL_LEN_FAST]));
3138 	seq_printf(seq, "\t\tuseless_loops: %llu\n",
3139 		   atomic64_read(&sbi->s_bal_cX_failed[CR_GOAL_LEN_FAST]));
3140 	seq_printf(seq, "\t\tbad_suggestions: %u\n",
3141 		   atomic_read(&sbi->s_bal_goal_fast_bad_suggestions));
3142 
3143 	/* CR_BEST_AVAIL_LEN stats */
3144 	seq_puts(seq, "\tcr_best_avail_stats:\n");
3145 	seq_printf(seq, "\t\thits: %llu\n",
3146 		   atomic64_read(&sbi->s_bal_cX_hits[CR_BEST_AVAIL_LEN]));
3147 	seq_printf(
3148 		seq, "\t\tgroups_considered: %llu\n",
3149 		atomic64_read(
3150 			&sbi->s_bal_cX_groups_considered[CR_BEST_AVAIL_LEN]));
3151 	seq_printf(seq, "\t\textents_scanned: %u\n",
3152 		   atomic_read(&sbi->s_bal_cX_ex_scanned[CR_BEST_AVAIL_LEN]));
3153 	seq_printf(seq, "\t\tuseless_loops: %llu\n",
3154 		   atomic64_read(&sbi->s_bal_cX_failed[CR_BEST_AVAIL_LEN]));
3155 	seq_printf(seq, "\t\tbad_suggestions: %u\n",
3156 		   atomic_read(&sbi->s_bal_best_avail_bad_suggestions));
3157 
3158 	/* CR_GOAL_LEN_SLOW stats */
3159 	seq_puts(seq, "\tcr_goal_slow_stats:\n");
3160 	seq_printf(seq, "\t\thits: %llu\n",
3161 		   atomic64_read(&sbi->s_bal_cX_hits[CR_GOAL_LEN_SLOW]));
3162 	seq_printf(seq, "\t\tgroups_considered: %llu\n",
3163 		   atomic64_read(
3164 			   &sbi->s_bal_cX_groups_considered[CR_GOAL_LEN_SLOW]));
3165 	seq_printf(seq, "\t\textents_scanned: %u\n",
3166 		   atomic_read(&sbi->s_bal_cX_ex_scanned[CR_GOAL_LEN_SLOW]));
3167 	seq_printf(seq, "\t\tuseless_loops: %llu\n",
3168 		   atomic64_read(&sbi->s_bal_cX_failed[CR_GOAL_LEN_SLOW]));
3169 
3170 	/* CR_ANY_FREE stats */
3171 	seq_puts(seq, "\tcr_any_free_stats:\n");
3172 	seq_printf(seq, "\t\thits: %llu\n",
3173 		   atomic64_read(&sbi->s_bal_cX_hits[CR_ANY_FREE]));
3174 	seq_printf(
3175 		seq, "\t\tgroups_considered: %llu\n",
3176 		atomic64_read(&sbi->s_bal_cX_groups_considered[CR_ANY_FREE]));
3177 	seq_printf(seq, "\t\textents_scanned: %u\n",
3178 		   atomic_read(&sbi->s_bal_cX_ex_scanned[CR_ANY_FREE]));
3179 	seq_printf(seq, "\t\tuseless_loops: %llu\n",
3180 		   atomic64_read(&sbi->s_bal_cX_failed[CR_ANY_FREE]));
3181 
3182 	/* Aggregates */
3183 	seq_printf(seq, "\textents_scanned: %u\n",
3184 		   atomic_read(&sbi->s_bal_ex_scanned));
3185 	seq_printf(seq, "\t\tgoal_hits: %u\n", atomic_read(&sbi->s_bal_goals));
3186 	seq_printf(seq, "\t\tlen_goal_hits: %u\n",
3187 		   atomic_read(&sbi->s_bal_len_goals));
3188 	seq_printf(seq, "\t\t2^n_hits: %u\n", atomic_read(&sbi->s_bal_2orders));
3189 	seq_printf(seq, "\t\tbreaks: %u\n", atomic_read(&sbi->s_bal_breaks));
3190 	seq_printf(seq, "\t\tlost: %u\n", atomic_read(&sbi->s_mb_lost_chunks));
3191 	seq_printf(seq, "\tbuddies_generated: %u/%u\n",
3192 		   atomic_read(&sbi->s_mb_buddies_generated),
3193 		   ext4_get_groups_count(sb));
3194 	seq_printf(seq, "\tbuddies_time_used: %llu\n",
3195 		   atomic64_read(&sbi->s_mb_generation_time));
3196 	seq_printf(seq, "\tpreallocated: %u\n",
3197 		   atomic_read(&sbi->s_mb_preallocated));
3198 	seq_printf(seq, "\tdiscarded: %u\n", atomic_read(&sbi->s_mb_discarded));
3199 	return 0;
3200 }
3201 
ext4_mb_seq_structs_summary_start(struct seq_file * seq,loff_t * pos)3202 static void *ext4_mb_seq_structs_summary_start(struct seq_file *seq, loff_t *pos)
3203 {
3204 	struct super_block *sb = pde_data(file_inode(seq->file));
3205 	unsigned long position;
3206 
3207 	if (*pos < 0 || *pos >= 2*MB_NUM_ORDERS(sb))
3208 		return NULL;
3209 	position = *pos + 1;
3210 	return (void *) ((unsigned long) position);
3211 }
3212 
ext4_mb_seq_structs_summary_next(struct seq_file * seq,void * v,loff_t * pos)3213 static void *ext4_mb_seq_structs_summary_next(struct seq_file *seq, void *v, loff_t *pos)
3214 {
3215 	struct super_block *sb = pde_data(file_inode(seq->file));
3216 	unsigned long position;
3217 
3218 	++*pos;
3219 	if (*pos < 0 || *pos >= 2*MB_NUM_ORDERS(sb))
3220 		return NULL;
3221 	position = *pos + 1;
3222 	return (void *) ((unsigned long) position);
3223 }
3224 
ext4_mb_seq_structs_summary_show(struct seq_file * seq,void * v)3225 static int ext4_mb_seq_structs_summary_show(struct seq_file *seq, void *v)
3226 {
3227 	struct super_block *sb = pde_data(file_inode(seq->file));
3228 	struct ext4_sb_info *sbi = EXT4_SB(sb);
3229 	unsigned long position = ((unsigned long) v);
3230 	struct ext4_group_info *grp;
3231 	unsigned int count;
3232 
3233 	position--;
3234 	if (position >= MB_NUM_ORDERS(sb)) {
3235 		position -= MB_NUM_ORDERS(sb);
3236 		if (position == 0)
3237 			seq_puts(seq, "avg_fragment_size_lists:\n");
3238 
3239 		count = 0;
3240 		read_lock(&sbi->s_mb_avg_fragment_size_locks[position]);
3241 		list_for_each_entry(grp, &sbi->s_mb_avg_fragment_size[position],
3242 				    bb_avg_fragment_size_node)
3243 			count++;
3244 		read_unlock(&sbi->s_mb_avg_fragment_size_locks[position]);
3245 		seq_printf(seq, "\tlist_order_%u_groups: %u\n",
3246 					(unsigned int)position, count);
3247 		return 0;
3248 	}
3249 
3250 	if (position == 0) {
3251 		seq_printf(seq, "optimize_scan: %d\n",
3252 			   test_opt2(sb, MB_OPTIMIZE_SCAN) ? 1 : 0);
3253 		seq_puts(seq, "max_free_order_lists:\n");
3254 	}
3255 	count = 0;
3256 	read_lock(&sbi->s_mb_largest_free_orders_locks[position]);
3257 	list_for_each_entry(grp, &sbi->s_mb_largest_free_orders[position],
3258 			    bb_largest_free_order_node)
3259 		count++;
3260 	read_unlock(&sbi->s_mb_largest_free_orders_locks[position]);
3261 	seq_printf(seq, "\tlist_order_%u_groups: %u\n",
3262 		   (unsigned int)position, count);
3263 
3264 	return 0;
3265 }
3266 
ext4_mb_seq_structs_summary_stop(struct seq_file * seq,void * v)3267 static void ext4_mb_seq_structs_summary_stop(struct seq_file *seq, void *v)
3268 {
3269 }
3270 
3271 const struct seq_operations ext4_mb_seq_structs_summary_ops = {
3272 	.start  = ext4_mb_seq_structs_summary_start,
3273 	.next   = ext4_mb_seq_structs_summary_next,
3274 	.stop   = ext4_mb_seq_structs_summary_stop,
3275 	.show   = ext4_mb_seq_structs_summary_show,
3276 };
3277 
get_groupinfo_cache(int blocksize_bits)3278 static struct kmem_cache *get_groupinfo_cache(int blocksize_bits)
3279 {
3280 	int cache_index = blocksize_bits - EXT4_MIN_BLOCK_LOG_SIZE;
3281 	struct kmem_cache *cachep = ext4_groupinfo_caches[cache_index];
3282 
3283 	BUG_ON(!cachep);
3284 	return cachep;
3285 }
3286 
3287 /*
3288  * Allocate the top-level s_group_info array for the specified number
3289  * of groups
3290  */
ext4_mb_alloc_groupinfo(struct super_block * sb,ext4_group_t ngroups)3291 int ext4_mb_alloc_groupinfo(struct super_block *sb, ext4_group_t ngroups)
3292 {
3293 	struct ext4_sb_info *sbi = EXT4_SB(sb);
3294 	unsigned size;
3295 	struct ext4_group_info ***old_groupinfo, ***new_groupinfo;
3296 
3297 	size = (ngroups + EXT4_DESC_PER_BLOCK(sb) - 1) >>
3298 		EXT4_DESC_PER_BLOCK_BITS(sb);
3299 	if (size <= sbi->s_group_info_size)
3300 		return 0;
3301 
3302 	size = roundup_pow_of_two(sizeof(*sbi->s_group_info) * size);
3303 	new_groupinfo = kvzalloc(size, GFP_KERNEL);
3304 	if (!new_groupinfo) {
3305 		ext4_msg(sb, KERN_ERR, "can't allocate buddy meta group");
3306 		return -ENOMEM;
3307 	}
3308 	rcu_read_lock();
3309 	old_groupinfo = rcu_dereference(sbi->s_group_info);
3310 	if (old_groupinfo)
3311 		memcpy(new_groupinfo, old_groupinfo,
3312 		       sbi->s_group_info_size * sizeof(*sbi->s_group_info));
3313 	rcu_read_unlock();
3314 	rcu_assign_pointer(sbi->s_group_info, new_groupinfo);
3315 	sbi->s_group_info_size = size / sizeof(*sbi->s_group_info);
3316 	if (old_groupinfo)
3317 		ext4_kvfree_array_rcu(old_groupinfo);
3318 	ext4_debug("allocated s_groupinfo array for %d meta_bg's\n",
3319 		   sbi->s_group_info_size);
3320 	return 0;
3321 }
3322 
3323 /* Create and initialize ext4_group_info data for the given group. */
ext4_mb_add_groupinfo(struct super_block * sb,ext4_group_t group,struct ext4_group_desc * desc)3324 int ext4_mb_add_groupinfo(struct super_block *sb, ext4_group_t group,
3325 			  struct ext4_group_desc *desc)
3326 {
3327 	int i;
3328 	int metalen = 0;
3329 	int idx = group >> EXT4_DESC_PER_BLOCK_BITS(sb);
3330 	struct ext4_sb_info *sbi = EXT4_SB(sb);
3331 	struct ext4_group_info **meta_group_info;
3332 	struct kmem_cache *cachep = get_groupinfo_cache(sb->s_blocksize_bits);
3333 
3334 	/*
3335 	 * First check if this group is the first of a reserved block.
3336 	 * If it's true, we have to allocate a new table of pointers
3337 	 * to ext4_group_info structures
3338 	 */
3339 	if (group % EXT4_DESC_PER_BLOCK(sb) == 0) {
3340 		metalen = sizeof(*meta_group_info) <<
3341 			EXT4_DESC_PER_BLOCK_BITS(sb);
3342 		meta_group_info = kmalloc(metalen, GFP_NOFS);
3343 		if (meta_group_info == NULL) {
3344 			ext4_msg(sb, KERN_ERR, "can't allocate mem "
3345 				 "for a buddy group");
3346 			return -ENOMEM;
3347 		}
3348 		rcu_read_lock();
3349 		rcu_dereference(sbi->s_group_info)[idx] = meta_group_info;
3350 		rcu_read_unlock();
3351 	}
3352 
3353 	meta_group_info = sbi_array_rcu_deref(sbi, s_group_info, idx);
3354 	i = group & (EXT4_DESC_PER_BLOCK(sb) - 1);
3355 
3356 	meta_group_info[i] = kmem_cache_zalloc(cachep, GFP_NOFS);
3357 	if (meta_group_info[i] == NULL) {
3358 		ext4_msg(sb, KERN_ERR, "can't allocate buddy mem");
3359 		goto exit_group_info;
3360 	}
3361 	set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT,
3362 		&(meta_group_info[i]->bb_state));
3363 
3364 	/*
3365 	 * initialize bb_free to be able to skip
3366 	 * empty groups without initialization
3367 	 */
3368 	if (ext4_has_group_desc_csum(sb) &&
3369 	    (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
3370 		meta_group_info[i]->bb_free =
3371 			ext4_free_clusters_after_init(sb, group, desc);
3372 	} else {
3373 		meta_group_info[i]->bb_free =
3374 			ext4_free_group_clusters(sb, desc);
3375 	}
3376 
3377 	INIT_LIST_HEAD(&meta_group_info[i]->bb_prealloc_list);
3378 	init_rwsem(&meta_group_info[i]->alloc_sem);
3379 	meta_group_info[i]->bb_free_root = RB_ROOT;
3380 	INIT_LIST_HEAD(&meta_group_info[i]->bb_largest_free_order_node);
3381 	INIT_LIST_HEAD(&meta_group_info[i]->bb_avg_fragment_size_node);
3382 	meta_group_info[i]->bb_largest_free_order = -1;  /* uninit */
3383 	meta_group_info[i]->bb_avg_fragment_size_order = -1;  /* uninit */
3384 	meta_group_info[i]->bb_group = group;
3385 
3386 	mb_group_bb_bitmap_alloc(sb, meta_group_info[i], group);
3387 	return 0;
3388 
3389 exit_group_info:
3390 	/* If a meta_group_info table has been allocated, release it now */
3391 	if (group % EXT4_DESC_PER_BLOCK(sb) == 0) {
3392 		struct ext4_group_info ***group_info;
3393 
3394 		rcu_read_lock();
3395 		group_info = rcu_dereference(sbi->s_group_info);
3396 		kfree(group_info[idx]);
3397 		group_info[idx] = NULL;
3398 		rcu_read_unlock();
3399 	}
3400 	return -ENOMEM;
3401 } /* ext4_mb_add_groupinfo */
3402 
ext4_mb_init_backend(struct super_block * sb)3403 static int ext4_mb_init_backend(struct super_block *sb)
3404 {
3405 	ext4_group_t ngroups = ext4_get_groups_count(sb);
3406 	ext4_group_t i;
3407 	struct ext4_sb_info *sbi = EXT4_SB(sb);
3408 	int err;
3409 	struct ext4_group_desc *desc;
3410 	struct ext4_group_info ***group_info;
3411 	struct kmem_cache *cachep;
3412 
3413 	err = ext4_mb_alloc_groupinfo(sb, ngroups);
3414 	if (err)
3415 		return err;
3416 
3417 	sbi->s_buddy_cache = new_inode(sb);
3418 	if (sbi->s_buddy_cache == NULL) {
3419 		ext4_msg(sb, KERN_ERR, "can't get new inode");
3420 		goto err_freesgi;
3421 	}
3422 	/* To avoid potentially colliding with an valid on-disk inode number,
3423 	 * use EXT4_BAD_INO for the buddy cache inode number.  This inode is
3424 	 * not in the inode hash, so it should never be found by iget(), but
3425 	 * this will avoid confusion if it ever shows up during debugging. */
3426 	sbi->s_buddy_cache->i_ino = EXT4_BAD_INO;
3427 	EXT4_I(sbi->s_buddy_cache)->i_disksize = 0;
3428 	for (i = 0; i < ngroups; i++) {
3429 		cond_resched();
3430 		desc = ext4_get_group_desc(sb, i, NULL);
3431 		if (desc == NULL) {
3432 			ext4_msg(sb, KERN_ERR, "can't read descriptor %u", i);
3433 			goto err_freebuddy;
3434 		}
3435 		if (ext4_mb_add_groupinfo(sb, i, desc) != 0)
3436 			goto err_freebuddy;
3437 	}
3438 
3439 	if (ext4_has_feature_flex_bg(sb)) {
3440 		/* a single flex group is supposed to be read by a single IO.
3441 		 * 2 ^ s_log_groups_per_flex != UINT_MAX as s_mb_prefetch is
3442 		 * unsigned integer, so the maximum shift is 32.
3443 		 */
3444 		if (sbi->s_es->s_log_groups_per_flex >= 32) {
3445 			ext4_msg(sb, KERN_ERR, "too many log groups per flexible block group");
3446 			goto err_freebuddy;
3447 		}
3448 		sbi->s_mb_prefetch = min_t(uint, 1 << sbi->s_es->s_log_groups_per_flex,
3449 			BLK_MAX_SEGMENT_SIZE >> (sb->s_blocksize_bits - 9));
3450 		sbi->s_mb_prefetch *= 8; /* 8 prefetch IOs in flight at most */
3451 	} else {
3452 		sbi->s_mb_prefetch = 32;
3453 	}
3454 	if (sbi->s_mb_prefetch > ext4_get_groups_count(sb))
3455 		sbi->s_mb_prefetch = ext4_get_groups_count(sb);
3456 	/*
3457 	 * now many real IOs to prefetch within a single allocation at
3458 	 * CR_POWER2_ALIGNED. Given CR_POWER2_ALIGNED is an CPU-related
3459 	 * optimization we shouldn't try to load too many groups, at some point
3460 	 * we should start to use what we've got in memory.
3461 	 * with an average random access time 5ms, it'd take a second to get
3462 	 * 200 groups (* N with flex_bg), so let's make this limit 4
3463 	 */
3464 	sbi->s_mb_prefetch_limit = sbi->s_mb_prefetch * 4;
3465 	if (sbi->s_mb_prefetch_limit > ext4_get_groups_count(sb))
3466 		sbi->s_mb_prefetch_limit = ext4_get_groups_count(sb);
3467 
3468 	return 0;
3469 
3470 err_freebuddy:
3471 	cachep = get_groupinfo_cache(sb->s_blocksize_bits);
3472 	while (i-- > 0) {
3473 		struct ext4_group_info *grp = ext4_get_group_info(sb, i);
3474 
3475 		if (grp)
3476 			kmem_cache_free(cachep, grp);
3477 	}
3478 	i = sbi->s_group_info_size;
3479 	rcu_read_lock();
3480 	group_info = rcu_dereference(sbi->s_group_info);
3481 	while (i-- > 0)
3482 		kfree(group_info[i]);
3483 	rcu_read_unlock();
3484 	iput(sbi->s_buddy_cache);
3485 err_freesgi:
3486 	rcu_read_lock();
3487 	kvfree(rcu_dereference(sbi->s_group_info));
3488 	rcu_read_unlock();
3489 	return -ENOMEM;
3490 }
3491 
ext4_groupinfo_destroy_slabs(void)3492 static void ext4_groupinfo_destroy_slabs(void)
3493 {
3494 	int i;
3495 
3496 	for (i = 0; i < NR_GRPINFO_CACHES; i++) {
3497 		kmem_cache_destroy(ext4_groupinfo_caches[i]);
3498 		ext4_groupinfo_caches[i] = NULL;
3499 	}
3500 }
3501 
ext4_groupinfo_create_slab(size_t size)3502 static int ext4_groupinfo_create_slab(size_t size)
3503 {
3504 	static DEFINE_MUTEX(ext4_grpinfo_slab_create_mutex);
3505 	int slab_size;
3506 	int blocksize_bits = order_base_2(size);
3507 	int cache_index = blocksize_bits - EXT4_MIN_BLOCK_LOG_SIZE;
3508 	struct kmem_cache *cachep;
3509 
3510 	if (cache_index >= NR_GRPINFO_CACHES)
3511 		return -EINVAL;
3512 
3513 	if (unlikely(cache_index < 0))
3514 		cache_index = 0;
3515 
3516 	mutex_lock(&ext4_grpinfo_slab_create_mutex);
3517 	if (ext4_groupinfo_caches[cache_index]) {
3518 		mutex_unlock(&ext4_grpinfo_slab_create_mutex);
3519 		return 0;	/* Already created */
3520 	}
3521 
3522 	slab_size = offsetof(struct ext4_group_info,
3523 				bb_counters[blocksize_bits + 2]);
3524 
3525 	cachep = kmem_cache_create(ext4_groupinfo_slab_names[cache_index],
3526 					slab_size, 0, SLAB_RECLAIM_ACCOUNT,
3527 					NULL);
3528 
3529 	ext4_groupinfo_caches[cache_index] = cachep;
3530 
3531 	mutex_unlock(&ext4_grpinfo_slab_create_mutex);
3532 	if (!cachep) {
3533 		printk(KERN_EMERG
3534 		       "EXT4-fs: no memory for groupinfo slab cache\n");
3535 		return -ENOMEM;
3536 	}
3537 
3538 	return 0;
3539 }
3540 
ext4_discard_work(struct work_struct * work)3541 static void ext4_discard_work(struct work_struct *work)
3542 {
3543 	struct ext4_sb_info *sbi = container_of(work,
3544 			struct ext4_sb_info, s_discard_work);
3545 	struct super_block *sb = sbi->s_sb;
3546 	struct ext4_free_data *fd, *nfd;
3547 	struct ext4_buddy e4b;
3548 	LIST_HEAD(discard_list);
3549 	ext4_group_t grp, load_grp;
3550 	int err = 0;
3551 
3552 	spin_lock(&sbi->s_md_lock);
3553 	list_splice_init(&sbi->s_discard_list, &discard_list);
3554 	spin_unlock(&sbi->s_md_lock);
3555 
3556 	load_grp = UINT_MAX;
3557 	list_for_each_entry_safe(fd, nfd, &discard_list, efd_list) {
3558 		/*
3559 		 * If filesystem is umounting or no memory or suffering
3560 		 * from no space, give up the discard
3561 		 */
3562 		if ((sb->s_flags & SB_ACTIVE) && !err &&
3563 		    !atomic_read(&sbi->s_retry_alloc_pending)) {
3564 			grp = fd->efd_group;
3565 			if (grp != load_grp) {
3566 				if (load_grp != UINT_MAX)
3567 					ext4_mb_unload_buddy(&e4b);
3568 
3569 				err = ext4_mb_load_buddy(sb, grp, &e4b);
3570 				if (err) {
3571 					kmem_cache_free(ext4_free_data_cachep, fd);
3572 					load_grp = UINT_MAX;
3573 					continue;
3574 				} else {
3575 					load_grp = grp;
3576 				}
3577 			}
3578 
3579 			ext4_lock_group(sb, grp);
3580 			ext4_try_to_trim_range(sb, &e4b, fd->efd_start_cluster,
3581 						fd->efd_start_cluster + fd->efd_count - 1, 1);
3582 			ext4_unlock_group(sb, grp);
3583 		}
3584 		kmem_cache_free(ext4_free_data_cachep, fd);
3585 	}
3586 
3587 	if (load_grp != UINT_MAX)
3588 		ext4_mb_unload_buddy(&e4b);
3589 }
3590 
ext4_mb_init(struct super_block * sb)3591 int ext4_mb_init(struct super_block *sb)
3592 {
3593 	struct ext4_sb_info *sbi = EXT4_SB(sb);
3594 	unsigned i, j;
3595 	unsigned offset, offset_incr;
3596 	unsigned max;
3597 	int ret;
3598 
3599 	i = MB_NUM_ORDERS(sb) * sizeof(*sbi->s_mb_offsets);
3600 
3601 	sbi->s_mb_offsets = kmalloc(i, GFP_KERNEL);
3602 	if (sbi->s_mb_offsets == NULL) {
3603 		ret = -ENOMEM;
3604 		goto out;
3605 	}
3606 
3607 	i = MB_NUM_ORDERS(sb) * sizeof(*sbi->s_mb_maxs);
3608 	sbi->s_mb_maxs = kmalloc(i, GFP_KERNEL);
3609 	if (sbi->s_mb_maxs == NULL) {
3610 		ret = -ENOMEM;
3611 		goto out;
3612 	}
3613 
3614 	ret = ext4_groupinfo_create_slab(sb->s_blocksize);
3615 	if (ret < 0)
3616 		goto out;
3617 
3618 	/* order 0 is regular bitmap */
3619 	sbi->s_mb_maxs[0] = sb->s_blocksize << 3;
3620 	sbi->s_mb_offsets[0] = 0;
3621 
3622 	i = 1;
3623 	offset = 0;
3624 	offset_incr = 1 << (sb->s_blocksize_bits - 1);
3625 	max = sb->s_blocksize << 2;
3626 	do {
3627 		sbi->s_mb_offsets[i] = offset;
3628 		sbi->s_mb_maxs[i] = max;
3629 		offset += offset_incr;
3630 		offset_incr = offset_incr >> 1;
3631 		max = max >> 1;
3632 		i++;
3633 	} while (i < MB_NUM_ORDERS(sb));
3634 
3635 	sbi->s_mb_avg_fragment_size =
3636 		kmalloc_array(MB_NUM_ORDERS(sb), sizeof(struct list_head),
3637 			GFP_KERNEL);
3638 	if (!sbi->s_mb_avg_fragment_size) {
3639 		ret = -ENOMEM;
3640 		goto out;
3641 	}
3642 	sbi->s_mb_avg_fragment_size_locks =
3643 		kmalloc_array(MB_NUM_ORDERS(sb), sizeof(rwlock_t),
3644 			GFP_KERNEL);
3645 	if (!sbi->s_mb_avg_fragment_size_locks) {
3646 		ret = -ENOMEM;
3647 		goto out;
3648 	}
3649 	for (i = 0; i < MB_NUM_ORDERS(sb); i++) {
3650 		INIT_LIST_HEAD(&sbi->s_mb_avg_fragment_size[i]);
3651 		rwlock_init(&sbi->s_mb_avg_fragment_size_locks[i]);
3652 	}
3653 	sbi->s_mb_largest_free_orders =
3654 		kmalloc_array(MB_NUM_ORDERS(sb), sizeof(struct list_head),
3655 			GFP_KERNEL);
3656 	if (!sbi->s_mb_largest_free_orders) {
3657 		ret = -ENOMEM;
3658 		goto out;
3659 	}
3660 	sbi->s_mb_largest_free_orders_locks =
3661 		kmalloc_array(MB_NUM_ORDERS(sb), sizeof(rwlock_t),
3662 			GFP_KERNEL);
3663 	if (!sbi->s_mb_largest_free_orders_locks) {
3664 		ret = -ENOMEM;
3665 		goto out;
3666 	}
3667 	for (i = 0; i < MB_NUM_ORDERS(sb); i++) {
3668 		INIT_LIST_HEAD(&sbi->s_mb_largest_free_orders[i]);
3669 		rwlock_init(&sbi->s_mb_largest_free_orders_locks[i]);
3670 	}
3671 
3672 	spin_lock_init(&sbi->s_md_lock);
3673 	sbi->s_mb_free_pending = 0;
3674 	INIT_LIST_HEAD(&sbi->s_freed_data_list[0]);
3675 	INIT_LIST_HEAD(&sbi->s_freed_data_list[1]);
3676 	INIT_LIST_HEAD(&sbi->s_discard_list);
3677 	INIT_WORK(&sbi->s_discard_work, ext4_discard_work);
3678 	atomic_set(&sbi->s_retry_alloc_pending, 0);
3679 
3680 	sbi->s_mb_max_to_scan = MB_DEFAULT_MAX_TO_SCAN;
3681 	sbi->s_mb_min_to_scan = MB_DEFAULT_MIN_TO_SCAN;
3682 	sbi->s_mb_stats = MB_DEFAULT_STATS;
3683 	sbi->s_mb_stream_request = MB_DEFAULT_STREAM_THRESHOLD;
3684 	sbi->s_mb_order2_reqs = MB_DEFAULT_ORDER2_REQS;
3685 	sbi->s_mb_best_avail_max_trim_order = MB_DEFAULT_BEST_AVAIL_TRIM_ORDER;
3686 
3687 	/*
3688 	 * The default group preallocation is 512, which for 4k block
3689 	 * sizes translates to 2 megabytes.  However for bigalloc file
3690 	 * systems, this is probably too big (i.e, if the cluster size
3691 	 * is 1 megabyte, then group preallocation size becomes half a
3692 	 * gigabyte!).  As a default, we will keep a two megabyte
3693 	 * group pralloc size for cluster sizes up to 64k, and after
3694 	 * that, we will force a minimum group preallocation size of
3695 	 * 32 clusters.  This translates to 8 megs when the cluster
3696 	 * size is 256k, and 32 megs when the cluster size is 1 meg,
3697 	 * which seems reasonable as a default.
3698 	 */
3699 	sbi->s_mb_group_prealloc = max(MB_DEFAULT_GROUP_PREALLOC >>
3700 				       sbi->s_cluster_bits, 32);
3701 	/*
3702 	 * If there is a s_stripe > 1, then we set the s_mb_group_prealloc
3703 	 * to the lowest multiple of s_stripe which is bigger than
3704 	 * the s_mb_group_prealloc as determined above. We want
3705 	 * the preallocation size to be an exact multiple of the
3706 	 * RAID stripe size so that preallocations don't fragment
3707 	 * the stripes.
3708 	 */
3709 	if (sbi->s_stripe > 1) {
3710 		sbi->s_mb_group_prealloc = roundup(
3711 			sbi->s_mb_group_prealloc, EXT4_NUM_B2C(sbi, sbi->s_stripe));
3712 	}
3713 
3714 	sbi->s_locality_groups = alloc_percpu(struct ext4_locality_group);
3715 	if (sbi->s_locality_groups == NULL) {
3716 		ret = -ENOMEM;
3717 		goto out;
3718 	}
3719 	for_each_possible_cpu(i) {
3720 		struct ext4_locality_group *lg;
3721 		lg = per_cpu_ptr(sbi->s_locality_groups, i);
3722 		mutex_init(&lg->lg_mutex);
3723 		for (j = 0; j < PREALLOC_TB_SIZE; j++)
3724 			INIT_LIST_HEAD(&lg->lg_prealloc_list[j]);
3725 		spin_lock_init(&lg->lg_prealloc_lock);
3726 	}
3727 
3728 	if (bdev_nonrot(sb->s_bdev))
3729 		sbi->s_mb_max_linear_groups = 0;
3730 	else
3731 		sbi->s_mb_max_linear_groups = MB_DEFAULT_LINEAR_LIMIT;
3732 	/* init file for buddy data */
3733 	ret = ext4_mb_init_backend(sb);
3734 	if (ret != 0)
3735 		goto out_free_locality_groups;
3736 
3737 	return 0;
3738 
3739 out_free_locality_groups:
3740 	free_percpu(sbi->s_locality_groups);
3741 	sbi->s_locality_groups = NULL;
3742 out:
3743 	kfree(sbi->s_mb_avg_fragment_size);
3744 	kfree(sbi->s_mb_avg_fragment_size_locks);
3745 	kfree(sbi->s_mb_largest_free_orders);
3746 	kfree(sbi->s_mb_largest_free_orders_locks);
3747 	kfree(sbi->s_mb_offsets);
3748 	sbi->s_mb_offsets = NULL;
3749 	kfree(sbi->s_mb_maxs);
3750 	sbi->s_mb_maxs = NULL;
3751 	return ret;
3752 }
3753 
3754 /* need to called with the ext4 group lock held */
ext4_mb_cleanup_pa(struct ext4_group_info * grp)3755 static int ext4_mb_cleanup_pa(struct ext4_group_info *grp)
3756 {
3757 	struct ext4_prealloc_space *pa;
3758 	struct list_head *cur, *tmp;
3759 	int count = 0;
3760 
3761 	list_for_each_safe(cur, tmp, &grp->bb_prealloc_list) {
3762 		pa = list_entry(cur, struct ext4_prealloc_space, pa_group_list);
3763 		list_del(&pa->pa_group_list);
3764 		count++;
3765 		kmem_cache_free(ext4_pspace_cachep, pa);
3766 	}
3767 	return count;
3768 }
3769 
ext4_mb_release(struct super_block * sb)3770 void ext4_mb_release(struct super_block *sb)
3771 {
3772 	ext4_group_t ngroups = ext4_get_groups_count(sb);
3773 	ext4_group_t i;
3774 	int num_meta_group_infos;
3775 	struct ext4_group_info *grinfo, ***group_info;
3776 	struct ext4_sb_info *sbi = EXT4_SB(sb);
3777 	struct kmem_cache *cachep = get_groupinfo_cache(sb->s_blocksize_bits);
3778 	int count;
3779 
3780 	if (test_opt(sb, DISCARD)) {
3781 		/*
3782 		 * wait the discard work to drain all of ext4_free_data
3783 		 */
3784 		flush_work(&sbi->s_discard_work);
3785 		WARN_ON_ONCE(!list_empty(&sbi->s_discard_list));
3786 	}
3787 
3788 	if (sbi->s_group_info) {
3789 		for (i = 0; i < ngroups; i++) {
3790 			cond_resched();
3791 			grinfo = ext4_get_group_info(sb, i);
3792 			if (!grinfo)
3793 				continue;
3794 			mb_group_bb_bitmap_free(grinfo);
3795 			ext4_lock_group(sb, i);
3796 			count = ext4_mb_cleanup_pa(grinfo);
3797 			if (count)
3798 				mb_debug(sb, "mballoc: %d PAs left\n",
3799 					 count);
3800 			ext4_unlock_group(sb, i);
3801 			kmem_cache_free(cachep, grinfo);
3802 		}
3803 		num_meta_group_infos = (ngroups +
3804 				EXT4_DESC_PER_BLOCK(sb) - 1) >>
3805 			EXT4_DESC_PER_BLOCK_BITS(sb);
3806 		rcu_read_lock();
3807 		group_info = rcu_dereference(sbi->s_group_info);
3808 		for (i = 0; i < num_meta_group_infos; i++)
3809 			kfree(group_info[i]);
3810 		kvfree(group_info);
3811 		rcu_read_unlock();
3812 	}
3813 	kfree(sbi->s_mb_avg_fragment_size);
3814 	kfree(sbi->s_mb_avg_fragment_size_locks);
3815 	kfree(sbi->s_mb_largest_free_orders);
3816 	kfree(sbi->s_mb_largest_free_orders_locks);
3817 	kfree(sbi->s_mb_offsets);
3818 	kfree(sbi->s_mb_maxs);
3819 	iput(sbi->s_buddy_cache);
3820 	if (sbi->s_mb_stats) {
3821 		ext4_msg(sb, KERN_INFO,
3822 		       "mballoc: %u blocks %u reqs (%u success)",
3823 				atomic_read(&sbi->s_bal_allocated),
3824 				atomic_read(&sbi->s_bal_reqs),
3825 				atomic_read(&sbi->s_bal_success));
3826 		ext4_msg(sb, KERN_INFO,
3827 		      "mballoc: %u extents scanned, %u groups scanned, %u goal hits, "
3828 				"%u 2^N hits, %u breaks, %u lost",
3829 				atomic_read(&sbi->s_bal_ex_scanned),
3830 				atomic_read(&sbi->s_bal_groups_scanned),
3831 				atomic_read(&sbi->s_bal_goals),
3832 				atomic_read(&sbi->s_bal_2orders),
3833 				atomic_read(&sbi->s_bal_breaks),
3834 				atomic_read(&sbi->s_mb_lost_chunks));
3835 		ext4_msg(sb, KERN_INFO,
3836 		       "mballoc: %u generated and it took %llu",
3837 				atomic_read(&sbi->s_mb_buddies_generated),
3838 				atomic64_read(&sbi->s_mb_generation_time));
3839 		ext4_msg(sb, KERN_INFO,
3840 		       "mballoc: %u preallocated, %u discarded",
3841 				atomic_read(&sbi->s_mb_preallocated),
3842 				atomic_read(&sbi->s_mb_discarded));
3843 	}
3844 
3845 	free_percpu(sbi->s_locality_groups);
3846 }
3847 
ext4_issue_discard(struct super_block * sb,ext4_group_t block_group,ext4_grpblk_t cluster,int count)3848 static inline int ext4_issue_discard(struct super_block *sb,
3849 		ext4_group_t block_group, ext4_grpblk_t cluster, int count)
3850 {
3851 	ext4_fsblk_t discard_block;
3852 
3853 	discard_block = (EXT4_C2B(EXT4_SB(sb), cluster) +
3854 			 ext4_group_first_block_no(sb, block_group));
3855 	count = EXT4_C2B(EXT4_SB(sb), count);
3856 	trace_ext4_discard_blocks(sb,
3857 			(unsigned long long) discard_block, count);
3858 
3859 	return sb_issue_discard(sb, discard_block, count, GFP_NOFS, 0);
3860 }
3861 
ext4_free_data_in_buddy(struct super_block * sb,struct ext4_free_data * entry)3862 static void ext4_free_data_in_buddy(struct super_block *sb,
3863 				    struct ext4_free_data *entry)
3864 {
3865 	struct ext4_buddy e4b;
3866 	struct ext4_group_info *db;
3867 	int err, count = 0;
3868 
3869 	mb_debug(sb, "gonna free %u blocks in group %u (0x%p):",
3870 		 entry->efd_count, entry->efd_group, entry);
3871 
3872 	err = ext4_mb_load_buddy(sb, entry->efd_group, &e4b);
3873 	/* we expect to find existing buddy because it's pinned */
3874 	BUG_ON(err != 0);
3875 
3876 	spin_lock(&EXT4_SB(sb)->s_md_lock);
3877 	EXT4_SB(sb)->s_mb_free_pending -= entry->efd_count;
3878 	spin_unlock(&EXT4_SB(sb)->s_md_lock);
3879 
3880 	db = e4b.bd_info;
3881 	/* there are blocks to put in buddy to make them really free */
3882 	count += entry->efd_count;
3883 	ext4_lock_group(sb, entry->efd_group);
3884 	/* Take it out of per group rb tree */
3885 	rb_erase(&entry->efd_node, &(db->bb_free_root));
3886 	mb_free_blocks(NULL, &e4b, entry->efd_start_cluster, entry->efd_count);
3887 
3888 	/*
3889 	 * Clear the trimmed flag for the group so that the next
3890 	 * ext4_trim_fs can trim it.
3891 	 */
3892 	EXT4_MB_GRP_CLEAR_TRIMMED(db);
3893 
3894 	if (!db->bb_free_root.rb_node) {
3895 		/* No more items in the per group rb tree
3896 		 * balance refcounts from ext4_mb_free_metadata()
3897 		 */
3898 		folio_put(e4b.bd_buddy_folio);
3899 		folio_put(e4b.bd_bitmap_folio);
3900 	}
3901 	ext4_unlock_group(sb, entry->efd_group);
3902 	ext4_mb_unload_buddy(&e4b);
3903 
3904 	mb_debug(sb, "freed %d blocks in 1 structures\n", count);
3905 }
3906 
3907 /*
3908  * This function is called by the jbd2 layer once the commit has finished,
3909  * so we know we can free the blocks that were released with that commit.
3910  */
ext4_process_freed_data(struct super_block * sb,tid_t commit_tid)3911 void ext4_process_freed_data(struct super_block *sb, tid_t commit_tid)
3912 {
3913 	struct ext4_sb_info *sbi = EXT4_SB(sb);
3914 	struct ext4_free_data *entry, *tmp;
3915 	LIST_HEAD(freed_data_list);
3916 	struct list_head *s_freed_head = &sbi->s_freed_data_list[commit_tid & 1];
3917 	bool wake;
3918 
3919 	list_replace_init(s_freed_head, &freed_data_list);
3920 
3921 	list_for_each_entry(entry, &freed_data_list, efd_list)
3922 		ext4_free_data_in_buddy(sb, entry);
3923 
3924 	if (test_opt(sb, DISCARD)) {
3925 		spin_lock(&sbi->s_md_lock);
3926 		wake = list_empty(&sbi->s_discard_list);
3927 		list_splice_tail(&freed_data_list, &sbi->s_discard_list);
3928 		spin_unlock(&sbi->s_md_lock);
3929 		if (wake)
3930 			queue_work(system_unbound_wq, &sbi->s_discard_work);
3931 	} else {
3932 		list_for_each_entry_safe(entry, tmp, &freed_data_list, efd_list)
3933 			kmem_cache_free(ext4_free_data_cachep, entry);
3934 	}
3935 }
3936 
ext4_init_mballoc(void)3937 int __init ext4_init_mballoc(void)
3938 {
3939 	ext4_pspace_cachep = KMEM_CACHE(ext4_prealloc_space,
3940 					SLAB_RECLAIM_ACCOUNT);
3941 	if (ext4_pspace_cachep == NULL)
3942 		goto out;
3943 
3944 	ext4_ac_cachep = KMEM_CACHE(ext4_allocation_context,
3945 				    SLAB_RECLAIM_ACCOUNT);
3946 	if (ext4_ac_cachep == NULL)
3947 		goto out_pa_free;
3948 
3949 	ext4_free_data_cachep = KMEM_CACHE(ext4_free_data,
3950 					   SLAB_RECLAIM_ACCOUNT);
3951 	if (ext4_free_data_cachep == NULL)
3952 		goto out_ac_free;
3953 
3954 	return 0;
3955 
3956 out_ac_free:
3957 	kmem_cache_destroy(ext4_ac_cachep);
3958 out_pa_free:
3959 	kmem_cache_destroy(ext4_pspace_cachep);
3960 out:
3961 	return -ENOMEM;
3962 }
3963 
ext4_exit_mballoc(void)3964 void ext4_exit_mballoc(void)
3965 {
3966 	/*
3967 	 * Wait for completion of call_rcu()'s on ext4_pspace_cachep
3968 	 * before destroying the slab cache.
3969 	 */
3970 	rcu_barrier();
3971 	kmem_cache_destroy(ext4_pspace_cachep);
3972 	kmem_cache_destroy(ext4_ac_cachep);
3973 	kmem_cache_destroy(ext4_free_data_cachep);
3974 	ext4_groupinfo_destroy_slabs();
3975 }
3976 
3977 #define EXT4_MB_BITMAP_MARKED_CHECK 0x0001
3978 #define EXT4_MB_SYNC_UPDATE 0x0002
3979 static int
ext4_mb_mark_context(handle_t * handle,struct super_block * sb,bool state,ext4_group_t group,ext4_grpblk_t blkoff,ext4_grpblk_t len,int flags,ext4_grpblk_t * ret_changed)3980 ext4_mb_mark_context(handle_t *handle, struct super_block *sb, bool state,
3981 		     ext4_group_t group, ext4_grpblk_t blkoff,
3982 		     ext4_grpblk_t len, int flags, ext4_grpblk_t *ret_changed)
3983 {
3984 	struct ext4_sb_info *sbi = EXT4_SB(sb);
3985 	struct buffer_head *bitmap_bh = NULL;
3986 	struct ext4_group_desc *gdp;
3987 	struct buffer_head *gdp_bh;
3988 	int err;
3989 	unsigned int i, already, changed = len;
3990 
3991 	KUNIT_STATIC_STUB_REDIRECT(ext4_mb_mark_context,
3992 				   handle, sb, state, group, blkoff, len,
3993 				   flags, ret_changed);
3994 
3995 	if (ret_changed)
3996 		*ret_changed = 0;
3997 	bitmap_bh = ext4_read_block_bitmap(sb, group);
3998 	if (IS_ERR(bitmap_bh))
3999 		return PTR_ERR(bitmap_bh);
4000 
4001 	if (handle) {
4002 		BUFFER_TRACE(bitmap_bh, "getting write access");
4003 		err = ext4_journal_get_write_access(handle, sb, bitmap_bh,
4004 						    EXT4_JTR_NONE);
4005 		if (err)
4006 			goto out_err;
4007 	}
4008 
4009 	err = -EIO;
4010 	gdp = ext4_get_group_desc(sb, group, &gdp_bh);
4011 	if (!gdp)
4012 		goto out_err;
4013 
4014 	if (handle) {
4015 		BUFFER_TRACE(gdp_bh, "get_write_access");
4016 		err = ext4_journal_get_write_access(handle, sb, gdp_bh,
4017 						    EXT4_JTR_NONE);
4018 		if (err)
4019 			goto out_err;
4020 	}
4021 
4022 	ext4_lock_group(sb, group);
4023 	if (ext4_has_group_desc_csum(sb) &&
4024 	    (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
4025 		gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
4026 		ext4_free_group_clusters_set(sb, gdp,
4027 			ext4_free_clusters_after_init(sb, group, gdp));
4028 	}
4029 
4030 	if (flags & EXT4_MB_BITMAP_MARKED_CHECK) {
4031 		already = 0;
4032 		for (i = 0; i < len; i++)
4033 			if (mb_test_bit(blkoff + i, bitmap_bh->b_data) ==
4034 					state)
4035 				already++;
4036 		changed = len - already;
4037 	}
4038 
4039 	if (state) {
4040 		mb_set_bits(bitmap_bh->b_data, blkoff, len);
4041 		ext4_free_group_clusters_set(sb, gdp,
4042 			ext4_free_group_clusters(sb, gdp) - changed);
4043 	} else {
4044 		mb_clear_bits(bitmap_bh->b_data, blkoff, len);
4045 		ext4_free_group_clusters_set(sb, gdp,
4046 			ext4_free_group_clusters(sb, gdp) + changed);
4047 	}
4048 
4049 	ext4_block_bitmap_csum_set(sb, gdp, bitmap_bh);
4050 	ext4_group_desc_csum_set(sb, group, gdp);
4051 	ext4_unlock_group(sb, group);
4052 	if (ret_changed)
4053 		*ret_changed = changed;
4054 
4055 	if (sbi->s_log_groups_per_flex) {
4056 		ext4_group_t flex_group = ext4_flex_group(sbi, group);
4057 		struct flex_groups *fg = sbi_array_rcu_deref(sbi,
4058 					   s_flex_groups, flex_group);
4059 
4060 		if (state)
4061 			atomic64_sub(changed, &fg->free_clusters);
4062 		else
4063 			atomic64_add(changed, &fg->free_clusters);
4064 	}
4065 
4066 	err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
4067 	if (err)
4068 		goto out_err;
4069 	err = ext4_handle_dirty_metadata(handle, NULL, gdp_bh);
4070 	if (err)
4071 		goto out_err;
4072 
4073 	if (flags & EXT4_MB_SYNC_UPDATE) {
4074 		sync_dirty_buffer(bitmap_bh);
4075 		sync_dirty_buffer(gdp_bh);
4076 	}
4077 
4078 out_err:
4079 	brelse(bitmap_bh);
4080 	return err;
4081 }
4082 
4083 /*
4084  * Check quota and mark chosen space (ac->ac_b_ex) non-free in bitmaps
4085  * Returns 0 if success or error code
4086  */
4087 static noinline_for_stack int
ext4_mb_mark_diskspace_used(struct ext4_allocation_context * ac,handle_t * handle,unsigned int reserv_clstrs)4088 ext4_mb_mark_diskspace_used(struct ext4_allocation_context *ac,
4089 				handle_t *handle, unsigned int reserv_clstrs)
4090 {
4091 	struct ext4_group_desc *gdp;
4092 	struct ext4_sb_info *sbi;
4093 	struct super_block *sb;
4094 	ext4_fsblk_t block;
4095 	int err, len;
4096 	int flags = 0;
4097 	ext4_grpblk_t changed;
4098 
4099 	BUG_ON(ac->ac_status != AC_STATUS_FOUND);
4100 	BUG_ON(ac->ac_b_ex.fe_len <= 0);
4101 
4102 	sb = ac->ac_sb;
4103 	sbi = EXT4_SB(sb);
4104 
4105 	gdp = ext4_get_group_desc(sb, ac->ac_b_ex.fe_group, NULL);
4106 	if (!gdp)
4107 		return -EIO;
4108 	ext4_debug("using block group %u(%d)\n", ac->ac_b_ex.fe_group,
4109 			ext4_free_group_clusters(sb, gdp));
4110 
4111 	block = ext4_grp_offs_to_block(sb, &ac->ac_b_ex);
4112 	len = EXT4_C2B(sbi, ac->ac_b_ex.fe_len);
4113 	if (!ext4_inode_block_valid(ac->ac_inode, block, len)) {
4114 		ext4_error(sb, "Allocating blocks %llu-%llu which overlap "
4115 			   "fs metadata", block, block+len);
4116 		/* File system mounted not to panic on error
4117 		 * Fix the bitmap and return EFSCORRUPTED
4118 		 * We leak some of the blocks here.
4119 		 */
4120 		err = ext4_mb_mark_context(handle, sb, true,
4121 					   ac->ac_b_ex.fe_group,
4122 					   ac->ac_b_ex.fe_start,
4123 					   ac->ac_b_ex.fe_len,
4124 					   0, NULL);
4125 		if (!err)
4126 			err = -EFSCORRUPTED;
4127 		return err;
4128 	}
4129 
4130 #ifdef AGGRESSIVE_CHECK
4131 	flags |= EXT4_MB_BITMAP_MARKED_CHECK;
4132 #endif
4133 	err = ext4_mb_mark_context(handle, sb, true, ac->ac_b_ex.fe_group,
4134 				   ac->ac_b_ex.fe_start, ac->ac_b_ex.fe_len,
4135 				   flags, &changed);
4136 
4137 	if (err && changed == 0)
4138 		return err;
4139 
4140 #ifdef AGGRESSIVE_CHECK
4141 	BUG_ON(changed != ac->ac_b_ex.fe_len);
4142 #endif
4143 	percpu_counter_sub(&sbi->s_freeclusters_counter, ac->ac_b_ex.fe_len);
4144 	/*
4145 	 * Now reduce the dirty block count also. Should not go negative
4146 	 */
4147 	if (!(ac->ac_flags & EXT4_MB_DELALLOC_RESERVED))
4148 		/* release all the reserved blocks if non delalloc */
4149 		percpu_counter_sub(&sbi->s_dirtyclusters_counter,
4150 				   reserv_clstrs);
4151 
4152 	return err;
4153 }
4154 
4155 /*
4156  * Idempotent helper for Ext4 fast commit replay path to set the state of
4157  * blocks in bitmaps and update counters.
4158  */
ext4_mb_mark_bb(struct super_block * sb,ext4_fsblk_t block,int len,bool state)4159 void ext4_mb_mark_bb(struct super_block *sb, ext4_fsblk_t block,
4160 		     int len, bool state)
4161 {
4162 	struct ext4_sb_info *sbi = EXT4_SB(sb);
4163 	ext4_group_t group;
4164 	ext4_grpblk_t blkoff;
4165 	int err = 0;
4166 	unsigned int clen, thisgrp_len;
4167 
4168 	while (len > 0) {
4169 		ext4_get_group_no_and_offset(sb, block, &group, &blkoff);
4170 
4171 		/*
4172 		 * Check to see if we are freeing blocks across a group
4173 		 * boundary.
4174 		 * In case of flex_bg, this can happen that (block, len) may
4175 		 * span across more than one group. In that case we need to
4176 		 * get the corresponding group metadata to work with.
4177 		 * For this we have goto again loop.
4178 		 */
4179 		thisgrp_len = min_t(unsigned int, (unsigned int)len,
4180 			EXT4_BLOCKS_PER_GROUP(sb) - EXT4_C2B(sbi, blkoff));
4181 		clen = EXT4_NUM_B2C(sbi, thisgrp_len);
4182 
4183 		if (!ext4_sb_block_valid(sb, NULL, block, thisgrp_len)) {
4184 			ext4_error(sb, "Marking blocks in system zone - "
4185 				   "Block = %llu, len = %u",
4186 				   block, thisgrp_len);
4187 			break;
4188 		}
4189 
4190 		err = ext4_mb_mark_context(NULL, sb, state,
4191 					   group, blkoff, clen,
4192 					   EXT4_MB_BITMAP_MARKED_CHECK |
4193 					   EXT4_MB_SYNC_UPDATE,
4194 					   NULL);
4195 		if (err)
4196 			break;
4197 
4198 		block += thisgrp_len;
4199 		len -= thisgrp_len;
4200 		BUG_ON(len < 0);
4201 	}
4202 }
4203 
4204 /*
4205  * here we normalize request for locality group
4206  * Group request are normalized to s_mb_group_prealloc, which goes to
4207  * s_strip if we set the same via mount option.
4208  * s_mb_group_prealloc can be configured via
4209  * /sys/fs/ext4/<partition>/mb_group_prealloc
4210  *
4211  * XXX: should we try to preallocate more than the group has now?
4212  */
ext4_mb_normalize_group_request(struct ext4_allocation_context * ac)4213 static void ext4_mb_normalize_group_request(struct ext4_allocation_context *ac)
4214 {
4215 	struct super_block *sb = ac->ac_sb;
4216 	struct ext4_locality_group *lg = ac->ac_lg;
4217 
4218 	BUG_ON(lg == NULL);
4219 	ac->ac_g_ex.fe_len = EXT4_SB(sb)->s_mb_group_prealloc;
4220 	mb_debug(sb, "goal %u blocks for locality group\n", ac->ac_g_ex.fe_len);
4221 }
4222 
4223 /*
4224  * This function returns the next element to look at during inode
4225  * PA rbtree walk. We assume that we have held the inode PA rbtree lock
4226  * (ei->i_prealloc_lock)
4227  *
4228  * new_start	The start of the range we want to compare
4229  * cur_start	The existing start that we are comparing against
4230  * node	The node of the rb_tree
4231  */
4232 static inline struct rb_node*
ext4_mb_pa_rb_next_iter(ext4_lblk_t new_start,ext4_lblk_t cur_start,struct rb_node * node)4233 ext4_mb_pa_rb_next_iter(ext4_lblk_t new_start, ext4_lblk_t cur_start, struct rb_node *node)
4234 {
4235 	if (new_start < cur_start)
4236 		return node->rb_left;
4237 	else
4238 		return node->rb_right;
4239 }
4240 
4241 static inline void
ext4_mb_pa_assert_overlap(struct ext4_allocation_context * ac,ext4_lblk_t start,loff_t end)4242 ext4_mb_pa_assert_overlap(struct ext4_allocation_context *ac,
4243 			  ext4_lblk_t start, loff_t end)
4244 {
4245 	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
4246 	struct ext4_inode_info *ei = EXT4_I(ac->ac_inode);
4247 	struct ext4_prealloc_space *tmp_pa;
4248 	ext4_lblk_t tmp_pa_start;
4249 	loff_t tmp_pa_end;
4250 	struct rb_node *iter;
4251 
4252 	read_lock(&ei->i_prealloc_lock);
4253 	for (iter = ei->i_prealloc_node.rb_node; iter;
4254 	     iter = ext4_mb_pa_rb_next_iter(start, tmp_pa_start, iter)) {
4255 		tmp_pa = rb_entry(iter, struct ext4_prealloc_space,
4256 				  pa_node.inode_node);
4257 		tmp_pa_start = tmp_pa->pa_lstart;
4258 		tmp_pa_end = pa_logical_end(sbi, tmp_pa);
4259 
4260 		spin_lock(&tmp_pa->pa_lock);
4261 		if (tmp_pa->pa_deleted == 0)
4262 			BUG_ON(!(start >= tmp_pa_end || end <= tmp_pa_start));
4263 		spin_unlock(&tmp_pa->pa_lock);
4264 	}
4265 	read_unlock(&ei->i_prealloc_lock);
4266 }
4267 
4268 /*
4269  * Given an allocation context "ac" and a range "start", "end", check
4270  * and adjust boundaries if the range overlaps with any of the existing
4271  * preallocatoins stored in the corresponding inode of the allocation context.
4272  *
4273  * Parameters:
4274  *	ac			allocation context
4275  *	start			start of the new range
4276  *	end			end of the new range
4277  */
4278 static inline void
ext4_mb_pa_adjust_overlap(struct ext4_allocation_context * ac,ext4_lblk_t * start,loff_t * end)4279 ext4_mb_pa_adjust_overlap(struct ext4_allocation_context *ac,
4280 			  ext4_lblk_t *start, loff_t *end)
4281 {
4282 	struct ext4_inode_info *ei = EXT4_I(ac->ac_inode);
4283 	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
4284 	struct ext4_prealloc_space *tmp_pa = NULL, *left_pa = NULL, *right_pa = NULL;
4285 	struct rb_node *iter;
4286 	ext4_lblk_t new_start, tmp_pa_start, right_pa_start = -1;
4287 	loff_t new_end, tmp_pa_end, left_pa_end = -1;
4288 
4289 	new_start = *start;
4290 	new_end = *end;
4291 
4292 	/*
4293 	 * Adjust the normalized range so that it doesn't overlap with any
4294 	 * existing preallocated blocks(PAs). Make sure to hold the rbtree lock
4295 	 * so it doesn't change underneath us.
4296 	 */
4297 	read_lock(&ei->i_prealloc_lock);
4298 
4299 	/* Step 1: find any one immediate neighboring PA of the normalized range */
4300 	for (iter = ei->i_prealloc_node.rb_node; iter;
4301 	     iter = ext4_mb_pa_rb_next_iter(ac->ac_o_ex.fe_logical,
4302 					    tmp_pa_start, iter)) {
4303 		tmp_pa = rb_entry(iter, struct ext4_prealloc_space,
4304 				  pa_node.inode_node);
4305 		tmp_pa_start = tmp_pa->pa_lstart;
4306 		tmp_pa_end = pa_logical_end(sbi, tmp_pa);
4307 
4308 		/* PA must not overlap original request */
4309 		spin_lock(&tmp_pa->pa_lock);
4310 		if (tmp_pa->pa_deleted == 0)
4311 			BUG_ON(!(ac->ac_o_ex.fe_logical >= tmp_pa_end ||
4312 				 ac->ac_o_ex.fe_logical < tmp_pa_start));
4313 		spin_unlock(&tmp_pa->pa_lock);
4314 	}
4315 
4316 	/*
4317 	 * Step 2: check if the found PA is left or right neighbor and
4318 	 * get the other neighbor
4319 	 */
4320 	if (tmp_pa) {
4321 		if (tmp_pa->pa_lstart < ac->ac_o_ex.fe_logical) {
4322 			struct rb_node *tmp;
4323 
4324 			left_pa = tmp_pa;
4325 			tmp = rb_next(&left_pa->pa_node.inode_node);
4326 			if (tmp) {
4327 				right_pa = rb_entry(tmp,
4328 						    struct ext4_prealloc_space,
4329 						    pa_node.inode_node);
4330 			}
4331 		} else {
4332 			struct rb_node *tmp;
4333 
4334 			right_pa = tmp_pa;
4335 			tmp = rb_prev(&right_pa->pa_node.inode_node);
4336 			if (tmp) {
4337 				left_pa = rb_entry(tmp,
4338 						   struct ext4_prealloc_space,
4339 						   pa_node.inode_node);
4340 			}
4341 		}
4342 	}
4343 
4344 	/* Step 3: get the non deleted neighbors */
4345 	if (left_pa) {
4346 		for (iter = &left_pa->pa_node.inode_node;;
4347 		     iter = rb_prev(iter)) {
4348 			if (!iter) {
4349 				left_pa = NULL;
4350 				break;
4351 			}
4352 
4353 			tmp_pa = rb_entry(iter, struct ext4_prealloc_space,
4354 					  pa_node.inode_node);
4355 			left_pa = tmp_pa;
4356 			spin_lock(&tmp_pa->pa_lock);
4357 			if (tmp_pa->pa_deleted == 0) {
4358 				spin_unlock(&tmp_pa->pa_lock);
4359 				break;
4360 			}
4361 			spin_unlock(&tmp_pa->pa_lock);
4362 		}
4363 	}
4364 
4365 	if (right_pa) {
4366 		for (iter = &right_pa->pa_node.inode_node;;
4367 		     iter = rb_next(iter)) {
4368 			if (!iter) {
4369 				right_pa = NULL;
4370 				break;
4371 			}
4372 
4373 			tmp_pa = rb_entry(iter, struct ext4_prealloc_space,
4374 					  pa_node.inode_node);
4375 			right_pa = tmp_pa;
4376 			spin_lock(&tmp_pa->pa_lock);
4377 			if (tmp_pa->pa_deleted == 0) {
4378 				spin_unlock(&tmp_pa->pa_lock);
4379 				break;
4380 			}
4381 			spin_unlock(&tmp_pa->pa_lock);
4382 		}
4383 	}
4384 
4385 	if (left_pa) {
4386 		left_pa_end = pa_logical_end(sbi, left_pa);
4387 		BUG_ON(left_pa_end > ac->ac_o_ex.fe_logical);
4388 	}
4389 
4390 	if (right_pa) {
4391 		right_pa_start = right_pa->pa_lstart;
4392 		BUG_ON(right_pa_start <= ac->ac_o_ex.fe_logical);
4393 	}
4394 
4395 	/* Step 4: trim our normalized range to not overlap with the neighbors */
4396 	if (left_pa) {
4397 		if (left_pa_end > new_start)
4398 			new_start = left_pa_end;
4399 	}
4400 
4401 	if (right_pa) {
4402 		if (right_pa_start < new_end)
4403 			new_end = right_pa_start;
4404 	}
4405 	read_unlock(&ei->i_prealloc_lock);
4406 
4407 	/* XXX: extra loop to check we really don't overlap preallocations */
4408 	ext4_mb_pa_assert_overlap(ac, new_start, new_end);
4409 
4410 	*start = new_start;
4411 	*end = new_end;
4412 }
4413 
4414 /*
4415  * Normalization means making request better in terms of
4416  * size and alignment
4417  */
4418 static noinline_for_stack void
ext4_mb_normalize_request(struct ext4_allocation_context * ac,struct ext4_allocation_request * ar)4419 ext4_mb_normalize_request(struct ext4_allocation_context *ac,
4420 				struct ext4_allocation_request *ar)
4421 {
4422 	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
4423 	struct ext4_super_block *es = sbi->s_es;
4424 	int bsbits, max;
4425 	loff_t size, start_off, end;
4426 	loff_t orig_size __maybe_unused;
4427 	ext4_lblk_t start;
4428 
4429 	/* do normalize only data requests, metadata requests
4430 	   do not need preallocation */
4431 	if (!(ac->ac_flags & EXT4_MB_HINT_DATA))
4432 		return;
4433 
4434 	/* sometime caller may want exact blocks */
4435 	if (unlikely(ac->ac_flags & EXT4_MB_HINT_GOAL_ONLY))
4436 		return;
4437 
4438 	/* caller may indicate that preallocation isn't
4439 	 * required (it's a tail, for example) */
4440 	if (ac->ac_flags & EXT4_MB_HINT_NOPREALLOC)
4441 		return;
4442 
4443 	if (ac->ac_flags & EXT4_MB_HINT_GROUP_ALLOC) {
4444 		ext4_mb_normalize_group_request(ac);
4445 		return ;
4446 	}
4447 
4448 	bsbits = ac->ac_sb->s_blocksize_bits;
4449 
4450 	/* first, let's learn actual file size
4451 	 * given current request is allocated */
4452 	size = extent_logical_end(sbi, &ac->ac_o_ex);
4453 	size = size << bsbits;
4454 	if (size < i_size_read(ac->ac_inode))
4455 		size = i_size_read(ac->ac_inode);
4456 	orig_size = size;
4457 
4458 	/* max size of free chunks */
4459 	max = 2 << bsbits;
4460 
4461 #define NRL_CHECK_SIZE(req, size, max, chunk_size)	\
4462 		(req <= (size) || max <= (chunk_size))
4463 
4464 	/* first, try to predict filesize */
4465 	/* XXX: should this table be tunable? */
4466 	start_off = 0;
4467 	if (size <= 16 * 1024) {
4468 		size = 16 * 1024;
4469 	} else if (size <= 32 * 1024) {
4470 		size = 32 * 1024;
4471 	} else if (size <= 64 * 1024) {
4472 		size = 64 * 1024;
4473 	} else if (size <= 128 * 1024) {
4474 		size = 128 * 1024;
4475 	} else if (size <= 256 * 1024) {
4476 		size = 256 * 1024;
4477 	} else if (size <= 512 * 1024) {
4478 		size = 512 * 1024;
4479 	} else if (size <= 1024 * 1024) {
4480 		size = 1024 * 1024;
4481 	} else if (NRL_CHECK_SIZE(size, 4 * 1024 * 1024, max, 2 * 1024)) {
4482 		start_off = ((loff_t)ac->ac_o_ex.fe_logical >>
4483 						(21 - bsbits)) << 21;
4484 		size = 2 * 1024 * 1024;
4485 	} else if (NRL_CHECK_SIZE(size, 8 * 1024 * 1024, max, 4 * 1024)) {
4486 		start_off = ((loff_t)ac->ac_o_ex.fe_logical >>
4487 							(22 - bsbits)) << 22;
4488 		size = 4 * 1024 * 1024;
4489 	} else if (NRL_CHECK_SIZE(EXT4_C2B(sbi, ac->ac_o_ex.fe_len),
4490 					(8<<20)>>bsbits, max, 8 * 1024)) {
4491 		start_off = ((loff_t)ac->ac_o_ex.fe_logical >>
4492 							(23 - bsbits)) << 23;
4493 		size = 8 * 1024 * 1024;
4494 	} else {
4495 		start_off = (loff_t) ac->ac_o_ex.fe_logical << bsbits;
4496 		size	  = (loff_t) EXT4_C2B(sbi,
4497 					      ac->ac_o_ex.fe_len) << bsbits;
4498 	}
4499 	size = size >> bsbits;
4500 	start = start_off >> bsbits;
4501 
4502 	/*
4503 	 * For tiny groups (smaller than 8MB) the chosen allocation
4504 	 * alignment may be larger than group size. Make sure the
4505 	 * alignment does not move allocation to a different group which
4506 	 * makes mballoc fail assertions later.
4507 	 */
4508 	start = max(start, rounddown(ac->ac_o_ex.fe_logical,
4509 			(ext4_lblk_t)EXT4_BLOCKS_PER_GROUP(ac->ac_sb)));
4510 
4511 	/* avoid unnecessary preallocation that may trigger assertions */
4512 	if (start + size > EXT_MAX_BLOCKS)
4513 		size = EXT_MAX_BLOCKS - start;
4514 
4515 	/* don't cover already allocated blocks in selected range */
4516 	if (ar->pleft && start <= ar->lleft) {
4517 		size -= ar->lleft + 1 - start;
4518 		start = ar->lleft + 1;
4519 	}
4520 	if (ar->pright && start + size - 1 >= ar->lright)
4521 		size -= start + size - ar->lright;
4522 
4523 	/*
4524 	 * Trim allocation request for filesystems with artificially small
4525 	 * groups.
4526 	 */
4527 	if (size > EXT4_BLOCKS_PER_GROUP(ac->ac_sb))
4528 		size = EXT4_BLOCKS_PER_GROUP(ac->ac_sb);
4529 
4530 	end = start + size;
4531 
4532 	ext4_mb_pa_adjust_overlap(ac, &start, &end);
4533 
4534 	size = end - start;
4535 
4536 	/*
4537 	 * In this function "start" and "size" are normalized for better
4538 	 * alignment and length such that we could preallocate more blocks.
4539 	 * This normalization is done such that original request of
4540 	 * ac->ac_o_ex.fe_logical & fe_len should always lie within "start" and
4541 	 * "size" boundaries.
4542 	 * (Note fe_len can be relaxed since FS block allocation API does not
4543 	 * provide gurantee on number of contiguous blocks allocation since that
4544 	 * depends upon free space left, etc).
4545 	 * In case of inode pa, later we use the allocated blocks
4546 	 * [pa_pstart + fe_logical - pa_lstart, fe_len/size] from the preallocated
4547 	 * range of goal/best blocks [start, size] to put it at the
4548 	 * ac_o_ex.fe_logical extent of this inode.
4549 	 * (See ext4_mb_use_inode_pa() for more details)
4550 	 */
4551 	if (start + size <= ac->ac_o_ex.fe_logical ||
4552 			start > ac->ac_o_ex.fe_logical) {
4553 		ext4_msg(ac->ac_sb, KERN_ERR,
4554 			 "start %lu, size %lu, fe_logical %lu",
4555 			 (unsigned long) start, (unsigned long) size,
4556 			 (unsigned long) ac->ac_o_ex.fe_logical);
4557 		BUG();
4558 	}
4559 	BUG_ON(size <= 0 || size > EXT4_BLOCKS_PER_GROUP(ac->ac_sb));
4560 
4561 	/* now prepare goal request */
4562 
4563 	/* XXX: is it better to align blocks WRT to logical
4564 	 * placement or satisfy big request as is */
4565 	ac->ac_g_ex.fe_logical = start;
4566 	ac->ac_g_ex.fe_len = EXT4_NUM_B2C(sbi, size);
4567 	ac->ac_orig_goal_len = ac->ac_g_ex.fe_len;
4568 
4569 	/* define goal start in order to merge */
4570 	if (ar->pright && (ar->lright == (start + size)) &&
4571 	    ar->pright >= size &&
4572 	    ar->pright - size >= le32_to_cpu(es->s_first_data_block)) {
4573 		/* merge to the right */
4574 		ext4_get_group_no_and_offset(ac->ac_sb, ar->pright - size,
4575 						&ac->ac_g_ex.fe_group,
4576 						&ac->ac_g_ex.fe_start);
4577 		ac->ac_flags |= EXT4_MB_HINT_TRY_GOAL;
4578 	}
4579 	if (ar->pleft && (ar->lleft + 1 == start) &&
4580 	    ar->pleft + 1 < ext4_blocks_count(es)) {
4581 		/* merge to the left */
4582 		ext4_get_group_no_and_offset(ac->ac_sb, ar->pleft + 1,
4583 						&ac->ac_g_ex.fe_group,
4584 						&ac->ac_g_ex.fe_start);
4585 		ac->ac_flags |= EXT4_MB_HINT_TRY_GOAL;
4586 	}
4587 
4588 	mb_debug(ac->ac_sb, "goal: %lld(was %lld) blocks at %u\n", size,
4589 		 orig_size, start);
4590 }
4591 
ext4_mb_collect_stats(struct ext4_allocation_context * ac)4592 static void ext4_mb_collect_stats(struct ext4_allocation_context *ac)
4593 {
4594 	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
4595 
4596 	if (sbi->s_mb_stats && ac->ac_g_ex.fe_len >= 1) {
4597 		atomic_inc(&sbi->s_bal_reqs);
4598 		atomic_add(ac->ac_b_ex.fe_len, &sbi->s_bal_allocated);
4599 		if (ac->ac_b_ex.fe_len >= ac->ac_o_ex.fe_len)
4600 			atomic_inc(&sbi->s_bal_success);
4601 
4602 		atomic_add(ac->ac_found, &sbi->s_bal_ex_scanned);
4603 		for (int i=0; i<EXT4_MB_NUM_CRS; i++) {
4604 			atomic_add(ac->ac_cX_found[i], &sbi->s_bal_cX_ex_scanned[i]);
4605 		}
4606 
4607 		atomic_add(ac->ac_groups_scanned, &sbi->s_bal_groups_scanned);
4608 		if (ac->ac_g_ex.fe_start == ac->ac_b_ex.fe_start &&
4609 				ac->ac_g_ex.fe_group == ac->ac_b_ex.fe_group)
4610 			atomic_inc(&sbi->s_bal_goals);
4611 		/* did we allocate as much as normalizer originally wanted? */
4612 		if (ac->ac_f_ex.fe_len == ac->ac_orig_goal_len)
4613 			atomic_inc(&sbi->s_bal_len_goals);
4614 
4615 		if (ac->ac_found > sbi->s_mb_max_to_scan)
4616 			atomic_inc(&sbi->s_bal_breaks);
4617 	}
4618 
4619 	if (ac->ac_op == EXT4_MB_HISTORY_ALLOC)
4620 		trace_ext4_mballoc_alloc(ac);
4621 	else
4622 		trace_ext4_mballoc_prealloc(ac);
4623 }
4624 
4625 /*
4626  * Called on failure; free up any blocks from the inode PA for this
4627  * context.  We don't need this for MB_GROUP_PA because we only change
4628  * pa_free in ext4_mb_release_context(), but on failure, we've already
4629  * zeroed out ac->ac_b_ex.fe_len, so group_pa->pa_free is not changed.
4630  */
ext4_discard_allocated_blocks(struct ext4_allocation_context * ac)4631 static void ext4_discard_allocated_blocks(struct ext4_allocation_context *ac)
4632 {
4633 	struct ext4_prealloc_space *pa = ac->ac_pa;
4634 	struct ext4_buddy e4b;
4635 	int err;
4636 
4637 	if (pa == NULL) {
4638 		if (ac->ac_f_ex.fe_len == 0)
4639 			return;
4640 		err = ext4_mb_load_buddy(ac->ac_sb, ac->ac_f_ex.fe_group, &e4b);
4641 		if (WARN_RATELIMIT(err,
4642 				   "ext4: mb_load_buddy failed (%d)", err))
4643 			/*
4644 			 * This should never happen since we pin the
4645 			 * pages in the ext4_allocation_context so
4646 			 * ext4_mb_load_buddy() should never fail.
4647 			 */
4648 			return;
4649 		ext4_lock_group(ac->ac_sb, ac->ac_f_ex.fe_group);
4650 		mb_free_blocks(ac->ac_inode, &e4b, ac->ac_f_ex.fe_start,
4651 			       ac->ac_f_ex.fe_len);
4652 		ext4_unlock_group(ac->ac_sb, ac->ac_f_ex.fe_group);
4653 		ext4_mb_unload_buddy(&e4b);
4654 		return;
4655 	}
4656 	if (pa->pa_type == MB_INODE_PA) {
4657 		spin_lock(&pa->pa_lock);
4658 		pa->pa_free += ac->ac_b_ex.fe_len;
4659 		spin_unlock(&pa->pa_lock);
4660 	}
4661 }
4662 
4663 /*
4664  * use blocks preallocated to inode
4665  */
ext4_mb_use_inode_pa(struct ext4_allocation_context * ac,struct ext4_prealloc_space * pa)4666 static void ext4_mb_use_inode_pa(struct ext4_allocation_context *ac,
4667 				struct ext4_prealloc_space *pa)
4668 {
4669 	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
4670 	ext4_fsblk_t start;
4671 	ext4_fsblk_t end;
4672 	int len;
4673 
4674 	/* found preallocated blocks, use them */
4675 	start = pa->pa_pstart + (ac->ac_o_ex.fe_logical - pa->pa_lstart);
4676 	end = min(pa->pa_pstart + EXT4_C2B(sbi, pa->pa_len),
4677 		  start + EXT4_C2B(sbi, ac->ac_o_ex.fe_len));
4678 	len = EXT4_NUM_B2C(sbi, end - start);
4679 	ext4_get_group_no_and_offset(ac->ac_sb, start, &ac->ac_b_ex.fe_group,
4680 					&ac->ac_b_ex.fe_start);
4681 	ac->ac_b_ex.fe_len = len;
4682 	ac->ac_status = AC_STATUS_FOUND;
4683 	ac->ac_pa = pa;
4684 
4685 	BUG_ON(start < pa->pa_pstart);
4686 	BUG_ON(end > pa->pa_pstart + EXT4_C2B(sbi, pa->pa_len));
4687 	BUG_ON(pa->pa_free < len);
4688 	BUG_ON(ac->ac_b_ex.fe_len <= 0);
4689 	pa->pa_free -= len;
4690 
4691 	mb_debug(ac->ac_sb, "use %llu/%d from inode pa %p\n", start, len, pa);
4692 }
4693 
4694 /*
4695  * use blocks preallocated to locality group
4696  */
ext4_mb_use_group_pa(struct ext4_allocation_context * ac,struct ext4_prealloc_space * pa)4697 static void ext4_mb_use_group_pa(struct ext4_allocation_context *ac,
4698 				struct ext4_prealloc_space *pa)
4699 {
4700 	unsigned int len = ac->ac_o_ex.fe_len;
4701 
4702 	ext4_get_group_no_and_offset(ac->ac_sb, pa->pa_pstart,
4703 					&ac->ac_b_ex.fe_group,
4704 					&ac->ac_b_ex.fe_start);
4705 	ac->ac_b_ex.fe_len = len;
4706 	ac->ac_status = AC_STATUS_FOUND;
4707 	ac->ac_pa = pa;
4708 
4709 	/* we don't correct pa_pstart or pa_len here to avoid
4710 	 * possible race when the group is being loaded concurrently
4711 	 * instead we correct pa later, after blocks are marked
4712 	 * in on-disk bitmap -- see ext4_mb_release_context()
4713 	 * Other CPUs are prevented from allocating from this pa by lg_mutex
4714 	 */
4715 	mb_debug(ac->ac_sb, "use %u/%u from group pa %p\n",
4716 		 pa->pa_lstart, len, pa);
4717 }
4718 
4719 /*
4720  * Return the prealloc space that have minimal distance
4721  * from the goal block. @cpa is the prealloc
4722  * space that is having currently known minimal distance
4723  * from the goal block.
4724  */
4725 static struct ext4_prealloc_space *
ext4_mb_check_group_pa(ext4_fsblk_t goal_block,struct ext4_prealloc_space * pa,struct ext4_prealloc_space * cpa)4726 ext4_mb_check_group_pa(ext4_fsblk_t goal_block,
4727 			struct ext4_prealloc_space *pa,
4728 			struct ext4_prealloc_space *cpa)
4729 {
4730 	ext4_fsblk_t cur_distance, new_distance;
4731 
4732 	if (cpa == NULL) {
4733 		atomic_inc(&pa->pa_count);
4734 		return pa;
4735 	}
4736 	cur_distance = abs(goal_block - cpa->pa_pstart);
4737 	new_distance = abs(goal_block - pa->pa_pstart);
4738 
4739 	if (cur_distance <= new_distance)
4740 		return cpa;
4741 
4742 	/* drop the previous reference */
4743 	atomic_dec(&cpa->pa_count);
4744 	atomic_inc(&pa->pa_count);
4745 	return pa;
4746 }
4747 
4748 /*
4749  * check if found pa meets EXT4_MB_HINT_GOAL_ONLY
4750  */
4751 static bool
ext4_mb_pa_goal_check(struct ext4_allocation_context * ac,struct ext4_prealloc_space * pa)4752 ext4_mb_pa_goal_check(struct ext4_allocation_context *ac,
4753 		      struct ext4_prealloc_space *pa)
4754 {
4755 	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
4756 	ext4_fsblk_t start;
4757 
4758 	if (likely(!(ac->ac_flags & EXT4_MB_HINT_GOAL_ONLY)))
4759 		return true;
4760 
4761 	/*
4762 	 * If EXT4_MB_HINT_GOAL_ONLY is set, ac_g_ex will not be adjusted
4763 	 * in ext4_mb_normalize_request and will keep same with ac_o_ex
4764 	 * from ext4_mb_initialize_context. Choose ac_g_ex here to keep
4765 	 * consistent with ext4_mb_find_by_goal.
4766 	 */
4767 	start = pa->pa_pstart +
4768 		(ac->ac_g_ex.fe_logical - pa->pa_lstart);
4769 	if (ext4_grp_offs_to_block(ac->ac_sb, &ac->ac_g_ex) != start)
4770 		return false;
4771 
4772 	if (ac->ac_g_ex.fe_len > pa->pa_len -
4773 	    EXT4_B2C(sbi, ac->ac_g_ex.fe_logical - pa->pa_lstart))
4774 		return false;
4775 
4776 	return true;
4777 }
4778 
4779 /*
4780  * search goal blocks in preallocated space
4781  */
4782 static noinline_for_stack bool
ext4_mb_use_preallocated(struct ext4_allocation_context * ac)4783 ext4_mb_use_preallocated(struct ext4_allocation_context *ac)
4784 {
4785 	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
4786 	int order, i;
4787 	struct ext4_inode_info *ei = EXT4_I(ac->ac_inode);
4788 	struct ext4_locality_group *lg;
4789 	struct ext4_prealloc_space *tmp_pa = NULL, *cpa = NULL;
4790 	struct rb_node *iter;
4791 	ext4_fsblk_t goal_block;
4792 
4793 	/* only data can be preallocated */
4794 	if (!(ac->ac_flags & EXT4_MB_HINT_DATA))
4795 		return false;
4796 
4797 	/*
4798 	 * first, try per-file preallocation by searching the inode pa rbtree.
4799 	 *
4800 	 * Here, we can't do a direct traversal of the tree because
4801 	 * ext4_mb_discard_group_preallocation() can paralelly mark the pa
4802 	 * deleted and that can cause direct traversal to skip some entries.
4803 	 */
4804 	read_lock(&ei->i_prealloc_lock);
4805 
4806 	if (RB_EMPTY_ROOT(&ei->i_prealloc_node)) {
4807 		goto try_group_pa;
4808 	}
4809 
4810 	/*
4811 	 * Step 1: Find a pa with logical start immediately adjacent to the
4812 	 * original logical start. This could be on the left or right.
4813 	 *
4814 	 * (tmp_pa->pa_lstart never changes so we can skip locking for it).
4815 	 */
4816 	for (iter = ei->i_prealloc_node.rb_node; iter;
4817 	     iter = ext4_mb_pa_rb_next_iter(ac->ac_o_ex.fe_logical,
4818 					    tmp_pa->pa_lstart, iter)) {
4819 		tmp_pa = rb_entry(iter, struct ext4_prealloc_space,
4820 				  pa_node.inode_node);
4821 	}
4822 
4823 	/*
4824 	 * Step 2: The adjacent pa might be to the right of logical start, find
4825 	 * the left adjacent pa. After this step we'd have a valid tmp_pa whose
4826 	 * logical start is towards the left of original request's logical start
4827 	 */
4828 	if (tmp_pa->pa_lstart > ac->ac_o_ex.fe_logical) {
4829 		struct rb_node *tmp;
4830 		tmp = rb_prev(&tmp_pa->pa_node.inode_node);
4831 
4832 		if (tmp) {
4833 			tmp_pa = rb_entry(tmp, struct ext4_prealloc_space,
4834 					    pa_node.inode_node);
4835 		} else {
4836 			/*
4837 			 * If there is no adjacent pa to the left then finding
4838 			 * an overlapping pa is not possible hence stop searching
4839 			 * inode pa tree
4840 			 */
4841 			goto try_group_pa;
4842 		}
4843 	}
4844 
4845 	BUG_ON(!(tmp_pa && tmp_pa->pa_lstart <= ac->ac_o_ex.fe_logical));
4846 
4847 	/*
4848 	 * Step 3: If the left adjacent pa is deleted, keep moving left to find
4849 	 * the first non deleted adjacent pa. After this step we should have a
4850 	 * valid tmp_pa which is guaranteed to be non deleted.
4851 	 */
4852 	for (iter = &tmp_pa->pa_node.inode_node;; iter = rb_prev(iter)) {
4853 		if (!iter) {
4854 			/*
4855 			 * no non deleted left adjacent pa, so stop searching
4856 			 * inode pa tree
4857 			 */
4858 			goto try_group_pa;
4859 		}
4860 		tmp_pa = rb_entry(iter, struct ext4_prealloc_space,
4861 				  pa_node.inode_node);
4862 		spin_lock(&tmp_pa->pa_lock);
4863 		if (tmp_pa->pa_deleted == 0) {
4864 			/*
4865 			 * We will keep holding the pa_lock from
4866 			 * this point on because we don't want group discard
4867 			 * to delete this pa underneath us. Since group
4868 			 * discard is anyways an ENOSPC operation it
4869 			 * should be okay for it to wait a few more cycles.
4870 			 */
4871 			break;
4872 		} else {
4873 			spin_unlock(&tmp_pa->pa_lock);
4874 		}
4875 	}
4876 
4877 	BUG_ON(!(tmp_pa && tmp_pa->pa_lstart <= ac->ac_o_ex.fe_logical));
4878 	BUG_ON(tmp_pa->pa_deleted == 1);
4879 
4880 	/*
4881 	 * Step 4: We now have the non deleted left adjacent pa. Only this
4882 	 * pa can possibly satisfy the request hence check if it overlaps
4883 	 * original logical start and stop searching if it doesn't.
4884 	 */
4885 	if (ac->ac_o_ex.fe_logical >= pa_logical_end(sbi, tmp_pa)) {
4886 		spin_unlock(&tmp_pa->pa_lock);
4887 		goto try_group_pa;
4888 	}
4889 
4890 	/* non-extent files can't have physical blocks past 2^32 */
4891 	if (!(ext4_test_inode_flag(ac->ac_inode, EXT4_INODE_EXTENTS)) &&
4892 	    (tmp_pa->pa_pstart + EXT4_C2B(sbi, tmp_pa->pa_len) >
4893 	     EXT4_MAX_BLOCK_FILE_PHYS)) {
4894 		/*
4895 		 * Since PAs don't overlap, we won't find any other PA to
4896 		 * satisfy this.
4897 		 */
4898 		spin_unlock(&tmp_pa->pa_lock);
4899 		goto try_group_pa;
4900 	}
4901 
4902 	if (tmp_pa->pa_free && likely(ext4_mb_pa_goal_check(ac, tmp_pa))) {
4903 		atomic_inc(&tmp_pa->pa_count);
4904 		ext4_mb_use_inode_pa(ac, tmp_pa);
4905 		spin_unlock(&tmp_pa->pa_lock);
4906 		read_unlock(&ei->i_prealloc_lock);
4907 		return true;
4908 	} else {
4909 		/*
4910 		 * We found a valid overlapping pa but couldn't use it because
4911 		 * it had no free blocks. This should ideally never happen
4912 		 * because:
4913 		 *
4914 		 * 1. When a new inode pa is added to rbtree it must have
4915 		 *    pa_free > 0 since otherwise we won't actually need
4916 		 *    preallocation.
4917 		 *
4918 		 * 2. An inode pa that is in the rbtree can only have it's
4919 		 *    pa_free become zero when another thread calls:
4920 		 *      ext4_mb_new_blocks
4921 		 *       ext4_mb_use_preallocated
4922 		 *        ext4_mb_use_inode_pa
4923 		 *
4924 		 * 3. Further, after the above calls make pa_free == 0, we will
4925 		 *    immediately remove it from the rbtree in:
4926 		 *      ext4_mb_new_blocks
4927 		 *       ext4_mb_release_context
4928 		 *        ext4_mb_put_pa
4929 		 *
4930 		 * 4. Since the pa_free becoming 0 and pa_free getting removed
4931 		 * from tree both happen in ext4_mb_new_blocks, which is always
4932 		 * called with i_data_sem held for data allocations, we can be
4933 		 * sure that another process will never see a pa in rbtree with
4934 		 * pa_free == 0.
4935 		 */
4936 		WARN_ON_ONCE(tmp_pa->pa_free == 0);
4937 	}
4938 	spin_unlock(&tmp_pa->pa_lock);
4939 try_group_pa:
4940 	read_unlock(&ei->i_prealloc_lock);
4941 
4942 	/* can we use group allocation? */
4943 	if (!(ac->ac_flags & EXT4_MB_HINT_GROUP_ALLOC))
4944 		return false;
4945 
4946 	/* inode may have no locality group for some reason */
4947 	lg = ac->ac_lg;
4948 	if (lg == NULL)
4949 		return false;
4950 	order  = fls(ac->ac_o_ex.fe_len) - 1;
4951 	if (order > PREALLOC_TB_SIZE - 1)
4952 		/* The max size of hash table is PREALLOC_TB_SIZE */
4953 		order = PREALLOC_TB_SIZE - 1;
4954 
4955 	goal_block = ext4_grp_offs_to_block(ac->ac_sb, &ac->ac_g_ex);
4956 	/*
4957 	 * search for the prealloc space that is having
4958 	 * minimal distance from the goal block.
4959 	 */
4960 	for (i = order; i < PREALLOC_TB_SIZE; i++) {
4961 		rcu_read_lock();
4962 		list_for_each_entry_rcu(tmp_pa, &lg->lg_prealloc_list[i],
4963 					pa_node.lg_list) {
4964 			spin_lock(&tmp_pa->pa_lock);
4965 			if (tmp_pa->pa_deleted == 0 &&
4966 					tmp_pa->pa_free >= ac->ac_o_ex.fe_len) {
4967 
4968 				cpa = ext4_mb_check_group_pa(goal_block,
4969 								tmp_pa, cpa);
4970 			}
4971 			spin_unlock(&tmp_pa->pa_lock);
4972 		}
4973 		rcu_read_unlock();
4974 	}
4975 	if (cpa) {
4976 		ext4_mb_use_group_pa(ac, cpa);
4977 		return true;
4978 	}
4979 	return false;
4980 }
4981 
4982 /*
4983  * the function goes through all preallocation in this group and marks them
4984  * used in in-core bitmap. buddy must be generated from this bitmap
4985  * Need to be called with ext4 group lock held
4986  */
4987 static noinline_for_stack
ext4_mb_generate_from_pa(struct super_block * sb,void * bitmap,ext4_group_t group)4988 void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
4989 					ext4_group_t group)
4990 {
4991 	struct ext4_group_info *grp = ext4_get_group_info(sb, group);
4992 	struct ext4_prealloc_space *pa;
4993 	struct list_head *cur;
4994 	ext4_group_t groupnr;
4995 	ext4_grpblk_t start;
4996 	int preallocated = 0;
4997 	int len;
4998 
4999 	if (!grp)
5000 		return;
5001 
5002 	/* all form of preallocation discards first load group,
5003 	 * so the only competing code is preallocation use.
5004 	 * we don't need any locking here
5005 	 * notice we do NOT ignore preallocations with pa_deleted
5006 	 * otherwise we could leave used blocks available for
5007 	 * allocation in buddy when concurrent ext4_mb_put_pa()
5008 	 * is dropping preallocation
5009 	 */
5010 	list_for_each(cur, &grp->bb_prealloc_list) {
5011 		pa = list_entry(cur, struct ext4_prealloc_space, pa_group_list);
5012 		spin_lock(&pa->pa_lock);
5013 		ext4_get_group_no_and_offset(sb, pa->pa_pstart,
5014 					     &groupnr, &start);
5015 		len = pa->pa_len;
5016 		spin_unlock(&pa->pa_lock);
5017 		if (unlikely(len == 0))
5018 			continue;
5019 		BUG_ON(groupnr != group);
5020 		mb_set_bits(bitmap, start, len);
5021 		preallocated += len;
5022 	}
5023 	mb_debug(sb, "preallocated %d for group %u\n", preallocated, group);
5024 }
5025 
ext4_mb_mark_pa_deleted(struct super_block * sb,struct ext4_prealloc_space * pa)5026 static void ext4_mb_mark_pa_deleted(struct super_block *sb,
5027 				    struct ext4_prealloc_space *pa)
5028 {
5029 	struct ext4_inode_info *ei;
5030 
5031 	if (pa->pa_deleted) {
5032 		ext4_warning(sb, "deleted pa, type:%d, pblk:%llu, lblk:%u, len:%d\n",
5033 			     pa->pa_type, pa->pa_pstart, pa->pa_lstart,
5034 			     pa->pa_len);
5035 		return;
5036 	}
5037 
5038 	pa->pa_deleted = 1;
5039 
5040 	if (pa->pa_type == MB_INODE_PA) {
5041 		ei = EXT4_I(pa->pa_inode);
5042 		atomic_dec(&ei->i_prealloc_active);
5043 	}
5044 }
5045 
ext4_mb_pa_free(struct ext4_prealloc_space * pa)5046 static inline void ext4_mb_pa_free(struct ext4_prealloc_space *pa)
5047 {
5048 	BUG_ON(!pa);
5049 	BUG_ON(atomic_read(&pa->pa_count));
5050 	BUG_ON(pa->pa_deleted == 0);
5051 	kmem_cache_free(ext4_pspace_cachep, pa);
5052 }
5053 
ext4_mb_pa_callback(struct rcu_head * head)5054 static void ext4_mb_pa_callback(struct rcu_head *head)
5055 {
5056 	struct ext4_prealloc_space *pa;
5057 
5058 	pa = container_of(head, struct ext4_prealloc_space, u.pa_rcu);
5059 	ext4_mb_pa_free(pa);
5060 }
5061 
5062 /*
5063  * drops a reference to preallocated space descriptor
5064  * if this was the last reference and the space is consumed
5065  */
ext4_mb_put_pa(struct ext4_allocation_context * ac,struct super_block * sb,struct ext4_prealloc_space * pa)5066 static void ext4_mb_put_pa(struct ext4_allocation_context *ac,
5067 			struct super_block *sb, struct ext4_prealloc_space *pa)
5068 {
5069 	ext4_group_t grp;
5070 	ext4_fsblk_t grp_blk;
5071 	struct ext4_inode_info *ei = EXT4_I(ac->ac_inode);
5072 
5073 	/* in this short window concurrent discard can set pa_deleted */
5074 	spin_lock(&pa->pa_lock);
5075 	if (!atomic_dec_and_test(&pa->pa_count) || pa->pa_free != 0) {
5076 		spin_unlock(&pa->pa_lock);
5077 		return;
5078 	}
5079 
5080 	if (pa->pa_deleted == 1) {
5081 		spin_unlock(&pa->pa_lock);
5082 		return;
5083 	}
5084 
5085 	ext4_mb_mark_pa_deleted(sb, pa);
5086 	spin_unlock(&pa->pa_lock);
5087 
5088 	grp_blk = pa->pa_pstart;
5089 	/*
5090 	 * If doing group-based preallocation, pa_pstart may be in the
5091 	 * next group when pa is used up
5092 	 */
5093 	if (pa->pa_type == MB_GROUP_PA)
5094 		grp_blk--;
5095 
5096 	grp = ext4_get_group_number(sb, grp_blk);
5097 
5098 	/*
5099 	 * possible race:
5100 	 *
5101 	 *  P1 (buddy init)			P2 (regular allocation)
5102 	 *					find block B in PA
5103 	 *  copy on-disk bitmap to buddy
5104 	 *  					mark B in on-disk bitmap
5105 	 *					drop PA from group
5106 	 *  mark all PAs in buddy
5107 	 *
5108 	 * thus, P1 initializes buddy with B available. to prevent this
5109 	 * we make "copy" and "mark all PAs" atomic and serialize "drop PA"
5110 	 * against that pair
5111 	 */
5112 	ext4_lock_group(sb, grp);
5113 	list_del(&pa->pa_group_list);
5114 	ext4_unlock_group(sb, grp);
5115 
5116 	if (pa->pa_type == MB_INODE_PA) {
5117 		write_lock(pa->pa_node_lock.inode_lock);
5118 		rb_erase(&pa->pa_node.inode_node, &ei->i_prealloc_node);
5119 		write_unlock(pa->pa_node_lock.inode_lock);
5120 		ext4_mb_pa_free(pa);
5121 	} else {
5122 		spin_lock(pa->pa_node_lock.lg_lock);
5123 		list_del_rcu(&pa->pa_node.lg_list);
5124 		spin_unlock(pa->pa_node_lock.lg_lock);
5125 		call_rcu(&(pa)->u.pa_rcu, ext4_mb_pa_callback);
5126 	}
5127 }
5128 
ext4_mb_pa_rb_insert(struct rb_root * root,struct rb_node * new)5129 static void ext4_mb_pa_rb_insert(struct rb_root *root, struct rb_node *new)
5130 {
5131 	struct rb_node **iter = &root->rb_node, *parent = NULL;
5132 	struct ext4_prealloc_space *iter_pa, *new_pa;
5133 	ext4_lblk_t iter_start, new_start;
5134 
5135 	while (*iter) {
5136 		iter_pa = rb_entry(*iter, struct ext4_prealloc_space,
5137 				   pa_node.inode_node);
5138 		new_pa = rb_entry(new, struct ext4_prealloc_space,
5139 				   pa_node.inode_node);
5140 		iter_start = iter_pa->pa_lstart;
5141 		new_start = new_pa->pa_lstart;
5142 
5143 		parent = *iter;
5144 		if (new_start < iter_start)
5145 			iter = &((*iter)->rb_left);
5146 		else
5147 			iter = &((*iter)->rb_right);
5148 	}
5149 
5150 	rb_link_node(new, parent, iter);
5151 	rb_insert_color(new, root);
5152 }
5153 
5154 /*
5155  * creates new preallocated space for given inode
5156  */
5157 static noinline_for_stack void
ext4_mb_new_inode_pa(struct ext4_allocation_context * ac)5158 ext4_mb_new_inode_pa(struct ext4_allocation_context *ac)
5159 {
5160 	struct super_block *sb = ac->ac_sb;
5161 	struct ext4_sb_info *sbi = EXT4_SB(sb);
5162 	struct ext4_prealloc_space *pa;
5163 	struct ext4_group_info *grp;
5164 	struct ext4_inode_info *ei;
5165 
5166 	/* preallocate only when found space is larger then requested */
5167 	BUG_ON(ac->ac_o_ex.fe_len >= ac->ac_b_ex.fe_len);
5168 	BUG_ON(ac->ac_status != AC_STATUS_FOUND);
5169 	BUG_ON(!S_ISREG(ac->ac_inode->i_mode));
5170 	BUG_ON(ac->ac_pa == NULL);
5171 
5172 	pa = ac->ac_pa;
5173 
5174 	if (ac->ac_b_ex.fe_len < ac->ac_orig_goal_len) {
5175 		struct ext4_free_extent ex = {
5176 			.fe_logical = ac->ac_g_ex.fe_logical,
5177 			.fe_len = ac->ac_orig_goal_len,
5178 		};
5179 		loff_t orig_goal_end = extent_logical_end(sbi, &ex);
5180 		loff_t o_ex_end = extent_logical_end(sbi, &ac->ac_o_ex);
5181 
5182 		/*
5183 		 * We can't allocate as much as normalizer wants, so we try
5184 		 * to get proper lstart to cover the original request, except
5185 		 * when the goal doesn't cover the original request as below:
5186 		 *
5187 		 * orig_ex:2045/2055(10), isize:8417280 -> normalized:0/2048
5188 		 * best_ex:0/200(200) -> adjusted: 1848/2048(200)
5189 		 */
5190 		BUG_ON(ac->ac_g_ex.fe_logical > ac->ac_o_ex.fe_logical);
5191 		BUG_ON(ac->ac_g_ex.fe_len < ac->ac_o_ex.fe_len);
5192 
5193 		/*
5194 		 * Use the below logic for adjusting best extent as it keeps
5195 		 * fragmentation in check while ensuring logical range of best
5196 		 * extent doesn't overflow out of goal extent:
5197 		 *
5198 		 * 1. Check if best ex can be kept at end of goal (before
5199 		 *    cr_best_avail trimmed it) and still cover original start
5200 		 * 2. Else, check if best ex can be kept at start of goal and
5201 		 *    still cover original end
5202 		 * 3. Else, keep the best ex at start of original request.
5203 		 */
5204 		ex.fe_len = ac->ac_b_ex.fe_len;
5205 
5206 		ex.fe_logical = orig_goal_end - EXT4_C2B(sbi, ex.fe_len);
5207 		if (ac->ac_o_ex.fe_logical >= ex.fe_logical)
5208 			goto adjust_bex;
5209 
5210 		ex.fe_logical = ac->ac_g_ex.fe_logical;
5211 		if (o_ex_end <= extent_logical_end(sbi, &ex))
5212 			goto adjust_bex;
5213 
5214 		ex.fe_logical = ac->ac_o_ex.fe_logical;
5215 adjust_bex:
5216 		ac->ac_b_ex.fe_logical = ex.fe_logical;
5217 
5218 		BUG_ON(ac->ac_o_ex.fe_logical < ac->ac_b_ex.fe_logical);
5219 		BUG_ON(extent_logical_end(sbi, &ex) > orig_goal_end);
5220 	}
5221 
5222 	pa->pa_lstart = ac->ac_b_ex.fe_logical;
5223 	pa->pa_pstart = ext4_grp_offs_to_block(sb, &ac->ac_b_ex);
5224 	pa->pa_len = ac->ac_b_ex.fe_len;
5225 	pa->pa_free = pa->pa_len;
5226 	spin_lock_init(&pa->pa_lock);
5227 	INIT_LIST_HEAD(&pa->pa_group_list);
5228 	pa->pa_deleted = 0;
5229 	pa->pa_type = MB_INODE_PA;
5230 
5231 	mb_debug(sb, "new inode pa %p: %llu/%d for %u\n", pa, pa->pa_pstart,
5232 		 pa->pa_len, pa->pa_lstart);
5233 	trace_ext4_mb_new_inode_pa(ac, pa);
5234 
5235 	atomic_add(pa->pa_free, &sbi->s_mb_preallocated);
5236 	ext4_mb_use_inode_pa(ac, pa);
5237 
5238 	ei = EXT4_I(ac->ac_inode);
5239 	grp = ext4_get_group_info(sb, ac->ac_b_ex.fe_group);
5240 	if (!grp)
5241 		return;
5242 
5243 	pa->pa_node_lock.inode_lock = &ei->i_prealloc_lock;
5244 	pa->pa_inode = ac->ac_inode;
5245 
5246 	list_add(&pa->pa_group_list, &grp->bb_prealloc_list);
5247 
5248 	write_lock(pa->pa_node_lock.inode_lock);
5249 	ext4_mb_pa_rb_insert(&ei->i_prealloc_node, &pa->pa_node.inode_node);
5250 	write_unlock(pa->pa_node_lock.inode_lock);
5251 	atomic_inc(&ei->i_prealloc_active);
5252 }
5253 
5254 /*
5255  * creates new preallocated space for locality group inodes belongs to
5256  */
5257 static noinline_for_stack void
ext4_mb_new_group_pa(struct ext4_allocation_context * ac)5258 ext4_mb_new_group_pa(struct ext4_allocation_context *ac)
5259 {
5260 	struct super_block *sb = ac->ac_sb;
5261 	struct ext4_locality_group *lg;
5262 	struct ext4_prealloc_space *pa;
5263 	struct ext4_group_info *grp;
5264 
5265 	/* preallocate only when found space is larger then requested */
5266 	BUG_ON(ac->ac_o_ex.fe_len >= ac->ac_b_ex.fe_len);
5267 	BUG_ON(ac->ac_status != AC_STATUS_FOUND);
5268 	BUG_ON(!S_ISREG(ac->ac_inode->i_mode));
5269 	BUG_ON(ac->ac_pa == NULL);
5270 
5271 	pa = ac->ac_pa;
5272 
5273 	pa->pa_pstart = ext4_grp_offs_to_block(sb, &ac->ac_b_ex);
5274 	pa->pa_lstart = pa->pa_pstart;
5275 	pa->pa_len = ac->ac_b_ex.fe_len;
5276 	pa->pa_free = pa->pa_len;
5277 	spin_lock_init(&pa->pa_lock);
5278 	INIT_LIST_HEAD(&pa->pa_node.lg_list);
5279 	INIT_LIST_HEAD(&pa->pa_group_list);
5280 	pa->pa_deleted = 0;
5281 	pa->pa_type = MB_GROUP_PA;
5282 
5283 	mb_debug(sb, "new group pa %p: %llu/%d for %u\n", pa, pa->pa_pstart,
5284 		 pa->pa_len, pa->pa_lstart);
5285 	trace_ext4_mb_new_group_pa(ac, pa);
5286 
5287 	ext4_mb_use_group_pa(ac, pa);
5288 	atomic_add(pa->pa_free, &EXT4_SB(sb)->s_mb_preallocated);
5289 
5290 	grp = ext4_get_group_info(sb, ac->ac_b_ex.fe_group);
5291 	if (!grp)
5292 		return;
5293 	lg = ac->ac_lg;
5294 	BUG_ON(lg == NULL);
5295 
5296 	pa->pa_node_lock.lg_lock = &lg->lg_prealloc_lock;
5297 	pa->pa_inode = NULL;
5298 
5299 	list_add(&pa->pa_group_list, &grp->bb_prealloc_list);
5300 
5301 	/*
5302 	 * We will later add the new pa to the right bucket
5303 	 * after updating the pa_free in ext4_mb_release_context
5304 	 */
5305 }
5306 
ext4_mb_new_preallocation(struct ext4_allocation_context * ac)5307 static void ext4_mb_new_preallocation(struct ext4_allocation_context *ac)
5308 {
5309 	if (ac->ac_flags & EXT4_MB_HINT_GROUP_ALLOC)
5310 		ext4_mb_new_group_pa(ac);
5311 	else
5312 		ext4_mb_new_inode_pa(ac);
5313 }
5314 
5315 /*
5316  * finds all unused blocks in on-disk bitmap, frees them in
5317  * in-core bitmap and buddy.
5318  * @pa must be unlinked from inode and group lists, so that
5319  * nobody else can find/use it.
5320  * the caller MUST hold group/inode locks.
5321  * TODO: optimize the case when there are no in-core structures yet
5322  */
5323 static noinline_for_stack void
ext4_mb_release_inode_pa(struct ext4_buddy * e4b,struct buffer_head * bitmap_bh,struct ext4_prealloc_space * pa)5324 ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh,
5325 			struct ext4_prealloc_space *pa)
5326 {
5327 	struct super_block *sb = e4b->bd_sb;
5328 	struct ext4_sb_info *sbi = EXT4_SB(sb);
5329 	unsigned int end;
5330 	unsigned int next;
5331 	ext4_group_t group;
5332 	ext4_grpblk_t bit;
5333 	unsigned long long grp_blk_start;
5334 	int free = 0;
5335 
5336 	BUG_ON(pa->pa_deleted == 0);
5337 	ext4_get_group_no_and_offset(sb, pa->pa_pstart, &group, &bit);
5338 	grp_blk_start = pa->pa_pstart - EXT4_C2B(sbi, bit);
5339 	BUG_ON(group != e4b->bd_group && pa->pa_len != 0);
5340 	end = bit + pa->pa_len;
5341 
5342 	while (bit < end) {
5343 		bit = mb_find_next_zero_bit(bitmap_bh->b_data, end, bit);
5344 		if (bit >= end)
5345 			break;
5346 		next = mb_find_next_bit(bitmap_bh->b_data, end, bit);
5347 		mb_debug(sb, "free preallocated %u/%u in group %u\n",
5348 			 (unsigned) ext4_group_first_block_no(sb, group) + bit,
5349 			 (unsigned) next - bit, (unsigned) group);
5350 		free += next - bit;
5351 
5352 		trace_ext4_mballoc_discard(sb, NULL, group, bit, next - bit);
5353 		trace_ext4_mb_release_inode_pa(pa, (grp_blk_start +
5354 						    EXT4_C2B(sbi, bit)),
5355 					       next - bit);
5356 		mb_free_blocks(pa->pa_inode, e4b, bit, next - bit);
5357 		bit = next + 1;
5358 	}
5359 	if (free != pa->pa_free) {
5360 		ext4_msg(e4b->bd_sb, KERN_CRIT,
5361 			 "pa %p: logic %lu, phys. %lu, len %d",
5362 			 pa, (unsigned long) pa->pa_lstart,
5363 			 (unsigned long) pa->pa_pstart,
5364 			 pa->pa_len);
5365 		ext4_grp_locked_error(sb, group, 0, 0, "free %u, pa_free %u",
5366 					free, pa->pa_free);
5367 		/*
5368 		 * pa is already deleted so we use the value obtained
5369 		 * from the bitmap and continue.
5370 		 */
5371 	}
5372 	atomic_add(free, &sbi->s_mb_discarded);
5373 }
5374 
5375 static noinline_for_stack void
ext4_mb_release_group_pa(struct ext4_buddy * e4b,struct ext4_prealloc_space * pa)5376 ext4_mb_release_group_pa(struct ext4_buddy *e4b,
5377 				struct ext4_prealloc_space *pa)
5378 {
5379 	struct super_block *sb = e4b->bd_sb;
5380 	ext4_group_t group;
5381 	ext4_grpblk_t bit;
5382 
5383 	trace_ext4_mb_release_group_pa(sb, pa);
5384 	BUG_ON(pa->pa_deleted == 0);
5385 	ext4_get_group_no_and_offset(sb, pa->pa_pstart, &group, &bit);
5386 	if (unlikely(group != e4b->bd_group && pa->pa_len != 0)) {
5387 		ext4_warning(sb, "bad group: expected %u, group %u, pa_start %llu",
5388 			     e4b->bd_group, group, pa->pa_pstart);
5389 		return;
5390 	}
5391 	mb_free_blocks(pa->pa_inode, e4b, bit, pa->pa_len);
5392 	atomic_add(pa->pa_len, &EXT4_SB(sb)->s_mb_discarded);
5393 	trace_ext4_mballoc_discard(sb, NULL, group, bit, pa->pa_len);
5394 }
5395 
5396 /*
5397  * releases all preallocations in given group
5398  *
5399  * first, we need to decide discard policy:
5400  * - when do we discard
5401  *   1) ENOSPC
5402  * - how many do we discard
5403  *   1) how many requested
5404  */
5405 static noinline_for_stack int
ext4_mb_discard_group_preallocations(struct super_block * sb,ext4_group_t group,int * busy)5406 ext4_mb_discard_group_preallocations(struct super_block *sb,
5407 				     ext4_group_t group, int *busy)
5408 {
5409 	struct ext4_group_info *grp = ext4_get_group_info(sb, group);
5410 	struct buffer_head *bitmap_bh = NULL;
5411 	struct ext4_prealloc_space *pa, *tmp;
5412 	LIST_HEAD(list);
5413 	struct ext4_buddy e4b;
5414 	struct ext4_inode_info *ei;
5415 	int err;
5416 	int free = 0;
5417 
5418 	if (!grp)
5419 		return 0;
5420 	mb_debug(sb, "discard preallocation for group %u\n", group);
5421 	if (list_empty(&grp->bb_prealloc_list))
5422 		goto out_dbg;
5423 
5424 	bitmap_bh = ext4_read_block_bitmap(sb, group);
5425 	if (IS_ERR(bitmap_bh)) {
5426 		err = PTR_ERR(bitmap_bh);
5427 		ext4_error_err(sb, -err,
5428 			       "Error %d reading block bitmap for %u",
5429 			       err, group);
5430 		goto out_dbg;
5431 	}
5432 
5433 	err = ext4_mb_load_buddy(sb, group, &e4b);
5434 	if (err) {
5435 		ext4_warning(sb, "Error %d loading buddy information for %u",
5436 			     err, group);
5437 		put_bh(bitmap_bh);
5438 		goto out_dbg;
5439 	}
5440 
5441 	ext4_lock_group(sb, group);
5442 	list_for_each_entry_safe(pa, tmp,
5443 				&grp->bb_prealloc_list, pa_group_list) {
5444 		spin_lock(&pa->pa_lock);
5445 		if (atomic_read(&pa->pa_count)) {
5446 			spin_unlock(&pa->pa_lock);
5447 			*busy = 1;
5448 			continue;
5449 		}
5450 		if (pa->pa_deleted) {
5451 			spin_unlock(&pa->pa_lock);
5452 			continue;
5453 		}
5454 
5455 		/* seems this one can be freed ... */
5456 		ext4_mb_mark_pa_deleted(sb, pa);
5457 
5458 		if (!free)
5459 			this_cpu_inc(discard_pa_seq);
5460 
5461 		/* we can trust pa_free ... */
5462 		free += pa->pa_free;
5463 
5464 		spin_unlock(&pa->pa_lock);
5465 
5466 		list_del(&pa->pa_group_list);
5467 		list_add(&pa->u.pa_tmp_list, &list);
5468 	}
5469 
5470 	/* now free all selected PAs */
5471 	list_for_each_entry_safe(pa, tmp, &list, u.pa_tmp_list) {
5472 
5473 		/* remove from object (inode or locality group) */
5474 		if (pa->pa_type == MB_GROUP_PA) {
5475 			spin_lock(pa->pa_node_lock.lg_lock);
5476 			list_del_rcu(&pa->pa_node.lg_list);
5477 			spin_unlock(pa->pa_node_lock.lg_lock);
5478 		} else {
5479 			write_lock(pa->pa_node_lock.inode_lock);
5480 			ei = EXT4_I(pa->pa_inode);
5481 			rb_erase(&pa->pa_node.inode_node, &ei->i_prealloc_node);
5482 			write_unlock(pa->pa_node_lock.inode_lock);
5483 		}
5484 
5485 		list_del(&pa->u.pa_tmp_list);
5486 
5487 		if (pa->pa_type == MB_GROUP_PA) {
5488 			ext4_mb_release_group_pa(&e4b, pa);
5489 			call_rcu(&(pa)->u.pa_rcu, ext4_mb_pa_callback);
5490 		} else {
5491 			ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa);
5492 			ext4_mb_pa_free(pa);
5493 		}
5494 	}
5495 
5496 	ext4_unlock_group(sb, group);
5497 	ext4_mb_unload_buddy(&e4b);
5498 	put_bh(bitmap_bh);
5499 out_dbg:
5500 	mb_debug(sb, "discarded (%d) blocks preallocated for group %u bb_free (%d)\n",
5501 		 free, group, grp->bb_free);
5502 	return free;
5503 }
5504 
5505 /*
5506  * releases all non-used preallocated blocks for given inode
5507  *
5508  * It's important to discard preallocations under i_data_sem
5509  * We don't want another block to be served from the prealloc
5510  * space when we are discarding the inode prealloc space.
5511  *
5512  * FIXME!! Make sure it is valid at all the call sites
5513  */
ext4_discard_preallocations(struct inode * inode)5514 void ext4_discard_preallocations(struct inode *inode)
5515 {
5516 	struct ext4_inode_info *ei = EXT4_I(inode);
5517 	struct super_block *sb = inode->i_sb;
5518 	struct buffer_head *bitmap_bh = NULL;
5519 	struct ext4_prealloc_space *pa, *tmp;
5520 	ext4_group_t group = 0;
5521 	LIST_HEAD(list);
5522 	struct ext4_buddy e4b;
5523 	struct rb_node *iter;
5524 	int err;
5525 
5526 	if (!S_ISREG(inode->i_mode))
5527 		return;
5528 
5529 	if (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY)
5530 		return;
5531 
5532 	mb_debug(sb, "discard preallocation for inode %lu\n",
5533 		 inode->i_ino);
5534 	trace_ext4_discard_preallocations(inode,
5535 			atomic_read(&ei->i_prealloc_active));
5536 
5537 repeat:
5538 	/* first, collect all pa's in the inode */
5539 	write_lock(&ei->i_prealloc_lock);
5540 	for (iter = rb_first(&ei->i_prealloc_node); iter;
5541 	     iter = rb_next(iter)) {
5542 		pa = rb_entry(iter, struct ext4_prealloc_space,
5543 			      pa_node.inode_node);
5544 		BUG_ON(pa->pa_node_lock.inode_lock != &ei->i_prealloc_lock);
5545 
5546 		spin_lock(&pa->pa_lock);
5547 		if (atomic_read(&pa->pa_count)) {
5548 			/* this shouldn't happen often - nobody should
5549 			 * use preallocation while we're discarding it */
5550 			spin_unlock(&pa->pa_lock);
5551 			write_unlock(&ei->i_prealloc_lock);
5552 			ext4_msg(sb, KERN_ERR,
5553 				 "uh-oh! used pa while discarding");
5554 			WARN_ON(1);
5555 			schedule_timeout_uninterruptible(HZ);
5556 			goto repeat;
5557 
5558 		}
5559 		if (pa->pa_deleted == 0) {
5560 			ext4_mb_mark_pa_deleted(sb, pa);
5561 			spin_unlock(&pa->pa_lock);
5562 			rb_erase(&pa->pa_node.inode_node, &ei->i_prealloc_node);
5563 			list_add(&pa->u.pa_tmp_list, &list);
5564 			continue;
5565 		}
5566 
5567 		/* someone is deleting pa right now */
5568 		spin_unlock(&pa->pa_lock);
5569 		write_unlock(&ei->i_prealloc_lock);
5570 
5571 		/* we have to wait here because pa_deleted
5572 		 * doesn't mean pa is already unlinked from
5573 		 * the list. as we might be called from
5574 		 * ->clear_inode() the inode will get freed
5575 		 * and concurrent thread which is unlinking
5576 		 * pa from inode's list may access already
5577 		 * freed memory, bad-bad-bad */
5578 
5579 		/* XXX: if this happens too often, we can
5580 		 * add a flag to force wait only in case
5581 		 * of ->clear_inode(), but not in case of
5582 		 * regular truncate */
5583 		schedule_timeout_uninterruptible(HZ);
5584 		goto repeat;
5585 	}
5586 	write_unlock(&ei->i_prealloc_lock);
5587 
5588 	list_for_each_entry_safe(pa, tmp, &list, u.pa_tmp_list) {
5589 		BUG_ON(pa->pa_type != MB_INODE_PA);
5590 		group = ext4_get_group_number(sb, pa->pa_pstart);
5591 
5592 		err = ext4_mb_load_buddy_gfp(sb, group, &e4b,
5593 					     GFP_NOFS|__GFP_NOFAIL);
5594 		if (err) {
5595 			ext4_error_err(sb, -err, "Error %d loading buddy information for %u",
5596 				       err, group);
5597 			continue;
5598 		}
5599 
5600 		bitmap_bh = ext4_read_block_bitmap(sb, group);
5601 		if (IS_ERR(bitmap_bh)) {
5602 			err = PTR_ERR(bitmap_bh);
5603 			ext4_error_err(sb, -err, "Error %d reading block bitmap for %u",
5604 				       err, group);
5605 			ext4_mb_unload_buddy(&e4b);
5606 			continue;
5607 		}
5608 
5609 		ext4_lock_group(sb, group);
5610 		list_del(&pa->pa_group_list);
5611 		ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa);
5612 		ext4_unlock_group(sb, group);
5613 
5614 		ext4_mb_unload_buddy(&e4b);
5615 		put_bh(bitmap_bh);
5616 
5617 		list_del(&pa->u.pa_tmp_list);
5618 		ext4_mb_pa_free(pa);
5619 	}
5620 }
5621 
ext4_mb_pa_alloc(struct ext4_allocation_context * ac)5622 static int ext4_mb_pa_alloc(struct ext4_allocation_context *ac)
5623 {
5624 	struct ext4_prealloc_space *pa;
5625 
5626 	BUG_ON(ext4_pspace_cachep == NULL);
5627 	pa = kmem_cache_zalloc(ext4_pspace_cachep, GFP_NOFS);
5628 	if (!pa)
5629 		return -ENOMEM;
5630 	atomic_set(&pa->pa_count, 1);
5631 	ac->ac_pa = pa;
5632 	return 0;
5633 }
5634 
ext4_mb_pa_put_free(struct ext4_allocation_context * ac)5635 static void ext4_mb_pa_put_free(struct ext4_allocation_context *ac)
5636 {
5637 	struct ext4_prealloc_space *pa = ac->ac_pa;
5638 
5639 	BUG_ON(!pa);
5640 	ac->ac_pa = NULL;
5641 	WARN_ON(!atomic_dec_and_test(&pa->pa_count));
5642 	/*
5643 	 * current function is only called due to an error or due to
5644 	 * len of found blocks < len of requested blocks hence the PA has not
5645 	 * been added to grp->bb_prealloc_list. So we don't need to lock it
5646 	 */
5647 	pa->pa_deleted = 1;
5648 	ext4_mb_pa_free(pa);
5649 }
5650 
5651 #ifdef CONFIG_EXT4_DEBUG
ext4_mb_show_pa(struct super_block * sb)5652 static inline void ext4_mb_show_pa(struct super_block *sb)
5653 {
5654 	ext4_group_t i, ngroups;
5655 
5656 	if (ext4_forced_shutdown(sb))
5657 		return;
5658 
5659 	ngroups = ext4_get_groups_count(sb);
5660 	mb_debug(sb, "groups: ");
5661 	for (i = 0; i < ngroups; i++) {
5662 		struct ext4_group_info *grp = ext4_get_group_info(sb, i);
5663 		struct ext4_prealloc_space *pa;
5664 		ext4_grpblk_t start;
5665 		struct list_head *cur;
5666 
5667 		if (!grp)
5668 			continue;
5669 		ext4_lock_group(sb, i);
5670 		list_for_each(cur, &grp->bb_prealloc_list) {
5671 			pa = list_entry(cur, struct ext4_prealloc_space,
5672 					pa_group_list);
5673 			spin_lock(&pa->pa_lock);
5674 			ext4_get_group_no_and_offset(sb, pa->pa_pstart,
5675 						     NULL, &start);
5676 			spin_unlock(&pa->pa_lock);
5677 			mb_debug(sb, "PA:%u:%d:%d\n", i, start,
5678 				 pa->pa_len);
5679 		}
5680 		ext4_unlock_group(sb, i);
5681 		mb_debug(sb, "%u: %d/%d\n", i, grp->bb_free,
5682 			 grp->bb_fragments);
5683 	}
5684 }
5685 
ext4_mb_show_ac(struct ext4_allocation_context * ac)5686 static void ext4_mb_show_ac(struct ext4_allocation_context *ac)
5687 {
5688 	struct super_block *sb = ac->ac_sb;
5689 
5690 	if (ext4_forced_shutdown(sb))
5691 		return;
5692 
5693 	mb_debug(sb, "Can't allocate:"
5694 			" Allocation context details:");
5695 	mb_debug(sb, "status %u flags 0x%x",
5696 			ac->ac_status, ac->ac_flags);
5697 	mb_debug(sb, "orig %lu/%lu/%lu@%lu, "
5698 			"goal %lu/%lu/%lu@%lu, "
5699 			"best %lu/%lu/%lu@%lu cr %d",
5700 			(unsigned long)ac->ac_o_ex.fe_group,
5701 			(unsigned long)ac->ac_o_ex.fe_start,
5702 			(unsigned long)ac->ac_o_ex.fe_len,
5703 			(unsigned long)ac->ac_o_ex.fe_logical,
5704 			(unsigned long)ac->ac_g_ex.fe_group,
5705 			(unsigned long)ac->ac_g_ex.fe_start,
5706 			(unsigned long)ac->ac_g_ex.fe_len,
5707 			(unsigned long)ac->ac_g_ex.fe_logical,
5708 			(unsigned long)ac->ac_b_ex.fe_group,
5709 			(unsigned long)ac->ac_b_ex.fe_start,
5710 			(unsigned long)ac->ac_b_ex.fe_len,
5711 			(unsigned long)ac->ac_b_ex.fe_logical,
5712 			(int)ac->ac_criteria);
5713 	mb_debug(sb, "%u found", ac->ac_found);
5714 	mb_debug(sb, "used pa: %s, ", ac->ac_pa ? "yes" : "no");
5715 	if (ac->ac_pa)
5716 		mb_debug(sb, "pa_type %s\n", ac->ac_pa->pa_type == MB_GROUP_PA ?
5717 			 "group pa" : "inode pa");
5718 	ext4_mb_show_pa(sb);
5719 }
5720 #else
ext4_mb_show_pa(struct super_block * sb)5721 static inline void ext4_mb_show_pa(struct super_block *sb)
5722 {
5723 }
ext4_mb_show_ac(struct ext4_allocation_context * ac)5724 static inline void ext4_mb_show_ac(struct ext4_allocation_context *ac)
5725 {
5726 	ext4_mb_show_pa(ac->ac_sb);
5727 }
5728 #endif
5729 
5730 /*
5731  * We use locality group preallocation for small size file. The size of the
5732  * file is determined by the current size or the resulting size after
5733  * allocation which ever is larger
5734  *
5735  * One can tune this size via /sys/fs/ext4/<partition>/mb_stream_req
5736  */
ext4_mb_group_or_file(struct ext4_allocation_context * ac)5737 static void ext4_mb_group_or_file(struct ext4_allocation_context *ac)
5738 {
5739 	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
5740 	int bsbits = ac->ac_sb->s_blocksize_bits;
5741 	loff_t size, isize;
5742 	bool inode_pa_eligible, group_pa_eligible;
5743 
5744 	if (!(ac->ac_flags & EXT4_MB_HINT_DATA))
5745 		return;
5746 
5747 	if (unlikely(ac->ac_flags & EXT4_MB_HINT_GOAL_ONLY))
5748 		return;
5749 
5750 	group_pa_eligible = sbi->s_mb_group_prealloc > 0;
5751 	inode_pa_eligible = true;
5752 	size = extent_logical_end(sbi, &ac->ac_o_ex);
5753 	isize = (i_size_read(ac->ac_inode) + ac->ac_sb->s_blocksize - 1)
5754 		>> bsbits;
5755 
5756 	/* No point in using inode preallocation for closed files */
5757 	if ((size == isize) && !ext4_fs_is_busy(sbi) &&
5758 	    !inode_is_open_for_write(ac->ac_inode))
5759 		inode_pa_eligible = false;
5760 
5761 	size = max(size, isize);
5762 	/* Don't use group allocation for large files */
5763 	if (size > sbi->s_mb_stream_request)
5764 		group_pa_eligible = false;
5765 
5766 	if (!group_pa_eligible) {
5767 		if (inode_pa_eligible)
5768 			ac->ac_flags |= EXT4_MB_STREAM_ALLOC;
5769 		else
5770 			ac->ac_flags |= EXT4_MB_HINT_NOPREALLOC;
5771 		return;
5772 	}
5773 
5774 	BUG_ON(ac->ac_lg != NULL);
5775 	/*
5776 	 * locality group prealloc space are per cpu. The reason for having
5777 	 * per cpu locality group is to reduce the contention between block
5778 	 * request from multiple CPUs.
5779 	 */
5780 	ac->ac_lg = raw_cpu_ptr(sbi->s_locality_groups);
5781 
5782 	/* we're going to use group allocation */
5783 	ac->ac_flags |= EXT4_MB_HINT_GROUP_ALLOC;
5784 
5785 	/* serialize all allocations in the group */
5786 	mutex_lock(&ac->ac_lg->lg_mutex);
5787 }
5788 
5789 static noinline_for_stack void
ext4_mb_initialize_context(struct ext4_allocation_context * ac,struct ext4_allocation_request * ar)5790 ext4_mb_initialize_context(struct ext4_allocation_context *ac,
5791 				struct ext4_allocation_request *ar)
5792 {
5793 	struct super_block *sb = ar->inode->i_sb;
5794 	struct ext4_sb_info *sbi = EXT4_SB(sb);
5795 	struct ext4_super_block *es = sbi->s_es;
5796 	ext4_group_t group;
5797 	unsigned int len;
5798 	ext4_fsblk_t goal;
5799 	ext4_grpblk_t block;
5800 
5801 	/* we can't allocate > group size */
5802 	len = ar->len;
5803 
5804 	/* just a dirty hack to filter too big requests  */
5805 	if (len >= EXT4_CLUSTERS_PER_GROUP(sb))
5806 		len = EXT4_CLUSTERS_PER_GROUP(sb);
5807 
5808 	/* start searching from the goal */
5809 	goal = ar->goal;
5810 	if (goal < le32_to_cpu(es->s_first_data_block) ||
5811 			goal >= ext4_blocks_count(es))
5812 		goal = le32_to_cpu(es->s_first_data_block);
5813 	ext4_get_group_no_and_offset(sb, goal, &group, &block);
5814 
5815 	/* set up allocation goals */
5816 	ac->ac_b_ex.fe_logical = EXT4_LBLK_CMASK(sbi, ar->logical);
5817 	ac->ac_status = AC_STATUS_CONTINUE;
5818 	ac->ac_sb = sb;
5819 	ac->ac_inode = ar->inode;
5820 	ac->ac_o_ex.fe_logical = ac->ac_b_ex.fe_logical;
5821 	ac->ac_o_ex.fe_group = group;
5822 	ac->ac_o_ex.fe_start = block;
5823 	ac->ac_o_ex.fe_len = len;
5824 	ac->ac_g_ex = ac->ac_o_ex;
5825 	ac->ac_orig_goal_len = ac->ac_g_ex.fe_len;
5826 	ac->ac_flags = ar->flags;
5827 
5828 	/* we have to define context: we'll work with a file or
5829 	 * locality group. this is a policy, actually */
5830 	ext4_mb_group_or_file(ac);
5831 
5832 	mb_debug(sb, "init ac: %u blocks @ %u, goal %u, flags 0x%x, 2^%d, "
5833 			"left: %u/%u, right %u/%u to %swritable\n",
5834 			(unsigned) ar->len, (unsigned) ar->logical,
5835 			(unsigned) ar->goal, ac->ac_flags, ac->ac_2order,
5836 			(unsigned) ar->lleft, (unsigned) ar->pleft,
5837 			(unsigned) ar->lright, (unsigned) ar->pright,
5838 			inode_is_open_for_write(ar->inode) ? "" : "non-");
5839 }
5840 
5841 static noinline_for_stack void
ext4_mb_discard_lg_preallocations(struct super_block * sb,struct ext4_locality_group * lg,int order,int total_entries)5842 ext4_mb_discard_lg_preallocations(struct super_block *sb,
5843 					struct ext4_locality_group *lg,
5844 					int order, int total_entries)
5845 {
5846 	ext4_group_t group = 0;
5847 	struct ext4_buddy e4b;
5848 	LIST_HEAD(discard_list);
5849 	struct ext4_prealloc_space *pa, *tmp;
5850 
5851 	mb_debug(sb, "discard locality group preallocation\n");
5852 
5853 	spin_lock(&lg->lg_prealloc_lock);
5854 	list_for_each_entry_rcu(pa, &lg->lg_prealloc_list[order],
5855 				pa_node.lg_list,
5856 				lockdep_is_held(&lg->lg_prealloc_lock)) {
5857 		spin_lock(&pa->pa_lock);
5858 		if (atomic_read(&pa->pa_count)) {
5859 			/*
5860 			 * This is the pa that we just used
5861 			 * for block allocation. So don't
5862 			 * free that
5863 			 */
5864 			spin_unlock(&pa->pa_lock);
5865 			continue;
5866 		}
5867 		if (pa->pa_deleted) {
5868 			spin_unlock(&pa->pa_lock);
5869 			continue;
5870 		}
5871 		/* only lg prealloc space */
5872 		BUG_ON(pa->pa_type != MB_GROUP_PA);
5873 
5874 		/* seems this one can be freed ... */
5875 		ext4_mb_mark_pa_deleted(sb, pa);
5876 		spin_unlock(&pa->pa_lock);
5877 
5878 		list_del_rcu(&pa->pa_node.lg_list);
5879 		list_add(&pa->u.pa_tmp_list, &discard_list);
5880 
5881 		total_entries--;
5882 		if (total_entries <= 5) {
5883 			/*
5884 			 * we want to keep only 5 entries
5885 			 * allowing it to grow to 8. This
5886 			 * mak sure we don't call discard
5887 			 * soon for this list.
5888 			 */
5889 			break;
5890 		}
5891 	}
5892 	spin_unlock(&lg->lg_prealloc_lock);
5893 
5894 	list_for_each_entry_safe(pa, tmp, &discard_list, u.pa_tmp_list) {
5895 		int err;
5896 
5897 		group = ext4_get_group_number(sb, pa->pa_pstart);
5898 		err = ext4_mb_load_buddy_gfp(sb, group, &e4b,
5899 					     GFP_NOFS|__GFP_NOFAIL);
5900 		if (err) {
5901 			ext4_error_err(sb, -err, "Error %d loading buddy information for %u",
5902 				       err, group);
5903 			continue;
5904 		}
5905 		ext4_lock_group(sb, group);
5906 		list_del(&pa->pa_group_list);
5907 		ext4_mb_release_group_pa(&e4b, pa);
5908 		ext4_unlock_group(sb, group);
5909 
5910 		ext4_mb_unload_buddy(&e4b);
5911 		list_del(&pa->u.pa_tmp_list);
5912 		call_rcu(&(pa)->u.pa_rcu, ext4_mb_pa_callback);
5913 	}
5914 }
5915 
5916 /*
5917  * We have incremented pa_count. So it cannot be freed at this
5918  * point. Also we hold lg_mutex. So no parallel allocation is
5919  * possible from this lg. That means pa_free cannot be updated.
5920  *
5921  * A parallel ext4_mb_discard_group_preallocations is possible.
5922  * which can cause the lg_prealloc_list to be updated.
5923  */
5924 
ext4_mb_add_n_trim(struct ext4_allocation_context * ac)5925 static void ext4_mb_add_n_trim(struct ext4_allocation_context *ac)
5926 {
5927 	int order, added = 0, lg_prealloc_count = 1;
5928 	struct super_block *sb = ac->ac_sb;
5929 	struct ext4_locality_group *lg = ac->ac_lg;
5930 	struct ext4_prealloc_space *tmp_pa, *pa = ac->ac_pa;
5931 
5932 	order = fls(pa->pa_free) - 1;
5933 	if (order > PREALLOC_TB_SIZE - 1)
5934 		/* The max size of hash table is PREALLOC_TB_SIZE */
5935 		order = PREALLOC_TB_SIZE - 1;
5936 	/* Add the prealloc space to lg */
5937 	spin_lock(&lg->lg_prealloc_lock);
5938 	list_for_each_entry_rcu(tmp_pa, &lg->lg_prealloc_list[order],
5939 				pa_node.lg_list,
5940 				lockdep_is_held(&lg->lg_prealloc_lock)) {
5941 		spin_lock(&tmp_pa->pa_lock);
5942 		if (tmp_pa->pa_deleted) {
5943 			spin_unlock(&tmp_pa->pa_lock);
5944 			continue;
5945 		}
5946 		if (!added && pa->pa_free < tmp_pa->pa_free) {
5947 			/* Add to the tail of the previous entry */
5948 			list_add_tail_rcu(&pa->pa_node.lg_list,
5949 						&tmp_pa->pa_node.lg_list);
5950 			added = 1;
5951 			/*
5952 			 * we want to count the total
5953 			 * number of entries in the list
5954 			 */
5955 		}
5956 		spin_unlock(&tmp_pa->pa_lock);
5957 		lg_prealloc_count++;
5958 	}
5959 	if (!added)
5960 		list_add_tail_rcu(&pa->pa_node.lg_list,
5961 					&lg->lg_prealloc_list[order]);
5962 	spin_unlock(&lg->lg_prealloc_lock);
5963 
5964 	/* Now trim the list to be not more than 8 elements */
5965 	if (lg_prealloc_count > 8)
5966 		ext4_mb_discard_lg_preallocations(sb, lg,
5967 						  order, lg_prealloc_count);
5968 }
5969 
5970 /*
5971  * release all resource we used in allocation
5972  */
ext4_mb_release_context(struct ext4_allocation_context * ac)5973 static void ext4_mb_release_context(struct ext4_allocation_context *ac)
5974 {
5975 	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
5976 	struct ext4_prealloc_space *pa = ac->ac_pa;
5977 	if (pa) {
5978 		if (pa->pa_type == MB_GROUP_PA) {
5979 			/* see comment in ext4_mb_use_group_pa() */
5980 			spin_lock(&pa->pa_lock);
5981 			pa->pa_pstart += EXT4_C2B(sbi, ac->ac_b_ex.fe_len);
5982 			pa->pa_lstart += EXT4_C2B(sbi, ac->ac_b_ex.fe_len);
5983 			pa->pa_free -= ac->ac_b_ex.fe_len;
5984 			pa->pa_len -= ac->ac_b_ex.fe_len;
5985 			spin_unlock(&pa->pa_lock);
5986 
5987 			/*
5988 			 * We want to add the pa to the right bucket.
5989 			 * Remove it from the list and while adding
5990 			 * make sure the list to which we are adding
5991 			 * doesn't grow big.
5992 			 */
5993 			if (likely(pa->pa_free)) {
5994 				spin_lock(pa->pa_node_lock.lg_lock);
5995 				list_del_rcu(&pa->pa_node.lg_list);
5996 				spin_unlock(pa->pa_node_lock.lg_lock);
5997 				ext4_mb_add_n_trim(ac);
5998 			}
5999 		}
6000 
6001 		ext4_mb_put_pa(ac, ac->ac_sb, pa);
6002 	}
6003 	if (ac->ac_bitmap_folio)
6004 		folio_put(ac->ac_bitmap_folio);
6005 	if (ac->ac_buddy_folio)
6006 		folio_put(ac->ac_buddy_folio);
6007 	if (ac->ac_flags & EXT4_MB_HINT_GROUP_ALLOC)
6008 		mutex_unlock(&ac->ac_lg->lg_mutex);
6009 	ext4_mb_collect_stats(ac);
6010 }
6011 
ext4_mb_discard_preallocations(struct super_block * sb,int needed)6012 static int ext4_mb_discard_preallocations(struct super_block *sb, int needed)
6013 {
6014 	ext4_group_t i, ngroups = ext4_get_groups_count(sb);
6015 	int ret;
6016 	int freed = 0, busy = 0;
6017 	int retry = 0;
6018 
6019 	trace_ext4_mb_discard_preallocations(sb, needed);
6020 
6021 	if (needed == 0)
6022 		needed = EXT4_CLUSTERS_PER_GROUP(sb) + 1;
6023  repeat:
6024 	for (i = 0; i < ngroups && needed > 0; i++) {
6025 		ret = ext4_mb_discard_group_preallocations(sb, i, &busy);
6026 		freed += ret;
6027 		needed -= ret;
6028 		cond_resched();
6029 	}
6030 
6031 	if (needed > 0 && busy && ++retry < 3) {
6032 		busy = 0;
6033 		goto repeat;
6034 	}
6035 
6036 	return freed;
6037 }
6038 
ext4_mb_discard_preallocations_should_retry(struct super_block * sb,struct ext4_allocation_context * ac,u64 * seq)6039 static bool ext4_mb_discard_preallocations_should_retry(struct super_block *sb,
6040 			struct ext4_allocation_context *ac, u64 *seq)
6041 {
6042 	int freed;
6043 	u64 seq_retry = 0;
6044 	bool ret = false;
6045 
6046 	freed = ext4_mb_discard_preallocations(sb, ac->ac_o_ex.fe_len);
6047 	if (freed) {
6048 		ret = true;
6049 		goto out_dbg;
6050 	}
6051 	seq_retry = ext4_get_discard_pa_seq_sum();
6052 	if (!(ac->ac_flags & EXT4_MB_STRICT_CHECK) || seq_retry != *seq) {
6053 		ac->ac_flags |= EXT4_MB_STRICT_CHECK;
6054 		*seq = seq_retry;
6055 		ret = true;
6056 	}
6057 
6058 out_dbg:
6059 	mb_debug(sb, "freed %d, retry ? %s\n", freed, ret ? "yes" : "no");
6060 	return ret;
6061 }
6062 
6063 /*
6064  * Simple allocator for Ext4 fast commit replay path. It searches for blocks
6065  * linearly starting at the goal block and also excludes the blocks which
6066  * are going to be in use after fast commit replay.
6067  */
6068 static ext4_fsblk_t
ext4_mb_new_blocks_simple(struct ext4_allocation_request * ar,int * errp)6069 ext4_mb_new_blocks_simple(struct ext4_allocation_request *ar, int *errp)
6070 {
6071 	struct buffer_head *bitmap_bh;
6072 	struct super_block *sb = ar->inode->i_sb;
6073 	struct ext4_sb_info *sbi = EXT4_SB(sb);
6074 	ext4_group_t group, nr;
6075 	ext4_grpblk_t blkoff;
6076 	ext4_grpblk_t max = EXT4_CLUSTERS_PER_GROUP(sb);
6077 	ext4_grpblk_t i = 0;
6078 	ext4_fsblk_t goal, block;
6079 	struct ext4_super_block *es = sbi->s_es;
6080 
6081 	goal = ar->goal;
6082 	if (goal < le32_to_cpu(es->s_first_data_block) ||
6083 			goal >= ext4_blocks_count(es))
6084 		goal = le32_to_cpu(es->s_first_data_block);
6085 
6086 	ar->len = 0;
6087 	ext4_get_group_no_and_offset(sb, goal, &group, &blkoff);
6088 	for (nr = ext4_get_groups_count(sb); nr > 0; nr--) {
6089 		bitmap_bh = ext4_read_block_bitmap(sb, group);
6090 		if (IS_ERR(bitmap_bh)) {
6091 			*errp = PTR_ERR(bitmap_bh);
6092 			pr_warn("Failed to read block bitmap\n");
6093 			return 0;
6094 		}
6095 
6096 		while (1) {
6097 			i = mb_find_next_zero_bit(bitmap_bh->b_data, max,
6098 						blkoff);
6099 			if (i >= max)
6100 				break;
6101 			if (ext4_fc_replay_check_excluded(sb,
6102 				ext4_group_first_block_no(sb, group) +
6103 				EXT4_C2B(sbi, i))) {
6104 				blkoff = i + 1;
6105 			} else
6106 				break;
6107 		}
6108 		brelse(bitmap_bh);
6109 		if (i < max)
6110 			break;
6111 
6112 		if (++group >= ext4_get_groups_count(sb))
6113 			group = 0;
6114 
6115 		blkoff = 0;
6116 	}
6117 
6118 	if (i >= max) {
6119 		*errp = -ENOSPC;
6120 		return 0;
6121 	}
6122 
6123 	block = ext4_group_first_block_no(sb, group) + EXT4_C2B(sbi, i);
6124 	ext4_mb_mark_bb(sb, block, 1, true);
6125 	ar->len = 1;
6126 
6127 	*errp = 0;
6128 	return block;
6129 }
6130 
6131 /*
6132  * Main entry point into mballoc to allocate blocks
6133  * it tries to use preallocation first, then falls back
6134  * to usual allocation
6135  */
ext4_mb_new_blocks(handle_t * handle,struct ext4_allocation_request * ar,int * errp)6136 ext4_fsblk_t ext4_mb_new_blocks(handle_t *handle,
6137 				struct ext4_allocation_request *ar, int *errp)
6138 {
6139 	struct ext4_allocation_context *ac = NULL;
6140 	struct ext4_sb_info *sbi;
6141 	struct super_block *sb;
6142 	ext4_fsblk_t block = 0;
6143 	unsigned int inquota = 0;
6144 	unsigned int reserv_clstrs = 0;
6145 	int retries = 0;
6146 	u64 seq;
6147 
6148 	might_sleep();
6149 	sb = ar->inode->i_sb;
6150 	sbi = EXT4_SB(sb);
6151 
6152 	trace_ext4_request_blocks(ar);
6153 	if (sbi->s_mount_state & EXT4_FC_REPLAY)
6154 		return ext4_mb_new_blocks_simple(ar, errp);
6155 
6156 	/* Allow to use superuser reservation for quota file */
6157 	if (ext4_is_quota_file(ar->inode))
6158 		ar->flags |= EXT4_MB_USE_ROOT_BLOCKS;
6159 
6160 	if ((ar->flags & EXT4_MB_DELALLOC_RESERVED) == 0) {
6161 		/* Without delayed allocation we need to verify
6162 		 * there is enough free blocks to do block allocation
6163 		 * and verify allocation doesn't exceed the quota limits.
6164 		 */
6165 		while (ar->len &&
6166 			ext4_claim_free_clusters(sbi, ar->len, ar->flags)) {
6167 
6168 			/* let others to free the space */
6169 			cond_resched();
6170 			ar->len = ar->len >> 1;
6171 		}
6172 		if (!ar->len) {
6173 			ext4_mb_show_pa(sb);
6174 			*errp = -ENOSPC;
6175 			return 0;
6176 		}
6177 		reserv_clstrs = ar->len;
6178 		if (ar->flags & EXT4_MB_USE_ROOT_BLOCKS) {
6179 			dquot_alloc_block_nofail(ar->inode,
6180 						 EXT4_C2B(sbi, ar->len));
6181 		} else {
6182 			while (ar->len &&
6183 				dquot_alloc_block(ar->inode,
6184 						  EXT4_C2B(sbi, ar->len))) {
6185 
6186 				ar->flags |= EXT4_MB_HINT_NOPREALLOC;
6187 				ar->len--;
6188 			}
6189 		}
6190 		inquota = ar->len;
6191 		if (ar->len == 0) {
6192 			*errp = -EDQUOT;
6193 			goto out;
6194 		}
6195 	}
6196 
6197 	ac = kmem_cache_zalloc(ext4_ac_cachep, GFP_NOFS);
6198 	if (!ac) {
6199 		ar->len = 0;
6200 		*errp = -ENOMEM;
6201 		goto out;
6202 	}
6203 
6204 	ext4_mb_initialize_context(ac, ar);
6205 
6206 	ac->ac_op = EXT4_MB_HISTORY_PREALLOC;
6207 	seq = this_cpu_read(discard_pa_seq);
6208 	if (!ext4_mb_use_preallocated(ac)) {
6209 		ac->ac_op = EXT4_MB_HISTORY_ALLOC;
6210 		ext4_mb_normalize_request(ac, ar);
6211 
6212 		*errp = ext4_mb_pa_alloc(ac);
6213 		if (*errp)
6214 			goto errout;
6215 repeat:
6216 		/* allocate space in core */
6217 		*errp = ext4_mb_regular_allocator(ac);
6218 		/*
6219 		 * pa allocated above is added to grp->bb_prealloc_list only
6220 		 * when we were able to allocate some block i.e. when
6221 		 * ac->ac_status == AC_STATUS_FOUND.
6222 		 * And error from above mean ac->ac_status != AC_STATUS_FOUND
6223 		 * So we have to free this pa here itself.
6224 		 */
6225 		if (*errp) {
6226 			ext4_mb_pa_put_free(ac);
6227 			ext4_discard_allocated_blocks(ac);
6228 			goto errout;
6229 		}
6230 		if (ac->ac_status == AC_STATUS_FOUND &&
6231 			ac->ac_o_ex.fe_len >= ac->ac_f_ex.fe_len)
6232 			ext4_mb_pa_put_free(ac);
6233 	}
6234 	if (likely(ac->ac_status == AC_STATUS_FOUND)) {
6235 		*errp = ext4_mb_mark_diskspace_used(ac, handle, reserv_clstrs);
6236 		if (*errp) {
6237 			ext4_discard_allocated_blocks(ac);
6238 			goto errout;
6239 		} else {
6240 			block = ext4_grp_offs_to_block(sb, &ac->ac_b_ex);
6241 			ar->len = ac->ac_b_ex.fe_len;
6242 		}
6243 	} else {
6244 		if (++retries < 3 &&
6245 		    ext4_mb_discard_preallocations_should_retry(sb, ac, &seq))
6246 			goto repeat;
6247 		/*
6248 		 * If block allocation fails then the pa allocated above
6249 		 * needs to be freed here itself.
6250 		 */
6251 		ext4_mb_pa_put_free(ac);
6252 		*errp = -ENOSPC;
6253 	}
6254 
6255 	if (*errp) {
6256 errout:
6257 		ac->ac_b_ex.fe_len = 0;
6258 		ar->len = 0;
6259 		ext4_mb_show_ac(ac);
6260 	}
6261 	ext4_mb_release_context(ac);
6262 	kmem_cache_free(ext4_ac_cachep, ac);
6263 out:
6264 	if (inquota && ar->len < inquota)
6265 		dquot_free_block(ar->inode, EXT4_C2B(sbi, inquota - ar->len));
6266 	if (!ar->len) {
6267 		if ((ar->flags & EXT4_MB_DELALLOC_RESERVED) == 0)
6268 			/* release all the reserved blocks if non delalloc */
6269 			percpu_counter_sub(&sbi->s_dirtyclusters_counter,
6270 						reserv_clstrs);
6271 	}
6272 
6273 	trace_ext4_allocate_blocks(ar, (unsigned long long)block);
6274 
6275 	return block;
6276 }
6277 
6278 /*
6279  * We can merge two free data extents only if the physical blocks
6280  * are contiguous, AND the extents were freed by the same transaction,
6281  * AND the blocks are associated with the same group.
6282  */
ext4_try_merge_freed_extent(struct ext4_sb_info * sbi,struct ext4_free_data * entry,struct ext4_free_data * new_entry,struct rb_root * entry_rb_root)6283 static void ext4_try_merge_freed_extent(struct ext4_sb_info *sbi,
6284 					struct ext4_free_data *entry,
6285 					struct ext4_free_data *new_entry,
6286 					struct rb_root *entry_rb_root)
6287 {
6288 	if ((entry->efd_tid != new_entry->efd_tid) ||
6289 	    (entry->efd_group != new_entry->efd_group))
6290 		return;
6291 	if (entry->efd_start_cluster + entry->efd_count ==
6292 	    new_entry->efd_start_cluster) {
6293 		new_entry->efd_start_cluster = entry->efd_start_cluster;
6294 		new_entry->efd_count += entry->efd_count;
6295 	} else if (new_entry->efd_start_cluster + new_entry->efd_count ==
6296 		   entry->efd_start_cluster) {
6297 		new_entry->efd_count += entry->efd_count;
6298 	} else
6299 		return;
6300 	spin_lock(&sbi->s_md_lock);
6301 	list_del(&entry->efd_list);
6302 	spin_unlock(&sbi->s_md_lock);
6303 	rb_erase(&entry->efd_node, entry_rb_root);
6304 	kmem_cache_free(ext4_free_data_cachep, entry);
6305 }
6306 
6307 static noinline_for_stack void
ext4_mb_free_metadata(handle_t * handle,struct ext4_buddy * e4b,struct ext4_free_data * new_entry)6308 ext4_mb_free_metadata(handle_t *handle, struct ext4_buddy *e4b,
6309 		      struct ext4_free_data *new_entry)
6310 {
6311 	ext4_group_t group = e4b->bd_group;
6312 	ext4_grpblk_t cluster;
6313 	ext4_grpblk_t clusters = new_entry->efd_count;
6314 	struct ext4_free_data *entry;
6315 	struct ext4_group_info *db = e4b->bd_info;
6316 	struct super_block *sb = e4b->bd_sb;
6317 	struct ext4_sb_info *sbi = EXT4_SB(sb);
6318 	struct rb_node **n = &db->bb_free_root.rb_node, *node;
6319 	struct rb_node *parent = NULL, *new_node;
6320 
6321 	BUG_ON(!ext4_handle_valid(handle));
6322 	BUG_ON(e4b->bd_bitmap_folio == NULL);
6323 	BUG_ON(e4b->bd_buddy_folio == NULL);
6324 
6325 	new_node = &new_entry->efd_node;
6326 	cluster = new_entry->efd_start_cluster;
6327 
6328 	if (!*n) {
6329 		/* first free block exent. We need to
6330 		   protect buddy cache from being freed,
6331 		 * otherwise we'll refresh it from
6332 		 * on-disk bitmap and lose not-yet-available
6333 		 * blocks */
6334 		folio_get(e4b->bd_buddy_folio);
6335 		folio_get(e4b->bd_bitmap_folio);
6336 	}
6337 	while (*n) {
6338 		parent = *n;
6339 		entry = rb_entry(parent, struct ext4_free_data, efd_node);
6340 		if (cluster < entry->efd_start_cluster)
6341 			n = &(*n)->rb_left;
6342 		else if (cluster >= (entry->efd_start_cluster + entry->efd_count))
6343 			n = &(*n)->rb_right;
6344 		else {
6345 			ext4_grp_locked_error(sb, group, 0,
6346 				ext4_group_first_block_no(sb, group) +
6347 				EXT4_C2B(sbi, cluster),
6348 				"Block already on to-be-freed list");
6349 			kmem_cache_free(ext4_free_data_cachep, new_entry);
6350 			return;
6351 		}
6352 	}
6353 
6354 	rb_link_node(new_node, parent, n);
6355 	rb_insert_color(new_node, &db->bb_free_root);
6356 
6357 	/* Now try to see the extent can be merged to left and right */
6358 	node = rb_prev(new_node);
6359 	if (node) {
6360 		entry = rb_entry(node, struct ext4_free_data, efd_node);
6361 		ext4_try_merge_freed_extent(sbi, entry, new_entry,
6362 					    &(db->bb_free_root));
6363 	}
6364 
6365 	node = rb_next(new_node);
6366 	if (node) {
6367 		entry = rb_entry(node, struct ext4_free_data, efd_node);
6368 		ext4_try_merge_freed_extent(sbi, entry, new_entry,
6369 					    &(db->bb_free_root));
6370 	}
6371 
6372 	spin_lock(&sbi->s_md_lock);
6373 	list_add_tail(&new_entry->efd_list, &sbi->s_freed_data_list[new_entry->efd_tid & 1]);
6374 	sbi->s_mb_free_pending += clusters;
6375 	spin_unlock(&sbi->s_md_lock);
6376 }
6377 
ext4_free_blocks_simple(struct inode * inode,ext4_fsblk_t block,unsigned long count)6378 static void ext4_free_blocks_simple(struct inode *inode, ext4_fsblk_t block,
6379 					unsigned long count)
6380 {
6381 	struct super_block *sb = inode->i_sb;
6382 	ext4_group_t group;
6383 	ext4_grpblk_t blkoff;
6384 
6385 	ext4_get_group_no_and_offset(sb, block, &group, &blkoff);
6386 	ext4_mb_mark_context(NULL, sb, false, group, blkoff, count,
6387 			     EXT4_MB_BITMAP_MARKED_CHECK |
6388 			     EXT4_MB_SYNC_UPDATE,
6389 			     NULL);
6390 }
6391 
6392 /**
6393  * ext4_mb_clear_bb() -- helper function for freeing blocks.
6394  *			Used by ext4_free_blocks()
6395  * @handle:		handle for this transaction
6396  * @inode:		inode
6397  * @block:		starting physical block to be freed
6398  * @count:		number of blocks to be freed
6399  * @flags:		flags used by ext4_free_blocks
6400  */
ext4_mb_clear_bb(handle_t * handle,struct inode * inode,ext4_fsblk_t block,unsigned long count,int flags)6401 static void ext4_mb_clear_bb(handle_t *handle, struct inode *inode,
6402 			       ext4_fsblk_t block, unsigned long count,
6403 			       int flags)
6404 {
6405 	struct super_block *sb = inode->i_sb;
6406 	struct ext4_group_info *grp;
6407 	unsigned int overflow;
6408 	ext4_grpblk_t bit;
6409 	ext4_group_t block_group;
6410 	struct ext4_sb_info *sbi;
6411 	struct ext4_buddy e4b;
6412 	unsigned int count_clusters;
6413 	int err = 0;
6414 	int mark_flags = 0;
6415 	ext4_grpblk_t changed;
6416 
6417 	sbi = EXT4_SB(sb);
6418 
6419 	if (!(flags & EXT4_FREE_BLOCKS_VALIDATED) &&
6420 	    !ext4_inode_block_valid(inode, block, count)) {
6421 		ext4_error(sb, "Freeing blocks in system zone - "
6422 			   "Block = %llu, count = %lu", block, count);
6423 		/* err = 0. ext4_std_error should be a no op */
6424 		goto error_out;
6425 	}
6426 	flags |= EXT4_FREE_BLOCKS_VALIDATED;
6427 
6428 do_more:
6429 	overflow = 0;
6430 	ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
6431 
6432 	grp = ext4_get_group_info(sb, block_group);
6433 	if (unlikely(!grp || EXT4_MB_GRP_BBITMAP_CORRUPT(grp)))
6434 		return;
6435 
6436 	/*
6437 	 * Check to see if we are freeing blocks across a group
6438 	 * boundary.
6439 	 */
6440 	if (EXT4_C2B(sbi, bit) + count > EXT4_BLOCKS_PER_GROUP(sb)) {
6441 		overflow = EXT4_C2B(sbi, bit) + count -
6442 			EXT4_BLOCKS_PER_GROUP(sb);
6443 		count -= overflow;
6444 		/* The range changed so it's no longer validated */
6445 		flags &= ~EXT4_FREE_BLOCKS_VALIDATED;
6446 	}
6447 	count_clusters = EXT4_NUM_B2C(sbi, count);
6448 	trace_ext4_mballoc_free(sb, inode, block_group, bit, count_clusters);
6449 
6450 	/* __GFP_NOFAIL: retry infinitely, ignore TIF_MEMDIE and memcg limit. */
6451 	err = ext4_mb_load_buddy_gfp(sb, block_group, &e4b,
6452 				     GFP_NOFS|__GFP_NOFAIL);
6453 	if (err)
6454 		goto error_out;
6455 
6456 	if (!(flags & EXT4_FREE_BLOCKS_VALIDATED) &&
6457 	    !ext4_inode_block_valid(inode, block, count)) {
6458 		ext4_error(sb, "Freeing blocks in system zone - "
6459 			   "Block = %llu, count = %lu", block, count);
6460 		/* err = 0. ext4_std_error should be a no op */
6461 		goto error_clean;
6462 	}
6463 
6464 #ifdef AGGRESSIVE_CHECK
6465 	mark_flags |= EXT4_MB_BITMAP_MARKED_CHECK;
6466 #endif
6467 	err = ext4_mb_mark_context(handle, sb, false, block_group, bit,
6468 				   count_clusters, mark_flags, &changed);
6469 
6470 
6471 	if (err && changed == 0)
6472 		goto error_clean;
6473 
6474 #ifdef AGGRESSIVE_CHECK
6475 	BUG_ON(changed != count_clusters);
6476 #endif
6477 
6478 	/*
6479 	 * We need to make sure we don't reuse the freed block until after the
6480 	 * transaction is committed. We make an exception if the inode is to be
6481 	 * written in writeback mode since writeback mode has weak data
6482 	 * consistency guarantees.
6483 	 */
6484 	if (ext4_handle_valid(handle) &&
6485 	    ((flags & EXT4_FREE_BLOCKS_METADATA) ||
6486 	     !ext4_should_writeback_data(inode))) {
6487 		struct ext4_free_data *new_entry;
6488 		/*
6489 		 * We use __GFP_NOFAIL because ext4_free_blocks() is not allowed
6490 		 * to fail.
6491 		 */
6492 		new_entry = kmem_cache_alloc(ext4_free_data_cachep,
6493 				GFP_NOFS|__GFP_NOFAIL);
6494 		new_entry->efd_start_cluster = bit;
6495 		new_entry->efd_group = block_group;
6496 		new_entry->efd_count = count_clusters;
6497 		new_entry->efd_tid = handle->h_transaction->t_tid;
6498 
6499 		ext4_lock_group(sb, block_group);
6500 		ext4_mb_free_metadata(handle, &e4b, new_entry);
6501 	} else {
6502 		if (test_opt(sb, DISCARD)) {
6503 			err = ext4_issue_discard(sb, block_group, bit,
6504 						 count_clusters);
6505 			/*
6506 			 * Ignore EOPNOTSUPP error. This is consistent with
6507 			 * what happens when using journal.
6508 			 */
6509 			if (err == -EOPNOTSUPP)
6510 				err = 0;
6511 			if (err)
6512 				ext4_msg(sb, KERN_WARNING, "discard request in"
6513 					 " group:%u block:%d count:%lu failed"
6514 					 " with %d", block_group, bit, count,
6515 					 err);
6516 		}
6517 
6518 		EXT4_MB_GRP_CLEAR_TRIMMED(e4b.bd_info);
6519 
6520 		ext4_lock_group(sb, block_group);
6521 		mb_free_blocks(inode, &e4b, bit, count_clusters);
6522 	}
6523 
6524 	ext4_unlock_group(sb, block_group);
6525 
6526 	/*
6527 	 * on a bigalloc file system, defer the s_freeclusters_counter
6528 	 * update to the caller (ext4_remove_space and friends) so they
6529 	 * can determine if a cluster freed here should be rereserved
6530 	 */
6531 	if (!(flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)) {
6532 		if (!(flags & EXT4_FREE_BLOCKS_NO_QUOT_UPDATE))
6533 			dquot_free_block(inode, EXT4_C2B(sbi, count_clusters));
6534 		percpu_counter_add(&sbi->s_freeclusters_counter,
6535 				   count_clusters);
6536 	}
6537 
6538 	if (overflow && !err) {
6539 		block += count;
6540 		count = overflow;
6541 		ext4_mb_unload_buddy(&e4b);
6542 		/* The range changed so it's no longer validated */
6543 		flags &= ~EXT4_FREE_BLOCKS_VALIDATED;
6544 		goto do_more;
6545 	}
6546 
6547 error_clean:
6548 	ext4_mb_unload_buddy(&e4b);
6549 error_out:
6550 	ext4_std_error(sb, err);
6551 }
6552 
6553 /**
6554  * ext4_free_blocks() -- Free given blocks and update quota
6555  * @handle:		handle for this transaction
6556  * @inode:		inode
6557  * @bh:			optional buffer of the block to be freed
6558  * @block:		starting physical block to be freed
6559  * @count:		number of blocks to be freed
6560  * @flags:		flags used by ext4_free_blocks
6561  */
ext4_free_blocks(handle_t * handle,struct inode * inode,struct buffer_head * bh,ext4_fsblk_t block,unsigned long count,int flags)6562 void ext4_free_blocks(handle_t *handle, struct inode *inode,
6563 		      struct buffer_head *bh, ext4_fsblk_t block,
6564 		      unsigned long count, int flags)
6565 {
6566 	struct super_block *sb = inode->i_sb;
6567 	unsigned int overflow;
6568 	struct ext4_sb_info *sbi;
6569 
6570 	sbi = EXT4_SB(sb);
6571 
6572 	if (bh) {
6573 		if (block)
6574 			BUG_ON(block != bh->b_blocknr);
6575 		else
6576 			block = bh->b_blocknr;
6577 	}
6578 
6579 	if (sbi->s_mount_state & EXT4_FC_REPLAY) {
6580 		ext4_free_blocks_simple(inode, block, EXT4_NUM_B2C(sbi, count));
6581 		return;
6582 	}
6583 
6584 	might_sleep();
6585 
6586 	if (!(flags & EXT4_FREE_BLOCKS_VALIDATED) &&
6587 	    !ext4_inode_block_valid(inode, block, count)) {
6588 		ext4_error(sb, "Freeing blocks not in datazone - "
6589 			   "block = %llu, count = %lu", block, count);
6590 		return;
6591 	}
6592 	flags |= EXT4_FREE_BLOCKS_VALIDATED;
6593 
6594 	ext4_debug("freeing block %llu\n", block);
6595 	trace_ext4_free_blocks(inode, block, count, flags);
6596 
6597 	if (bh && (flags & EXT4_FREE_BLOCKS_FORGET)) {
6598 		BUG_ON(count > 1);
6599 
6600 		ext4_forget(handle, flags & EXT4_FREE_BLOCKS_METADATA,
6601 			    inode, bh, block);
6602 	}
6603 
6604 	/*
6605 	 * If the extent to be freed does not begin on a cluster
6606 	 * boundary, we need to deal with partial clusters at the
6607 	 * beginning and end of the extent.  Normally we will free
6608 	 * blocks at the beginning or the end unless we are explicitly
6609 	 * requested to avoid doing so.
6610 	 */
6611 	overflow = EXT4_PBLK_COFF(sbi, block);
6612 	if (overflow) {
6613 		if (flags & EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER) {
6614 			overflow = sbi->s_cluster_ratio - overflow;
6615 			block += overflow;
6616 			if (count > overflow)
6617 				count -= overflow;
6618 			else
6619 				return;
6620 		} else {
6621 			block -= overflow;
6622 			count += overflow;
6623 		}
6624 		/* The range changed so it's no longer validated */
6625 		flags &= ~EXT4_FREE_BLOCKS_VALIDATED;
6626 	}
6627 	overflow = EXT4_LBLK_COFF(sbi, count);
6628 	if (overflow) {
6629 		if (flags & EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER) {
6630 			if (count > overflow)
6631 				count -= overflow;
6632 			else
6633 				return;
6634 		} else
6635 			count += sbi->s_cluster_ratio - overflow;
6636 		/* The range changed so it's no longer validated */
6637 		flags &= ~EXT4_FREE_BLOCKS_VALIDATED;
6638 	}
6639 
6640 	if (!bh && (flags & EXT4_FREE_BLOCKS_FORGET)) {
6641 		int i;
6642 		int is_metadata = flags & EXT4_FREE_BLOCKS_METADATA;
6643 
6644 		for (i = 0; i < count; i++) {
6645 			cond_resched();
6646 			if (is_metadata)
6647 				bh = sb_find_get_block(inode->i_sb, block + i);
6648 			ext4_forget(handle, is_metadata, inode, bh, block + i);
6649 		}
6650 	}
6651 
6652 	ext4_mb_clear_bb(handle, inode, block, count, flags);
6653 }
6654 
6655 /**
6656  * ext4_group_add_blocks() -- Add given blocks to an existing group
6657  * @handle:			handle to this transaction
6658  * @sb:				super block
6659  * @block:			start physical block to add to the block group
6660  * @count:			number of blocks to free
6661  *
6662  * This marks the blocks as free in the bitmap and buddy.
6663  */
ext4_group_add_blocks(handle_t * handle,struct super_block * sb,ext4_fsblk_t block,unsigned long count)6664 int ext4_group_add_blocks(handle_t *handle, struct super_block *sb,
6665 			 ext4_fsblk_t block, unsigned long count)
6666 {
6667 	ext4_group_t block_group;
6668 	ext4_grpblk_t bit;
6669 	struct ext4_sb_info *sbi = EXT4_SB(sb);
6670 	struct ext4_buddy e4b;
6671 	int err = 0;
6672 	ext4_fsblk_t first_cluster = EXT4_B2C(sbi, block);
6673 	ext4_fsblk_t last_cluster = EXT4_B2C(sbi, block + count - 1);
6674 	unsigned long cluster_count = last_cluster - first_cluster + 1;
6675 	ext4_grpblk_t changed;
6676 
6677 	ext4_debug("Adding block(s) %llu-%llu\n", block, block + count - 1);
6678 
6679 	if (cluster_count == 0)
6680 		return 0;
6681 
6682 	ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
6683 	/*
6684 	 * Check to see if we are freeing blocks across a group
6685 	 * boundary.
6686 	 */
6687 	if (bit + cluster_count > EXT4_CLUSTERS_PER_GROUP(sb)) {
6688 		ext4_warning(sb, "too many blocks added to group %u",
6689 			     block_group);
6690 		err = -EINVAL;
6691 		goto error_out;
6692 	}
6693 
6694 	err = ext4_mb_load_buddy(sb, block_group, &e4b);
6695 	if (err)
6696 		goto error_out;
6697 
6698 	if (!ext4_sb_block_valid(sb, NULL, block, count)) {
6699 		ext4_error(sb, "Adding blocks in system zones - "
6700 			   "Block = %llu, count = %lu",
6701 			   block, count);
6702 		err = -EINVAL;
6703 		goto error_clean;
6704 	}
6705 
6706 	err = ext4_mb_mark_context(handle, sb, false, block_group, bit,
6707 				   cluster_count, EXT4_MB_BITMAP_MARKED_CHECK,
6708 				   &changed);
6709 	if (err && changed == 0)
6710 		goto error_clean;
6711 
6712 	if (changed != cluster_count)
6713 		ext4_error(sb, "bit already cleared in group %u", block_group);
6714 
6715 	ext4_lock_group(sb, block_group);
6716 	mb_free_blocks(NULL, &e4b, bit, cluster_count);
6717 	ext4_unlock_group(sb, block_group);
6718 	percpu_counter_add(&sbi->s_freeclusters_counter,
6719 			   changed);
6720 
6721 error_clean:
6722 	ext4_mb_unload_buddy(&e4b);
6723 error_out:
6724 	ext4_std_error(sb, err);
6725 	return err;
6726 }
6727 
6728 /**
6729  * ext4_trim_extent -- function to TRIM one single free extent in the group
6730  * @sb:		super block for the file system
6731  * @start:	starting block of the free extent in the alloc. group
6732  * @count:	number of blocks to TRIM
6733  * @e4b:	ext4 buddy for the group
6734  *
6735  * Trim "count" blocks starting at "start" in the "group". To assure that no
6736  * one will allocate those blocks, mark it as used in buddy bitmap. This must
6737  * be called with under the group lock.
6738  */
ext4_trim_extent(struct super_block * sb,int start,int count,struct ext4_buddy * e4b)6739 static int ext4_trim_extent(struct super_block *sb,
6740 		int start, int count, struct ext4_buddy *e4b)
6741 __releases(bitlock)
6742 __acquires(bitlock)
6743 {
6744 	struct ext4_free_extent ex;
6745 	ext4_group_t group = e4b->bd_group;
6746 	int ret = 0;
6747 
6748 	trace_ext4_trim_extent(sb, group, start, count);
6749 
6750 	assert_spin_locked(ext4_group_lock_ptr(sb, group));
6751 
6752 	ex.fe_start = start;
6753 	ex.fe_group = group;
6754 	ex.fe_len = count;
6755 
6756 	/*
6757 	 * Mark blocks used, so no one can reuse them while
6758 	 * being trimmed.
6759 	 */
6760 	mb_mark_used(e4b, &ex);
6761 	ext4_unlock_group(sb, group);
6762 	ret = ext4_issue_discard(sb, group, start, count);
6763 	ext4_lock_group(sb, group);
6764 	mb_free_blocks(NULL, e4b, start, ex.fe_len);
6765 	return ret;
6766 }
6767 
ext4_last_grp_cluster(struct super_block * sb,ext4_group_t grp)6768 static ext4_grpblk_t ext4_last_grp_cluster(struct super_block *sb,
6769 					   ext4_group_t grp)
6770 {
6771 	unsigned long nr_clusters_in_group;
6772 
6773 	if (grp < (ext4_get_groups_count(sb) - 1))
6774 		nr_clusters_in_group = EXT4_CLUSTERS_PER_GROUP(sb);
6775 	else
6776 		nr_clusters_in_group = (ext4_blocks_count(EXT4_SB(sb)->s_es) -
6777 					ext4_group_first_block_no(sb, grp))
6778 				       >> EXT4_CLUSTER_BITS(sb);
6779 
6780 	return nr_clusters_in_group - 1;
6781 }
6782 
ext4_trim_interrupted(void)6783 static bool ext4_trim_interrupted(void)
6784 {
6785 	return fatal_signal_pending(current) || freezing(current);
6786 }
6787 
ext4_try_to_trim_range(struct super_block * sb,struct ext4_buddy * e4b,ext4_grpblk_t start,ext4_grpblk_t max,ext4_grpblk_t minblocks)6788 static int ext4_try_to_trim_range(struct super_block *sb,
6789 		struct ext4_buddy *e4b, ext4_grpblk_t start,
6790 		ext4_grpblk_t max, ext4_grpblk_t minblocks)
6791 __acquires(ext4_group_lock_ptr(sb, e4b->bd_group))
6792 __releases(ext4_group_lock_ptr(sb, e4b->bd_group))
6793 {
6794 	ext4_grpblk_t next, count, free_count, last, origin_start;
6795 	bool set_trimmed = false;
6796 	void *bitmap;
6797 
6798 	if (unlikely(EXT4_MB_GRP_BBITMAP_CORRUPT(e4b->bd_info)))
6799 		return 0;
6800 
6801 	last = ext4_last_grp_cluster(sb, e4b->bd_group);
6802 	bitmap = e4b->bd_bitmap;
6803 	if (start == 0 && max >= last)
6804 		set_trimmed = true;
6805 	origin_start = start;
6806 	start = max(e4b->bd_info->bb_first_free, start);
6807 	count = 0;
6808 	free_count = 0;
6809 
6810 	while (start <= max) {
6811 		start = mb_find_next_zero_bit(bitmap, max + 1, start);
6812 		if (start > max)
6813 			break;
6814 
6815 		next = mb_find_next_bit(bitmap, last + 1, start);
6816 		if (origin_start == 0 && next >= last)
6817 			set_trimmed = true;
6818 
6819 		if ((next - start) >= minblocks) {
6820 			int ret = ext4_trim_extent(sb, start, next - start, e4b);
6821 
6822 			if (ret && ret != -EOPNOTSUPP)
6823 				return count;
6824 			count += next - start;
6825 		}
6826 		free_count += next - start;
6827 		start = next + 1;
6828 
6829 		if (ext4_trim_interrupted())
6830 			return count;
6831 
6832 		if (need_resched()) {
6833 			ext4_unlock_group(sb, e4b->bd_group);
6834 			cond_resched();
6835 			ext4_lock_group(sb, e4b->bd_group);
6836 		}
6837 
6838 		if ((e4b->bd_info->bb_free - free_count) < minblocks)
6839 			break;
6840 	}
6841 
6842 	if (set_trimmed)
6843 		EXT4_MB_GRP_SET_TRIMMED(e4b->bd_info);
6844 
6845 	return count;
6846 }
6847 
6848 /**
6849  * ext4_trim_all_free -- function to trim all free space in alloc. group
6850  * @sb:			super block for file system
6851  * @group:		group to be trimmed
6852  * @start:		first group block to examine
6853  * @max:		last group block to examine
6854  * @minblocks:		minimum extent block count
6855  *
6856  * ext4_trim_all_free walks through group's block bitmap searching for free
6857  * extents. When the free extent is found, mark it as used in group buddy
6858  * bitmap. Then issue a TRIM command on this extent and free the extent in
6859  * the group buddy bitmap.
6860  */
6861 static ext4_grpblk_t
ext4_trim_all_free(struct super_block * sb,ext4_group_t group,ext4_grpblk_t start,ext4_grpblk_t max,ext4_grpblk_t minblocks)6862 ext4_trim_all_free(struct super_block *sb, ext4_group_t group,
6863 		   ext4_grpblk_t start, ext4_grpblk_t max,
6864 		   ext4_grpblk_t minblocks)
6865 {
6866 	struct ext4_buddy e4b;
6867 	int ret;
6868 
6869 	trace_ext4_trim_all_free(sb, group, start, max);
6870 
6871 	ret = ext4_mb_load_buddy(sb, group, &e4b);
6872 	if (ret) {
6873 		ext4_warning(sb, "Error %d loading buddy information for %u",
6874 			     ret, group);
6875 		return ret;
6876 	}
6877 
6878 	ext4_lock_group(sb, group);
6879 
6880 	if (!EXT4_MB_GRP_WAS_TRIMMED(e4b.bd_info) ||
6881 	    minblocks < EXT4_SB(sb)->s_last_trim_minblks)
6882 		ret = ext4_try_to_trim_range(sb, &e4b, start, max, minblocks);
6883 	else
6884 		ret = 0;
6885 
6886 	ext4_unlock_group(sb, group);
6887 	ext4_mb_unload_buddy(&e4b);
6888 
6889 	ext4_debug("trimmed %d blocks in the group %d\n",
6890 		ret, group);
6891 
6892 	return ret;
6893 }
6894 
6895 /**
6896  * ext4_trim_fs() -- trim ioctl handle function
6897  * @sb:			superblock for filesystem
6898  * @range:		fstrim_range structure
6899  *
6900  * start:	First Byte to trim
6901  * len:		number of Bytes to trim from start
6902  * minlen:	minimum extent length in Bytes
6903  * ext4_trim_fs goes through all allocation groups containing Bytes from
6904  * start to start+len. For each such a group ext4_trim_all_free function
6905  * is invoked to trim all free space.
6906  */
ext4_trim_fs(struct super_block * sb,struct fstrim_range * range)6907 int ext4_trim_fs(struct super_block *sb, struct fstrim_range *range)
6908 {
6909 	unsigned int discard_granularity = bdev_discard_granularity(sb->s_bdev);
6910 	struct ext4_group_info *grp;
6911 	ext4_group_t group, first_group, last_group;
6912 	ext4_grpblk_t cnt = 0, first_cluster, last_cluster;
6913 	uint64_t start, end, minlen, trimmed = 0;
6914 	ext4_fsblk_t first_data_blk =
6915 			le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block);
6916 	ext4_fsblk_t max_blks = ext4_blocks_count(EXT4_SB(sb)->s_es);
6917 	int ret = 0;
6918 
6919 	start = range->start >> sb->s_blocksize_bits;
6920 	end = start + (range->len >> sb->s_blocksize_bits) - 1;
6921 	minlen = EXT4_NUM_B2C(EXT4_SB(sb),
6922 			      range->minlen >> sb->s_blocksize_bits);
6923 
6924 	if (minlen > EXT4_CLUSTERS_PER_GROUP(sb) ||
6925 	    start >= max_blks ||
6926 	    range->len < sb->s_blocksize)
6927 		return -EINVAL;
6928 	/* No point to try to trim less than discard granularity */
6929 	if (range->minlen < discard_granularity) {
6930 		minlen = EXT4_NUM_B2C(EXT4_SB(sb),
6931 				discard_granularity >> sb->s_blocksize_bits);
6932 		if (minlen > EXT4_CLUSTERS_PER_GROUP(sb))
6933 			goto out;
6934 	}
6935 	if (end >= max_blks - 1)
6936 		end = max_blks - 1;
6937 	if (end <= first_data_blk)
6938 		goto out;
6939 	if (start < first_data_blk)
6940 		start = first_data_blk;
6941 
6942 	/* Determine first and last group to examine based on start and end */
6943 	ext4_get_group_no_and_offset(sb, (ext4_fsblk_t) start,
6944 				     &first_group, &first_cluster);
6945 	ext4_get_group_no_and_offset(sb, (ext4_fsblk_t) end,
6946 				     &last_group, &last_cluster);
6947 
6948 	/* end now represents the last cluster to discard in this group */
6949 	end = EXT4_CLUSTERS_PER_GROUP(sb) - 1;
6950 
6951 	for (group = first_group; group <= last_group; group++) {
6952 		if (ext4_trim_interrupted())
6953 			break;
6954 		grp = ext4_get_group_info(sb, group);
6955 		if (!grp)
6956 			continue;
6957 		/* We only do this if the grp has never been initialized */
6958 		if (unlikely(EXT4_MB_GRP_NEED_INIT(grp))) {
6959 			ret = ext4_mb_init_group(sb, group, GFP_NOFS);
6960 			if (ret)
6961 				break;
6962 		}
6963 
6964 		/*
6965 		 * For all the groups except the last one, last cluster will
6966 		 * always be EXT4_CLUSTERS_PER_GROUP(sb)-1, so we only need to
6967 		 * change it for the last group, note that last_cluster is
6968 		 * already computed earlier by ext4_get_group_no_and_offset()
6969 		 */
6970 		if (group == last_group)
6971 			end = last_cluster;
6972 		if (grp->bb_free >= minlen) {
6973 			cnt = ext4_trim_all_free(sb, group, first_cluster,
6974 						 end, minlen);
6975 			if (cnt < 0) {
6976 				ret = cnt;
6977 				break;
6978 			}
6979 			trimmed += cnt;
6980 		}
6981 
6982 		/*
6983 		 * For every group except the first one, we are sure
6984 		 * that the first cluster to discard will be cluster #0.
6985 		 */
6986 		first_cluster = 0;
6987 	}
6988 
6989 	if (!ret)
6990 		EXT4_SB(sb)->s_last_trim_minblks = minlen;
6991 
6992 out:
6993 	range->len = EXT4_C2B(EXT4_SB(sb), trimmed) << sb->s_blocksize_bits;
6994 	return ret;
6995 }
6996 
6997 /* Iterate all the free extents in the group. */
6998 int
ext4_mballoc_query_range(struct super_block * sb,ext4_group_t group,ext4_grpblk_t start,ext4_grpblk_t end,ext4_mballoc_query_range_fn formatter,void * priv)6999 ext4_mballoc_query_range(
7000 	struct super_block		*sb,
7001 	ext4_group_t			group,
7002 	ext4_grpblk_t			start,
7003 	ext4_grpblk_t			end,
7004 	ext4_mballoc_query_range_fn	formatter,
7005 	void				*priv)
7006 {
7007 	void				*bitmap;
7008 	ext4_grpblk_t			next;
7009 	struct ext4_buddy		e4b;
7010 	int				error;
7011 
7012 	error = ext4_mb_load_buddy(sb, group, &e4b);
7013 	if (error)
7014 		return error;
7015 	bitmap = e4b.bd_bitmap;
7016 
7017 	ext4_lock_group(sb, group);
7018 
7019 	start = max(e4b.bd_info->bb_first_free, start);
7020 	if (end >= EXT4_CLUSTERS_PER_GROUP(sb))
7021 		end = EXT4_CLUSTERS_PER_GROUP(sb) - 1;
7022 
7023 	while (start <= end) {
7024 		start = mb_find_next_zero_bit(bitmap, end + 1, start);
7025 		if (start > end)
7026 			break;
7027 		next = mb_find_next_bit(bitmap, end + 1, start);
7028 
7029 		ext4_unlock_group(sb, group);
7030 		error = formatter(sb, group, start, next - start, priv);
7031 		if (error)
7032 			goto out_unload;
7033 		ext4_lock_group(sb, group);
7034 
7035 		start = next + 1;
7036 	}
7037 
7038 	ext4_unlock_group(sb, group);
7039 out_unload:
7040 	ext4_mb_unload_buddy(&e4b);
7041 
7042 	return error;
7043 }
7044 
7045 #ifdef CONFIG_EXT4_KUNIT_TESTS
7046 #include "mballoc-test.c"
7047 #endif
7048