aboutsummaryrefslogtreecommitdiffstats
path: root/fs/ubifs/budget.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/ubifs/budget.c')
-rw-r--r--fs/ubifs/budget.c208
1 files changed, 69 insertions, 139 deletions
diff --git a/fs/ubifs/budget.c b/fs/ubifs/budget.c
index 4a18f084cc42..0e5e54d82924 100644
--- a/fs/ubifs/budget.c
+++ b/fs/ubifs/budget.c
@@ -32,18 +32,15 @@
32 32
33#include "ubifs.h" 33#include "ubifs.h"
34#include <linux/writeback.h> 34#include <linux/writeback.h>
35#include <asm/div64.h> 35#include <linux/math64.h>
36 36
37/* 37/*
38 * When pessimistic budget calculations say that there is no enough space, 38 * When pessimistic budget calculations say that there is no enough space,
39 * UBIFS starts writing back dirty inodes and pages, doing garbage collection, 39 * UBIFS starts writing back dirty inodes and pages, doing garbage collection,
40 * or committing. The below constants define maximum number of times UBIFS 40 * or committing. The below constant defines maximum number of times UBIFS
41 * repeats the operations. 41 * repeats the operations.
42 */ 42 */
43#define MAX_SHRINK_RETRIES 8 43#define MAX_MKSPC_RETRIES 3
44#define MAX_GC_RETRIES 4
45#define MAX_CMT_RETRIES 2
46#define MAX_NOSPC_RETRIES 1
47 44
48/* 45/*
49 * The below constant defines amount of dirty pages which should be written 46 * The below constant defines amount of dirty pages which should be written
@@ -52,30 +49,6 @@
52#define NR_TO_WRITE 16 49#define NR_TO_WRITE 16
53 50
54/** 51/**
55 * struct retries_info - information about re-tries while making free space.
56 * @prev_liability: previous liability
57 * @shrink_cnt: how many times the liability was shrinked
58 * @shrink_retries: count of liability shrink re-tries (increased when
59 * liability does not shrink)
60 * @try_gc: GC should be tried first
61 * @gc_retries: how many times GC was run
62 * @cmt_retries: how many times commit has been done
63 * @nospc_retries: how many times GC returned %-ENOSPC
64 *
65 * Since we consider budgeting to be the fast-path, and this structure has to
66 * be allocated on stack and zeroed out, we make it smaller using bit-fields.
67 */
68struct retries_info {
69 long long prev_liability;
70 unsigned int shrink_cnt;
71 unsigned int shrink_retries:5;
72 unsigned int try_gc:1;
73 unsigned int gc_retries:4;
74 unsigned int cmt_retries:3;
75 unsigned int nospc_retries:1;
76};
77
78/**
79 * shrink_liability - write-back some dirty pages/inodes. 52 * shrink_liability - write-back some dirty pages/inodes.
80 * @c: UBIFS file-system description object 53 * @c: UBIFS file-system description object
81 * @nr_to_write: how many dirty pages to write-back 54 * @nr_to_write: how many dirty pages to write-back
@@ -147,9 +120,25 @@ static int run_gc(struct ubifs_info *c)
147} 120}
148 121
149/** 122/**
123 * get_liability - calculate current liability.
124 * @c: UBIFS file-system description object
125 *
126 * This function calculates and returns current UBIFS liability, i.e. the
127 * amount of bytes UBIFS has "promised" to write to the media.
128 */
129static long long get_liability(struct ubifs_info *c)
130{
131 long long liab;
132
133 spin_lock(&c->space_lock);
134 liab = c->budg_idx_growth + c->budg_data_growth + c->budg_dd_growth;
135 spin_unlock(&c->space_lock);
136 return liab;
137}
138
139/**
150 * make_free_space - make more free space on the file-system. 140 * make_free_space - make more free space on the file-system.
151 * @c: UBIFS file-system description object 141 * @c: UBIFS file-system description object
152 * @ri: information about previous invocations of this function
153 * 142 *
154 * This function is called when an operation cannot be budgeted because there 143 * This function is called when an operation cannot be budgeted because there
155 * is supposedly no free space. But in most cases there is some free space: 144 * is supposedly no free space. But in most cases there is some free space:
@@ -165,87 +154,42 @@ static int run_gc(struct ubifs_info *c)
165 * Returns %-ENOSPC if it couldn't do more free space, and other negative error 154 * Returns %-ENOSPC if it couldn't do more free space, and other negative error
166 * codes on failures. 155 * codes on failures.
167 */ 156 */
168static int make_free_space(struct ubifs_info *c, struct retries_info *ri) 157static int make_free_space(struct ubifs_info *c)
169{ 158{
170 int err; 159 int err, retries = 0;
171 160 long long liab1, liab2;
172 /*
173 * If we have some dirty pages and inodes (liability), try to write
174 * them back unless this was tried too many times without effect
175 * already.
176 */
177 if (ri->shrink_retries < MAX_SHRINK_RETRIES && !ri->try_gc) {
178 long long liability;
179
180 spin_lock(&c->space_lock);
181 liability = c->budg_idx_growth + c->budg_data_growth +
182 c->budg_dd_growth;
183 spin_unlock(&c->space_lock);
184 161
185 if (ri->prev_liability >= liability) { 162 do {
186 /* Liability does not shrink, next time try GC then */ 163 liab1 = get_liability(c);
187 ri->shrink_retries += 1; 164 /*
188 if (ri->gc_retries < MAX_GC_RETRIES) 165 * We probably have some dirty pages or inodes (liability), try
189 ri->try_gc = 1; 166 * to write them back.
190 dbg_budg("liability did not shrink: retries %d of %d", 167 */
191 ri->shrink_retries, MAX_SHRINK_RETRIES); 168 dbg_budg("liability %lld, run write-back", liab1);
192 } 169 shrink_liability(c, NR_TO_WRITE);
193 170
194 dbg_budg("force write-back (count %d)", ri->shrink_cnt); 171 liab2 = get_liability(c);
195 shrink_liability(c, NR_TO_WRITE + ri->shrink_cnt); 172 if (liab2 < liab1)
173 return -EAGAIN;
196 174
197 ri->prev_liability = liability; 175 dbg_budg("new liability %lld (not shrinked)", liab2);
198 ri->shrink_cnt += 1;
199 return -EAGAIN;
200 }
201 176
202 /* 177 /* Liability did not shrink again, try GC */
203 * Try to run garbage collector unless it was already tried too many 178 dbg_budg("Run GC");
204 * times.
205 */
206 if (ri->gc_retries < MAX_GC_RETRIES) {
207 ri->gc_retries += 1;
208 dbg_budg("run GC, retries %d of %d",
209 ri->gc_retries, MAX_GC_RETRIES);
210
211 ri->try_gc = 0;
212 err = run_gc(c); 179 err = run_gc(c);
213 if (!err) 180 if (!err)
214 return -EAGAIN; 181 return -EAGAIN;
215 182
216 if (err == -EAGAIN) { 183 if (err != -EAGAIN && err != -ENOSPC)
217 dbg_budg("GC asked to commit"); 184 /* Some real error happened */
218 err = ubifs_run_commit(c);
219 if (err)
220 return err;
221 return -EAGAIN;
222 }
223
224 if (err != -ENOSPC)
225 return err;
226
227 /*
228 * GC could not make any progress. If this is the first time,
229 * then it makes sense to try to commit, because it might make
230 * some dirty space.
231 */
232 dbg_budg("GC returned -ENOSPC, retries %d",
233 ri->nospc_retries);
234 if (ri->nospc_retries >= MAX_NOSPC_RETRIES)
235 return err; 185 return err;
236 ri->nospc_retries += 1;
237 }
238 186
239 /* Neither GC nor write-back helped, try to commit */ 187 dbg_budg("Run commit (retries %d)", retries);
240 if (ri->cmt_retries < MAX_CMT_RETRIES) {
241 ri->cmt_retries += 1;
242 dbg_budg("run commit, retries %d of %d",
243 ri->cmt_retries, MAX_CMT_RETRIES);
244 err = ubifs_run_commit(c); 188 err = ubifs_run_commit(c);
245 if (err) 189 if (err)
246 return err; 190 return err;
247 return -EAGAIN; 191 } while (retries++ < MAX_MKSPC_RETRIES);
248 } 192
249 return -ENOSPC; 193 return -ENOSPC;
250} 194}
251 195
@@ -258,8 +202,8 @@ static int make_free_space(struct ubifs_info *c, struct retries_info *ri)
258 */ 202 */
259int ubifs_calc_min_idx_lebs(struct ubifs_info *c) 203int ubifs_calc_min_idx_lebs(struct ubifs_info *c)
260{ 204{
261 int ret; 205 int idx_lebs, eff_leb_size = c->leb_size - c->max_idx_node_sz;
262 uint64_t idx_size; 206 long long idx_size;
263 207
264 idx_size = c->old_idx_sz + c->budg_idx_growth + c->budg_uncommitted_idx; 208 idx_size = c->old_idx_sz + c->budg_idx_growth + c->budg_uncommitted_idx;
265 209
@@ -271,23 +215,16 @@ int ubifs_calc_min_idx_lebs(struct ubifs_info *c)
271 * pair, nor similarly the two variables for the new index size, so we 215 * pair, nor similarly the two variables for the new index size, so we
272 * have to do this costly 64-bit division on fast-path. 216 * have to do this costly 64-bit division on fast-path.
273 */ 217 */
274 if (do_div(idx_size, c->leb_size - c->max_idx_node_sz)) 218 idx_size += eff_leb_size - 1;
275 ret = idx_size + 1; 219 idx_lebs = div_u64(idx_size, eff_leb_size);
276 else
277 ret = idx_size;
278 /* 220 /*
279 * The index head is not available for the in-the-gaps method, so add an 221 * The index head is not available for the in-the-gaps method, so add an
280 * extra LEB to compensate. 222 * extra LEB to compensate.
281 */ 223 */
282 ret += 1; 224 idx_lebs += 1;
283 /* 225 if (idx_lebs < MIN_INDEX_LEBS)
284 * At present the index needs at least 2 LEBs: one for the index head 226 idx_lebs = MIN_INDEX_LEBS;
285 * and one for in-the-gaps method (which currently does not cater for 227 return idx_lebs;
286 * the index head and so excludes it from consideration).
287 */
288 if (ret < 2)
289 ret = 2;
290 return ret;
291} 228}
292 229
293/** 230/**
@@ -530,8 +467,7 @@ static int calc_dd_growth(const struct ubifs_info *c,
530int ubifs_budget_space(struct ubifs_info *c, struct ubifs_budget_req *req) 467int ubifs_budget_space(struct ubifs_info *c, struct ubifs_budget_req *req)
531{ 468{
532 int uninitialized_var(cmt_retries), uninitialized_var(wb_retries); 469 int uninitialized_var(cmt_retries), uninitialized_var(wb_retries);
533 int err, idx_growth, data_growth, dd_growth; 470 int err, idx_growth, data_growth, dd_growth, retried = 0;
534 struct retries_info ri;
535 471
536 ubifs_assert(req->new_page <= 1); 472 ubifs_assert(req->new_page <= 1);
537 ubifs_assert(req->dirtied_page <= 1); 473 ubifs_assert(req->dirtied_page <= 1);
@@ -549,7 +485,6 @@ int ubifs_budget_space(struct ubifs_info *c, struct ubifs_budget_req *req)
549 if (!data_growth && !dd_growth) 485 if (!data_growth && !dd_growth)
550 return 0; 486 return 0;
551 idx_growth = calc_idx_growth(c, req); 487 idx_growth = calc_idx_growth(c, req);
552 memset(&ri, 0, sizeof(struct retries_info));
553 488
554again: 489again:
555 spin_lock(&c->space_lock); 490 spin_lock(&c->space_lock);
@@ -587,12 +522,17 @@ again:
587 return err; 522 return err;
588 } 523 }
589 524
590 err = make_free_space(c, &ri); 525 err = make_free_space(c);
526 cond_resched();
591 if (err == -EAGAIN) { 527 if (err == -EAGAIN) {
592 dbg_budg("try again"); 528 dbg_budg("try again");
593 cond_resched();
594 goto again; 529 goto again;
595 } else if (err == -ENOSPC) { 530 } else if (err == -ENOSPC) {
531 if (!retried) {
532 retried = 1;
533 dbg_budg("-ENOSPC, but anyway try once again");
534 goto again;
535 }
596 dbg_budg("FS is full, -ENOSPC"); 536 dbg_budg("FS is full, -ENOSPC");
597 c->nospace = 1; 537 c->nospace = 1;
598 if (can_use_rp(c) || c->rp_size == 0) 538 if (can_use_rp(c) || c->rp_size == 0)
@@ -712,9 +652,9 @@ void ubifs_release_dirty_inode_budget(struct ubifs_info *c,
712 * user-space. User-space application tend to expect that if the file-system 652 * user-space. User-space application tend to expect that if the file-system
713 * (e.g., via the 'statfs()' call) reports that it has N bytes available, they 653 * (e.g., via the 'statfs()' call) reports that it has N bytes available, they
714 * are able to write a file of size N. UBIFS attaches node headers to each data 654 * are able to write a file of size N. UBIFS attaches node headers to each data
715 * node and it has to write indexind nodes as well. This introduces additional 655 * node and it has to write indexing nodes as well. This introduces additional
716 * overhead, and UBIFS it has to report sligtly less free space to meet the 656 * overhead, and UBIFS has to report slightly less free space to meet the above
717 * above expectetion. 657 * expectations.
718 * 658 *
719 * This function assumes free space is made up of uncompressed data nodes and 659 * This function assumes free space is made up of uncompressed data nodes and
720 * full index nodes (one per data node, tripled because we always allow enough 660 * full index nodes (one per data node, tripled because we always allow enough
@@ -723,7 +663,7 @@ void ubifs_release_dirty_inode_budget(struct ubifs_info *c,
723 * Note, the calculation is pessimistic, which means that most of the time 663 * Note, the calculation is pessimistic, which means that most of the time
724 * UBIFS reports less space than it actually has. 664 * UBIFS reports less space than it actually has.
725 */ 665 */
726long long ubifs_reported_space(const struct ubifs_info *c, uint64_t free) 666long long ubifs_reported_space(const struct ubifs_info *c, long long free)
727{ 667{
728 int divisor, factor, f; 668 int divisor, factor, f;
729 669
@@ -737,7 +677,7 @@ long long ubifs_reported_space(const struct ubifs_info *c, uint64_t free)
737 * of data nodes, f - fanout. Because effective UBIFS fanout is twice 677 * of data nodes, f - fanout. Because effective UBIFS fanout is twice
738 * as less than maximum fanout, we assume that each data node 678 * as less than maximum fanout, we assume that each data node
739 * introduces 3 * @c->max_idx_node_sz / (@c->fanout/2 - 1) bytes. 679 * introduces 3 * @c->max_idx_node_sz / (@c->fanout/2 - 1) bytes.
740 * Note, the multiplier 3 is because UBIFS reseves thrice as more space 680 * Note, the multiplier 3 is because UBIFS reserves thrice as more space
741 * for the index. 681 * for the index.
742 */ 682 */
743 f = c->fanout > 3 ? c->fanout >> 1 : 2; 683 f = c->fanout > 3 ? c->fanout >> 1 : 2;
@@ -745,8 +685,7 @@ long long ubifs_reported_space(const struct ubifs_info *c, uint64_t free)
745 divisor = UBIFS_MAX_DATA_NODE_SZ; 685 divisor = UBIFS_MAX_DATA_NODE_SZ;
746 divisor += (c->max_idx_node_sz * 3) / (f - 1); 686 divisor += (c->max_idx_node_sz * 3) / (f - 1);
747 free *= factor; 687 free *= factor;
748 do_div(free, divisor); 688 return div_u64(free, divisor);
749 return free;
750} 689}
751 690
752/** 691/**
@@ -756,10 +695,10 @@ long long ubifs_reported_space(const struct ubifs_info *c, uint64_t free)
756 * This function calculates amount of free space to report to user-space. 695 * This function calculates amount of free space to report to user-space.
757 * 696 *
758 * Because UBIFS may introduce substantial overhead (the index, node headers, 697 * Because UBIFS may introduce substantial overhead (the index, node headers,
759 * alighment, wastage at the end of eraseblocks, etc), it cannot report real 698 * alignment, wastage at the end of eraseblocks, etc), it cannot report real
760 * amount of free flash space it has (well, because not all dirty space is 699 * amount of free flash space it has (well, because not all dirty space is
761 * reclamable, UBIFS does not actually know the real amount). If UBIFS did so, 700 * reclaimable, UBIFS does not actually know the real amount). If UBIFS did so,
762 * it would bread user expectetion about what free space is. Users seem to 701 * it would bread user expectations about what free space is. Users seem to
763 * accustomed to assume that if the file-system reports N bytes of free space, 702 * accustomed to assume that if the file-system reports N bytes of free space,
764 * they would be able to fit a file of N bytes to the FS. This almost works for 703 * they would be able to fit a file of N bytes to the FS. This almost works for
765 * traditional file-systems, because they have way less overhead than UBIFS. 704 * traditional file-systems, because they have way less overhead than UBIFS.
@@ -771,18 +710,9 @@ long long ubifs_get_free_space(struct ubifs_info *c)
771 long long available, outstanding, free; 710 long long available, outstanding, free;
772 711
773 spin_lock(&c->space_lock); 712 spin_lock(&c->space_lock);
774 min_idx_lebs = ubifs_calc_min_idx_lebs(c); 713 min_idx_lebs = c->min_idx_lebs;
714 ubifs_assert(min_idx_lebs == ubifs_calc_min_idx_lebs(c));
775 outstanding = c->budg_data_growth + c->budg_dd_growth; 715 outstanding = c->budg_data_growth + c->budg_dd_growth;
776
777 /*
778 * Force the amount available to the total size reported if the used
779 * space is zero.
780 */
781 if (c->lst.total_used <= UBIFS_INO_NODE_SZ && !outstanding) {
782 spin_unlock(&c->space_lock);
783 return (long long)c->block_cnt << UBIFS_BLOCK_SHIFT;
784 }
785
786 available = ubifs_calc_available(c, min_idx_lebs); 716 available = ubifs_calc_available(c, min_idx_lebs);
787 717
788 /* 718 /*