]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
1da177e4 | 20 | #include "xfs_types.h" |
a844f451 | 21 | #include "xfs_bit.h" |
1da177e4 | 22 | #include "xfs_log.h" |
a844f451 | 23 | #include "xfs_inum.h" |
1da177e4 LT |
24 | #include "xfs_trans.h" |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_ag.h" | |
a844f451 | 27 | #include "xfs_dir2.h" |
1da177e4 LT |
28 | #include "xfs_dmapi.h" |
29 | #include "xfs_mount.h" | |
1da177e4 | 30 | #include "xfs_bmap_btree.h" |
a844f451 | 31 | #include "xfs_alloc_btree.h" |
1da177e4 | 32 | #include "xfs_ialloc_btree.h" |
a844f451 NS |
33 | #include "xfs_dir2_sf.h" |
34 | #include "xfs_attr_sf.h" | |
35 | #include "xfs_dinode.h" | |
36 | #include "xfs_inode.h" | |
1da177e4 LT |
37 | #include "xfs_btree.h" |
38 | #include "xfs_ialloc.h" | |
39 | #include "xfs_alloc.h" | |
1da177e4 LT |
40 | #include "xfs_error.h" |
41 | ||
42 | ||
43 | #define XFS_ABSDIFF(a,b) (((a) <= (b)) ? ((b) - (a)) : ((a) - (b))) | |
44 | ||
45 | #define XFSA_FIXUP_BNO_OK 1 | |
46 | #define XFSA_FIXUP_CNT_OK 2 | |
47 | ||
f4586e40 | 48 | STATIC void |
1da177e4 LT |
49 | xfs_alloc_search_busy(xfs_trans_t *tp, |
50 | xfs_agnumber_t agno, | |
51 | xfs_agblock_t bno, | |
52 | xfs_extlen_t len); | |
53 | ||
54 | #if defined(XFS_ALLOC_TRACE) | |
55 | ktrace_t *xfs_alloc_trace_buf; | |
56 | ||
57 | #define TRACE_ALLOC(s,a) \ | |
34a622b2 | 58 | xfs_alloc_trace_alloc(__func__, s, a, __LINE__) |
1da177e4 | 59 | #define TRACE_FREE(s,a,b,x,f) \ |
34a622b2 | 60 | xfs_alloc_trace_free(__func__, s, mp, a, b, x, f, __LINE__) |
1da177e4 | 61 | #define TRACE_MODAGF(s,a,f) \ |
34a622b2 HH |
62 | xfs_alloc_trace_modagf(__func__, s, mp, a, f, __LINE__) |
63 | #define TRACE_BUSY(__func__,s,ag,agb,l,sl,tp) \ | |
64 | xfs_alloc_trace_busy(__func__, s, mp, ag, agb, l, sl, tp, XFS_ALLOC_KTRACE_BUSY, __LINE__) | |
65 | #define TRACE_UNBUSY(__func__,s,ag,sl,tp) \ | |
66 | xfs_alloc_trace_busy(__func__, s, mp, ag, -1, -1, sl, tp, XFS_ALLOC_KTRACE_UNBUSY, __LINE__) | |
f4586e40 DC |
67 | #define TRACE_BUSYSEARCH(__func__,s,ag,agb,l,tp) \ |
68 | xfs_alloc_trace_busy(__func__, s, mp, ag, agb, l, 0, tp, XFS_ALLOC_KTRACE_BUSYSEARCH, __LINE__) | |
1da177e4 LT |
69 | #else |
70 | #define TRACE_ALLOC(s,a) | |
71 | #define TRACE_FREE(s,a,b,x,f) | |
72 | #define TRACE_MODAGF(s,a,f) | |
73 | #define TRACE_BUSY(s,a,ag,agb,l,sl,tp) | |
74 | #define TRACE_UNBUSY(fname,s,ag,sl,tp) | |
f4586e40 | 75 | #define TRACE_BUSYSEARCH(fname,s,ag,agb,l,tp) |
1da177e4 LT |
76 | #endif /* XFS_ALLOC_TRACE */ |
77 | ||
78 | /* | |
79 | * Prototypes for per-ag allocation routines | |
80 | */ | |
81 | ||
82 | STATIC int xfs_alloc_ag_vextent_exact(xfs_alloc_arg_t *); | |
83 | STATIC int xfs_alloc_ag_vextent_near(xfs_alloc_arg_t *); | |
84 | STATIC int xfs_alloc_ag_vextent_size(xfs_alloc_arg_t *); | |
85 | STATIC int xfs_alloc_ag_vextent_small(xfs_alloc_arg_t *, | |
86 | xfs_btree_cur_t *, xfs_agblock_t *, xfs_extlen_t *, int *); | |
87 | ||
88 | /* | |
89 | * Internal functions. | |
90 | */ | |
91 | ||
92 | /* | |
93 | * Compute aligned version of the found extent. | |
94 | * Takes alignment and min length into account. | |
95 | */ | |
12375c82 | 96 | STATIC void |
1da177e4 LT |
97 | xfs_alloc_compute_aligned( |
98 | xfs_agblock_t foundbno, /* starting block in found extent */ | |
99 | xfs_extlen_t foundlen, /* length in found extent */ | |
100 | xfs_extlen_t alignment, /* alignment for allocation */ | |
101 | xfs_extlen_t minlen, /* minimum length for allocation */ | |
102 | xfs_agblock_t *resbno, /* result block number */ | |
103 | xfs_extlen_t *reslen) /* result length */ | |
104 | { | |
105 | xfs_agblock_t bno; | |
106 | xfs_extlen_t diff; | |
107 | xfs_extlen_t len; | |
108 | ||
109 | if (alignment > 1 && foundlen >= minlen) { | |
110 | bno = roundup(foundbno, alignment); | |
111 | diff = bno - foundbno; | |
112 | len = diff >= foundlen ? 0 : foundlen - diff; | |
113 | } else { | |
114 | bno = foundbno; | |
115 | len = foundlen; | |
116 | } | |
117 | *resbno = bno; | |
118 | *reslen = len; | |
1da177e4 LT |
119 | } |
120 | ||
121 | /* | |
122 | * Compute best start block and diff for "near" allocations. | |
123 | * freelen >= wantlen already checked by caller. | |
124 | */ | |
125 | STATIC xfs_extlen_t /* difference value (absolute) */ | |
126 | xfs_alloc_compute_diff( | |
127 | xfs_agblock_t wantbno, /* target starting block */ | |
128 | xfs_extlen_t wantlen, /* target length */ | |
129 | xfs_extlen_t alignment, /* target alignment */ | |
130 | xfs_agblock_t freebno, /* freespace's starting block */ | |
131 | xfs_extlen_t freelen, /* freespace's length */ | |
132 | xfs_agblock_t *newbnop) /* result: best start block from free */ | |
133 | { | |
134 | xfs_agblock_t freeend; /* end of freespace extent */ | |
135 | xfs_agblock_t newbno1; /* return block number */ | |
136 | xfs_agblock_t newbno2; /* other new block number */ | |
137 | xfs_extlen_t newlen1=0; /* length with newbno1 */ | |
138 | xfs_extlen_t newlen2=0; /* length with newbno2 */ | |
139 | xfs_agblock_t wantend; /* end of target extent */ | |
140 | ||
141 | ASSERT(freelen >= wantlen); | |
142 | freeend = freebno + freelen; | |
143 | wantend = wantbno + wantlen; | |
144 | if (freebno >= wantbno) { | |
145 | if ((newbno1 = roundup(freebno, alignment)) >= freeend) | |
146 | newbno1 = NULLAGBLOCK; | |
147 | } else if (freeend >= wantend && alignment > 1) { | |
148 | newbno1 = roundup(wantbno, alignment); | |
149 | newbno2 = newbno1 - alignment; | |
150 | if (newbno1 >= freeend) | |
151 | newbno1 = NULLAGBLOCK; | |
152 | else | |
153 | newlen1 = XFS_EXTLEN_MIN(wantlen, freeend - newbno1); | |
154 | if (newbno2 < freebno) | |
155 | newbno2 = NULLAGBLOCK; | |
156 | else | |
157 | newlen2 = XFS_EXTLEN_MIN(wantlen, freeend - newbno2); | |
158 | if (newbno1 != NULLAGBLOCK && newbno2 != NULLAGBLOCK) { | |
159 | if (newlen1 < newlen2 || | |
160 | (newlen1 == newlen2 && | |
161 | XFS_ABSDIFF(newbno1, wantbno) > | |
162 | XFS_ABSDIFF(newbno2, wantbno))) | |
163 | newbno1 = newbno2; | |
164 | } else if (newbno2 != NULLAGBLOCK) | |
165 | newbno1 = newbno2; | |
166 | } else if (freeend >= wantend) { | |
167 | newbno1 = wantbno; | |
168 | } else if (alignment > 1) { | |
169 | newbno1 = roundup(freeend - wantlen, alignment); | |
170 | if (newbno1 > freeend - wantlen && | |
171 | newbno1 - alignment >= freebno) | |
172 | newbno1 -= alignment; | |
173 | else if (newbno1 >= freeend) | |
174 | newbno1 = NULLAGBLOCK; | |
175 | } else | |
176 | newbno1 = freeend - wantlen; | |
177 | *newbnop = newbno1; | |
178 | return newbno1 == NULLAGBLOCK ? 0 : XFS_ABSDIFF(newbno1, wantbno); | |
179 | } | |
180 | ||
181 | /* | |
182 | * Fix up the length, based on mod and prod. | |
183 | * len should be k * prod + mod for some k. | |
184 | * If len is too small it is returned unchanged. | |
185 | * If len hits maxlen it is left alone. | |
186 | */ | |
187 | STATIC void | |
188 | xfs_alloc_fix_len( | |
189 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
190 | { | |
191 | xfs_extlen_t k; | |
192 | xfs_extlen_t rlen; | |
193 | ||
194 | ASSERT(args->mod < args->prod); | |
195 | rlen = args->len; | |
196 | ASSERT(rlen >= args->minlen); | |
197 | ASSERT(rlen <= args->maxlen); | |
198 | if (args->prod <= 1 || rlen < args->mod || rlen == args->maxlen || | |
199 | (args->mod == 0 && rlen < args->prod)) | |
200 | return; | |
201 | k = rlen % args->prod; | |
202 | if (k == args->mod) | |
203 | return; | |
204 | if (k > args->mod) { | |
205 | if ((int)(rlen = rlen - k - args->mod) < (int)args->minlen) | |
206 | return; | |
207 | } else { | |
208 | if ((int)(rlen = rlen - args->prod - (args->mod - k)) < | |
209 | (int)args->minlen) | |
210 | return; | |
211 | } | |
212 | ASSERT(rlen >= args->minlen); | |
213 | ASSERT(rlen <= args->maxlen); | |
214 | args->len = rlen; | |
215 | } | |
216 | ||
217 | /* | |
218 | * Fix up length if there is too little space left in the a.g. | |
219 | * Return 1 if ok, 0 if too little, should give up. | |
220 | */ | |
221 | STATIC int | |
222 | xfs_alloc_fix_minleft( | |
223 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
224 | { | |
225 | xfs_agf_t *agf; /* a.g. freelist header */ | |
226 | int diff; /* free space difference */ | |
227 | ||
228 | if (args->minleft == 0) | |
229 | return 1; | |
230 | agf = XFS_BUF_TO_AGF(args->agbp); | |
16259e7d CH |
231 | diff = be32_to_cpu(agf->agf_freeblks) |
232 | + be32_to_cpu(agf->agf_flcount) | |
1da177e4 LT |
233 | - args->len - args->minleft; |
234 | if (diff >= 0) | |
235 | return 1; | |
236 | args->len += diff; /* shrink the allocated space */ | |
237 | if (args->len >= args->minlen) | |
238 | return 1; | |
239 | args->agbno = NULLAGBLOCK; | |
240 | return 0; | |
241 | } | |
242 | ||
243 | /* | |
244 | * Update the two btrees, logically removing from freespace the extent | |
245 | * starting at rbno, rlen blocks. The extent is contained within the | |
246 | * actual (current) free extent fbno for flen blocks. | |
247 | * Flags are passed in indicating whether the cursors are set to the | |
248 | * relevant records. | |
249 | */ | |
250 | STATIC int /* error code */ | |
251 | xfs_alloc_fixup_trees( | |
252 | xfs_btree_cur_t *cnt_cur, /* cursor for by-size btree */ | |
253 | xfs_btree_cur_t *bno_cur, /* cursor for by-block btree */ | |
254 | xfs_agblock_t fbno, /* starting block of free extent */ | |
255 | xfs_extlen_t flen, /* length of free extent */ | |
256 | xfs_agblock_t rbno, /* starting block of returned extent */ | |
257 | xfs_extlen_t rlen, /* length of returned extent */ | |
258 | int flags) /* flags, XFSA_FIXUP_... */ | |
259 | { | |
260 | int error; /* error code */ | |
261 | int i; /* operation results */ | |
262 | xfs_agblock_t nfbno1; /* first new free startblock */ | |
263 | xfs_agblock_t nfbno2; /* second new free startblock */ | |
264 | xfs_extlen_t nflen1=0; /* first new free length */ | |
265 | xfs_extlen_t nflen2=0; /* second new free length */ | |
266 | ||
267 | /* | |
268 | * Look up the record in the by-size tree if necessary. | |
269 | */ | |
270 | if (flags & XFSA_FIXUP_CNT_OK) { | |
271 | #ifdef DEBUG | |
272 | if ((error = xfs_alloc_get_rec(cnt_cur, &nfbno1, &nflen1, &i))) | |
273 | return error; | |
274 | XFS_WANT_CORRUPTED_RETURN( | |
275 | i == 1 && nfbno1 == fbno && nflen1 == flen); | |
276 | #endif | |
277 | } else { | |
278 | if ((error = xfs_alloc_lookup_eq(cnt_cur, fbno, flen, &i))) | |
279 | return error; | |
280 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
281 | } | |
282 | /* | |
283 | * Look up the record in the by-block tree if necessary. | |
284 | */ | |
285 | if (flags & XFSA_FIXUP_BNO_OK) { | |
286 | #ifdef DEBUG | |
287 | if ((error = xfs_alloc_get_rec(bno_cur, &nfbno1, &nflen1, &i))) | |
288 | return error; | |
289 | XFS_WANT_CORRUPTED_RETURN( | |
290 | i == 1 && nfbno1 == fbno && nflen1 == flen); | |
291 | #endif | |
292 | } else { | |
293 | if ((error = xfs_alloc_lookup_eq(bno_cur, fbno, flen, &i))) | |
294 | return error; | |
295 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
296 | } | |
297 | #ifdef DEBUG | |
298 | { | |
299 | xfs_alloc_block_t *bnoblock; | |
300 | xfs_alloc_block_t *cntblock; | |
301 | ||
302 | if (bno_cur->bc_nlevels == 1 && | |
303 | cnt_cur->bc_nlevels == 1) { | |
304 | bnoblock = XFS_BUF_TO_ALLOC_BLOCK(bno_cur->bc_bufs[0]); | |
305 | cntblock = XFS_BUF_TO_ALLOC_BLOCK(cnt_cur->bc_bufs[0]); | |
306 | XFS_WANT_CORRUPTED_RETURN( | |
16259e7d CH |
307 | be16_to_cpu(bnoblock->bb_numrecs) == |
308 | be16_to_cpu(cntblock->bb_numrecs)); | |
1da177e4 LT |
309 | } |
310 | } | |
311 | #endif | |
312 | /* | |
313 | * Deal with all four cases: the allocated record is contained | |
314 | * within the freespace record, so we can have new freespace | |
315 | * at either (or both) end, or no freespace remaining. | |
316 | */ | |
317 | if (rbno == fbno && rlen == flen) | |
318 | nfbno1 = nfbno2 = NULLAGBLOCK; | |
319 | else if (rbno == fbno) { | |
320 | nfbno1 = rbno + rlen; | |
321 | nflen1 = flen - rlen; | |
322 | nfbno2 = NULLAGBLOCK; | |
323 | } else if (rbno + rlen == fbno + flen) { | |
324 | nfbno1 = fbno; | |
325 | nflen1 = flen - rlen; | |
326 | nfbno2 = NULLAGBLOCK; | |
327 | } else { | |
328 | nfbno1 = fbno; | |
329 | nflen1 = rbno - fbno; | |
330 | nfbno2 = rbno + rlen; | |
331 | nflen2 = (fbno + flen) - nfbno2; | |
332 | } | |
333 | /* | |
334 | * Delete the entry from the by-size btree. | |
335 | */ | |
336 | if ((error = xfs_alloc_delete(cnt_cur, &i))) | |
337 | return error; | |
338 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
339 | /* | |
340 | * Add new by-size btree entry(s). | |
341 | */ | |
342 | if (nfbno1 != NULLAGBLOCK) { | |
343 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno1, nflen1, &i))) | |
344 | return error; | |
345 | XFS_WANT_CORRUPTED_RETURN(i == 0); | |
346 | if ((error = xfs_alloc_insert(cnt_cur, &i))) | |
347 | return error; | |
348 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
349 | } | |
350 | if (nfbno2 != NULLAGBLOCK) { | |
351 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno2, nflen2, &i))) | |
352 | return error; | |
353 | XFS_WANT_CORRUPTED_RETURN(i == 0); | |
354 | if ((error = xfs_alloc_insert(cnt_cur, &i))) | |
355 | return error; | |
356 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
357 | } | |
358 | /* | |
359 | * Fix up the by-block btree entry(s). | |
360 | */ | |
361 | if (nfbno1 == NULLAGBLOCK) { | |
362 | /* | |
363 | * No remaining freespace, just delete the by-block tree entry. | |
364 | */ | |
365 | if ((error = xfs_alloc_delete(bno_cur, &i))) | |
366 | return error; | |
367 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
368 | } else { | |
369 | /* | |
370 | * Update the by-block entry to start later|be shorter. | |
371 | */ | |
372 | if ((error = xfs_alloc_update(bno_cur, nfbno1, nflen1))) | |
373 | return error; | |
374 | } | |
375 | if (nfbno2 != NULLAGBLOCK) { | |
376 | /* | |
377 | * 2 resulting free entries, need to add one. | |
378 | */ | |
379 | if ((error = xfs_alloc_lookup_eq(bno_cur, nfbno2, nflen2, &i))) | |
380 | return error; | |
381 | XFS_WANT_CORRUPTED_RETURN(i == 0); | |
382 | if ((error = xfs_alloc_insert(bno_cur, &i))) | |
383 | return error; | |
384 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
385 | } | |
386 | return 0; | |
387 | } | |
388 | ||
389 | /* | |
390 | * Read in the allocation group free block array. | |
391 | */ | |
392 | STATIC int /* error */ | |
393 | xfs_alloc_read_agfl( | |
394 | xfs_mount_t *mp, /* mount point structure */ | |
395 | xfs_trans_t *tp, /* transaction pointer */ | |
396 | xfs_agnumber_t agno, /* allocation group number */ | |
397 | xfs_buf_t **bpp) /* buffer for the ag free block array */ | |
398 | { | |
399 | xfs_buf_t *bp; /* return value */ | |
400 | int error; | |
401 | ||
402 | ASSERT(agno != NULLAGNUMBER); | |
403 | error = xfs_trans_read_buf( | |
404 | mp, tp, mp->m_ddev_targp, | |
405 | XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)), | |
406 | XFS_FSS_TO_BB(mp, 1), 0, &bp); | |
407 | if (error) | |
408 | return error; | |
409 | ASSERT(bp); | |
410 | ASSERT(!XFS_BUF_GETERROR(bp)); | |
411 | XFS_BUF_SET_VTYPE_REF(bp, B_FS_AGFL, XFS_AGFL_REF); | |
412 | *bpp = bp; | |
413 | return 0; | |
414 | } | |
415 | ||
416 | #if defined(XFS_ALLOC_TRACE) | |
417 | /* | |
418 | * Add an allocation trace entry for an alloc call. | |
419 | */ | |
420 | STATIC void | |
421 | xfs_alloc_trace_alloc( | |
3a59c94c | 422 | const char *name, /* function tag string */ |
1da177e4 LT |
423 | char *str, /* additional string */ |
424 | xfs_alloc_arg_t *args, /* allocation argument structure */ | |
425 | int line) /* source line number */ | |
426 | { | |
427 | ktrace_enter(xfs_alloc_trace_buf, | |
428 | (void *)(__psint_t)(XFS_ALLOC_KTRACE_ALLOC | (line << 16)), | |
429 | (void *)name, | |
430 | (void *)str, | |
431 | (void *)args->mp, | |
432 | (void *)(__psunsigned_t)args->agno, | |
433 | (void *)(__psunsigned_t)args->agbno, | |
434 | (void *)(__psunsigned_t)args->minlen, | |
435 | (void *)(__psunsigned_t)args->maxlen, | |
436 | (void *)(__psunsigned_t)args->mod, | |
437 | (void *)(__psunsigned_t)args->prod, | |
438 | (void *)(__psunsigned_t)args->minleft, | |
439 | (void *)(__psunsigned_t)args->total, | |
440 | (void *)(__psunsigned_t)args->alignment, | |
441 | (void *)(__psunsigned_t)args->len, | |
442 | (void *)((((__psint_t)args->type) << 16) | | |
443 | (__psint_t)args->otype), | |
444 | (void *)(__psint_t)((args->wasdel << 3) | | |
445 | (args->wasfromfl << 2) | | |
446 | (args->isfl << 1) | | |
447 | (args->userdata << 0))); | |
448 | } | |
449 | ||
450 | /* | |
451 | * Add an allocation trace entry for a free call. | |
452 | */ | |
453 | STATIC void | |
454 | xfs_alloc_trace_free( | |
3a59c94c | 455 | const char *name, /* function tag string */ |
1da177e4 LT |
456 | char *str, /* additional string */ |
457 | xfs_mount_t *mp, /* file system mount point */ | |
458 | xfs_agnumber_t agno, /* allocation group number */ | |
459 | xfs_agblock_t agbno, /* a.g. relative block number */ | |
460 | xfs_extlen_t len, /* length of extent */ | |
461 | int isfl, /* set if is freelist allocation/free */ | |
462 | int line) /* source line number */ | |
463 | { | |
464 | ktrace_enter(xfs_alloc_trace_buf, | |
465 | (void *)(__psint_t)(XFS_ALLOC_KTRACE_FREE | (line << 16)), | |
466 | (void *)name, | |
467 | (void *)str, | |
468 | (void *)mp, | |
469 | (void *)(__psunsigned_t)agno, | |
470 | (void *)(__psunsigned_t)agbno, | |
471 | (void *)(__psunsigned_t)len, | |
472 | (void *)(__psint_t)isfl, | |
473 | NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL); | |
474 | } | |
475 | ||
476 | /* | |
477 | * Add an allocation trace entry for modifying an agf. | |
478 | */ | |
479 | STATIC void | |
480 | xfs_alloc_trace_modagf( | |
3a59c94c | 481 | const char *name, /* function tag string */ |
1da177e4 LT |
482 | char *str, /* additional string */ |
483 | xfs_mount_t *mp, /* file system mount point */ | |
484 | xfs_agf_t *agf, /* new agf value */ | |
485 | int flags, /* logging flags for agf */ | |
486 | int line) /* source line number */ | |
487 | { | |
488 | ktrace_enter(xfs_alloc_trace_buf, | |
489 | (void *)(__psint_t)(XFS_ALLOC_KTRACE_MODAGF | (line << 16)), | |
490 | (void *)name, | |
491 | (void *)str, | |
492 | (void *)mp, | |
493 | (void *)(__psint_t)flags, | |
16259e7d CH |
494 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_seqno), |
495 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_length), | |
496 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_roots[XFS_BTNUM_BNO]), | |
497 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_roots[XFS_BTNUM_CNT]), | |
498 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]), | |
499 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]), | |
500 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_flfirst), | |
501 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_fllast), | |
502 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_flcount), | |
503 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_freeblks), | |
504 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_longest)); | |
1da177e4 LT |
505 | } |
506 | ||
507 | STATIC void | |
508 | xfs_alloc_trace_busy( | |
3a59c94c | 509 | const char *name, /* function tag string */ |
1da177e4 | 510 | char *str, /* additional string */ |
c41564b5 | 511 | xfs_mount_t *mp, /* file system mount point */ |
1da177e4 LT |
512 | xfs_agnumber_t agno, /* allocation group number */ |
513 | xfs_agblock_t agbno, /* a.g. relative block number */ | |
514 | xfs_extlen_t len, /* length of extent */ | |
515 | int slot, /* perag Busy slot */ | |
516 | xfs_trans_t *tp, | |
517 | int trtype, /* type: add, delete, search */ | |
518 | int line) /* source line number */ | |
519 | { | |
520 | ktrace_enter(xfs_alloc_trace_buf, | |
521 | (void *)(__psint_t)(trtype | (line << 16)), | |
522 | (void *)name, | |
523 | (void *)str, | |
524 | (void *)mp, | |
525 | (void *)(__psunsigned_t)agno, | |
526 | (void *)(__psunsigned_t)agbno, | |
527 | (void *)(__psunsigned_t)len, | |
528 | (void *)(__psint_t)slot, | |
529 | (void *)tp, | |
530 | NULL, NULL, NULL, NULL, NULL, NULL, NULL); | |
531 | } | |
532 | #endif /* XFS_ALLOC_TRACE */ | |
533 | ||
534 | /* | |
535 | * Allocation group level functions. | |
536 | */ | |
537 | ||
538 | /* | |
539 | * Allocate a variable extent in the allocation group agno. | |
540 | * Type and bno are used to determine where in the allocation group the | |
541 | * extent will start. | |
542 | * Extent's length (returned in *len) will be between minlen and maxlen, | |
543 | * and of the form k * prod + mod unless there's nothing that large. | |
544 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
545 | */ | |
546 | STATIC int /* error */ | |
547 | xfs_alloc_ag_vextent( | |
548 | xfs_alloc_arg_t *args) /* argument structure for allocation */ | |
549 | { | |
550 | int error=0; | |
1da177e4 LT |
551 | |
552 | ASSERT(args->minlen > 0); | |
553 | ASSERT(args->maxlen > 0); | |
554 | ASSERT(args->minlen <= args->maxlen); | |
555 | ASSERT(args->mod < args->prod); | |
556 | ASSERT(args->alignment > 0); | |
557 | /* | |
558 | * Branch to correct routine based on the type. | |
559 | */ | |
560 | args->wasfromfl = 0; | |
561 | switch (args->type) { | |
562 | case XFS_ALLOCTYPE_THIS_AG: | |
563 | error = xfs_alloc_ag_vextent_size(args); | |
564 | break; | |
565 | case XFS_ALLOCTYPE_NEAR_BNO: | |
566 | error = xfs_alloc_ag_vextent_near(args); | |
567 | break; | |
568 | case XFS_ALLOCTYPE_THIS_BNO: | |
569 | error = xfs_alloc_ag_vextent_exact(args); | |
570 | break; | |
571 | default: | |
572 | ASSERT(0); | |
573 | /* NOTREACHED */ | |
574 | } | |
575 | if (error) | |
576 | return error; | |
577 | /* | |
578 | * If the allocation worked, need to change the agf structure | |
579 | * (and log it), and the superblock. | |
580 | */ | |
581 | if (args->agbno != NULLAGBLOCK) { | |
582 | xfs_agf_t *agf; /* allocation group freelist header */ | |
583 | #ifdef XFS_ALLOC_TRACE | |
584 | xfs_mount_t *mp = args->mp; | |
585 | #endif | |
586 | long slen = (long)args->len; | |
587 | ||
588 | ASSERT(args->len >= args->minlen && args->len <= args->maxlen); | |
589 | ASSERT(!(args->wasfromfl) || !args->isfl); | |
590 | ASSERT(args->agbno % args->alignment == 0); | |
591 | if (!(args->wasfromfl)) { | |
592 | ||
593 | agf = XFS_BUF_TO_AGF(args->agbp); | |
413d57c9 | 594 | be32_add_cpu(&agf->agf_freeblks, -(args->len)); |
1da177e4 LT |
595 | xfs_trans_agblocks_delta(args->tp, |
596 | -((long)(args->len))); | |
597 | args->pag->pagf_freeblks -= args->len; | |
16259e7d CH |
598 | ASSERT(be32_to_cpu(agf->agf_freeblks) <= |
599 | be32_to_cpu(agf->agf_length)); | |
1da177e4 LT |
600 | TRACE_MODAGF(NULL, agf, XFS_AGF_FREEBLKS); |
601 | xfs_alloc_log_agf(args->tp, args->agbp, | |
602 | XFS_AGF_FREEBLKS); | |
603 | /* search the busylist for these blocks */ | |
604 | xfs_alloc_search_busy(args->tp, args->agno, | |
605 | args->agbno, args->len); | |
606 | } | |
607 | if (!args->isfl) | |
608 | xfs_trans_mod_sb(args->tp, | |
609 | args->wasdel ? XFS_TRANS_SB_RES_FDBLOCKS : | |
610 | XFS_TRANS_SB_FDBLOCKS, -slen); | |
611 | XFS_STATS_INC(xs_allocx); | |
612 | XFS_STATS_ADD(xs_allocb, args->len); | |
613 | } | |
614 | return 0; | |
615 | } | |
616 | ||
617 | /* | |
618 | * Allocate a variable extent at exactly agno/bno. | |
619 | * Extent's length (returned in *len) will be between minlen and maxlen, | |
620 | * and of the form k * prod + mod unless there's nothing that large. | |
621 | * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it. | |
622 | */ | |
623 | STATIC int /* error */ | |
624 | xfs_alloc_ag_vextent_exact( | |
625 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
626 | { | |
627 | xfs_btree_cur_t *bno_cur;/* by block-number btree cursor */ | |
628 | xfs_btree_cur_t *cnt_cur;/* by count btree cursor */ | |
629 | xfs_agblock_t end; /* end of allocated extent */ | |
630 | int error; | |
631 | xfs_agblock_t fbno; /* start block of found extent */ | |
632 | xfs_agblock_t fend; /* end block of found extent */ | |
633 | xfs_extlen_t flen; /* length of found extent */ | |
1da177e4 LT |
634 | int i; /* success/failure of operation */ |
635 | xfs_agblock_t maxend; /* end of maximal extent */ | |
636 | xfs_agblock_t minend; /* end of minimal extent */ | |
637 | xfs_extlen_t rlen; /* length of returned extent */ | |
638 | ||
639 | ASSERT(args->alignment == 1); | |
640 | /* | |
641 | * Allocate/initialize a cursor for the by-number freespace btree. | |
642 | */ | |
561f7d17 CH |
643 | bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
644 | args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
645 | /* |
646 | * Lookup bno and minlen in the btree (minlen is irrelevant, really). | |
647 | * Look for the closest free block <= bno, it must contain bno | |
648 | * if any free block does. | |
649 | */ | |
650 | if ((error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i))) | |
651 | goto error0; | |
652 | if (!i) { | |
653 | /* | |
654 | * Didn't find it, return null. | |
655 | */ | |
656 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
657 | args->agbno = NULLAGBLOCK; | |
658 | return 0; | |
659 | } | |
660 | /* | |
661 | * Grab the freespace record. | |
662 | */ | |
663 | if ((error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i))) | |
664 | goto error0; | |
665 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
666 | ASSERT(fbno <= args->agbno); | |
667 | minend = args->agbno + args->minlen; | |
668 | maxend = args->agbno + args->maxlen; | |
669 | fend = fbno + flen; | |
670 | /* | |
671 | * Give up if the freespace isn't long enough for the minimum request. | |
672 | */ | |
673 | if (fend < minend) { | |
674 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
675 | args->agbno = NULLAGBLOCK; | |
676 | return 0; | |
677 | } | |
678 | /* | |
679 | * End of extent will be smaller of the freespace end and the | |
680 | * maximal requested end. | |
681 | */ | |
682 | end = XFS_AGBLOCK_MIN(fend, maxend); | |
683 | /* | |
684 | * Fix the length according to mod and prod if given. | |
685 | */ | |
686 | args->len = end - args->agbno; | |
687 | xfs_alloc_fix_len(args); | |
688 | if (!xfs_alloc_fix_minleft(args)) { | |
689 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
690 | return 0; | |
691 | } | |
692 | rlen = args->len; | |
693 | ASSERT(args->agbno + rlen <= fend); | |
694 | end = args->agbno + rlen; | |
695 | /* | |
696 | * We are allocating agbno for rlen [agbno .. end] | |
697 | * Allocate/initialize a cursor for the by-size btree. | |
698 | */ | |
561f7d17 CH |
699 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
700 | args->agno, XFS_BTNUM_CNT); | |
1da177e4 | 701 | ASSERT(args->agbno + args->len <= |
16259e7d | 702 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
1da177e4 LT |
703 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, |
704 | args->agbno, args->len, XFSA_FIXUP_BNO_OK))) { | |
705 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
706 | goto error0; | |
707 | } | |
708 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
709 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
710 | TRACE_ALLOC("normal", args); | |
711 | args->wasfromfl = 0; | |
712 | return 0; | |
713 | ||
714 | error0: | |
715 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
716 | TRACE_ALLOC("error", args); | |
717 | return error; | |
718 | } | |
719 | ||
720 | /* | |
721 | * Allocate a variable extent near bno in the allocation group agno. | |
722 | * Extent's length (returned in len) will be between minlen and maxlen, | |
723 | * and of the form k * prod + mod unless there's nothing that large. | |
724 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
725 | */ | |
726 | STATIC int /* error */ | |
727 | xfs_alloc_ag_vextent_near( | |
728 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
729 | { | |
730 | xfs_btree_cur_t *bno_cur_gt; /* cursor for bno btree, right side */ | |
731 | xfs_btree_cur_t *bno_cur_lt; /* cursor for bno btree, left side */ | |
732 | xfs_btree_cur_t *cnt_cur; /* cursor for count btree */ | |
1da177e4 LT |
733 | xfs_agblock_t gtbno; /* start bno of right side entry */ |
734 | xfs_agblock_t gtbnoa; /* aligned ... */ | |
735 | xfs_extlen_t gtdiff; /* difference to right side entry */ | |
736 | xfs_extlen_t gtlen; /* length of right side entry */ | |
737 | xfs_extlen_t gtlena; /* aligned ... */ | |
738 | xfs_agblock_t gtnew; /* useful start bno of right side */ | |
739 | int error; /* error code */ | |
740 | int i; /* result code, temporary */ | |
741 | int j; /* result code, temporary */ | |
742 | xfs_agblock_t ltbno; /* start bno of left side entry */ | |
743 | xfs_agblock_t ltbnoa; /* aligned ... */ | |
744 | xfs_extlen_t ltdiff; /* difference to left side entry */ | |
745 | /*REFERENCED*/ | |
746 | xfs_agblock_t ltend; /* end bno of left side entry */ | |
747 | xfs_extlen_t ltlen; /* length of left side entry */ | |
748 | xfs_extlen_t ltlena; /* aligned ... */ | |
749 | xfs_agblock_t ltnew; /* useful start bno of left side */ | |
750 | xfs_extlen_t rlen; /* length of returned extent */ | |
751 | #if defined(DEBUG) && defined(__KERNEL__) | |
752 | /* | |
753 | * Randomly don't execute the first algorithm. | |
754 | */ | |
755 | int dofirst; /* set to do first algorithm */ | |
756 | ||
e7a23a9b | 757 | dofirst = random32() & 1; |
1da177e4 LT |
758 | #endif |
759 | /* | |
760 | * Get a cursor for the by-size btree. | |
761 | */ | |
561f7d17 CH |
762 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
763 | args->agno, XFS_BTNUM_CNT); | |
1da177e4 LT |
764 | ltlen = 0; |
765 | bno_cur_lt = bno_cur_gt = NULL; | |
766 | /* | |
767 | * See if there are any free extents as big as maxlen. | |
768 | */ | |
769 | if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, args->maxlen, &i))) | |
770 | goto error0; | |
771 | /* | |
772 | * If none, then pick up the last entry in the tree unless the | |
773 | * tree is empty. | |
774 | */ | |
775 | if (!i) { | |
776 | if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, <bno, | |
777 | <len, &i))) | |
778 | goto error0; | |
779 | if (i == 0 || ltlen == 0) { | |
780 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
781 | return 0; | |
782 | } | |
783 | ASSERT(i == 1); | |
784 | } | |
785 | args->wasfromfl = 0; | |
786 | /* | |
787 | * First algorithm. | |
788 | * If the requested extent is large wrt the freespaces available | |
789 | * in this a.g., then the cursor will be pointing to a btree entry | |
790 | * near the right edge of the tree. If it's in the last btree leaf | |
791 | * block, then we just examine all the entries in that block | |
792 | * that are big enough, and pick the best one. | |
793 | * This is written as a while loop so we can break out of it, | |
794 | * but we never loop back to the top. | |
795 | */ | |
796 | while (xfs_btree_islastblock(cnt_cur, 0)) { | |
797 | xfs_extlen_t bdiff; | |
798 | int besti=0; | |
799 | xfs_extlen_t blen=0; | |
800 | xfs_agblock_t bnew=0; | |
801 | ||
802 | #if defined(DEBUG) && defined(__KERNEL__) | |
803 | if (!dofirst) | |
804 | break; | |
805 | #endif | |
806 | /* | |
807 | * Start from the entry that lookup found, sequence through | |
808 | * all larger free blocks. If we're actually pointing at a | |
809 | * record smaller than maxlen, go to the start of this block, | |
810 | * and skip all those smaller than minlen. | |
811 | */ | |
812 | if (ltlen || args->alignment > 1) { | |
813 | cnt_cur->bc_ptrs[0] = 1; | |
814 | do { | |
815 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, | |
816 | <len, &i))) | |
817 | goto error0; | |
818 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
819 | if (ltlen >= args->minlen) | |
820 | break; | |
637aa50f | 821 | if ((error = xfs_btree_increment(cnt_cur, 0, &i))) |
1da177e4 LT |
822 | goto error0; |
823 | } while (i); | |
824 | ASSERT(ltlen >= args->minlen); | |
825 | if (!i) | |
826 | break; | |
827 | } | |
828 | i = cnt_cur->bc_ptrs[0]; | |
829 | for (j = 1, blen = 0, bdiff = 0; | |
830 | !error && j && (blen < args->maxlen || bdiff > 0); | |
637aa50f | 831 | error = xfs_btree_increment(cnt_cur, 0, &j)) { |
1da177e4 LT |
832 | /* |
833 | * For each entry, decide if it's better than | |
834 | * the previous best entry. | |
835 | */ | |
836 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i))) | |
837 | goto error0; | |
838 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
12375c82 DC |
839 | xfs_alloc_compute_aligned(ltbno, ltlen, args->alignment, |
840 | args->minlen, <bnoa, <lena); | |
e6430037 | 841 | if (ltlena < args->minlen) |
1da177e4 LT |
842 | continue; |
843 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); | |
844 | xfs_alloc_fix_len(args); | |
845 | ASSERT(args->len >= args->minlen); | |
846 | if (args->len < blen) | |
847 | continue; | |
848 | ltdiff = xfs_alloc_compute_diff(args->agbno, args->len, | |
849 | args->alignment, ltbno, ltlen, <new); | |
850 | if (ltnew != NULLAGBLOCK && | |
851 | (args->len > blen || ltdiff < bdiff)) { | |
852 | bdiff = ltdiff; | |
853 | bnew = ltnew; | |
854 | blen = args->len; | |
855 | besti = cnt_cur->bc_ptrs[0]; | |
856 | } | |
857 | } | |
858 | /* | |
859 | * It didn't work. We COULD be in a case where | |
860 | * there's a good record somewhere, so try again. | |
861 | */ | |
862 | if (blen == 0) | |
863 | break; | |
864 | /* | |
865 | * Point at the best entry, and retrieve it again. | |
866 | */ | |
867 | cnt_cur->bc_ptrs[0] = besti; | |
868 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i))) | |
869 | goto error0; | |
870 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
871 | ltend = ltbno + ltlen; | |
16259e7d | 872 | ASSERT(ltend <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
1da177e4 LT |
873 | args->len = blen; |
874 | if (!xfs_alloc_fix_minleft(args)) { | |
875 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
876 | TRACE_ALLOC("nominleft", args); | |
877 | return 0; | |
878 | } | |
879 | blen = args->len; | |
880 | /* | |
881 | * We are allocating starting at bnew for blen blocks. | |
882 | */ | |
883 | args->agbno = bnew; | |
884 | ASSERT(bnew >= ltbno); | |
885 | ASSERT(bnew + blen <= ltend); | |
886 | /* | |
887 | * Set up a cursor for the by-bno tree. | |
888 | */ | |
561f7d17 CH |
889 | bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, |
890 | args->agbp, args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
891 | /* |
892 | * Fix up the btree entries. | |
893 | */ | |
894 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, | |
895 | ltlen, bnew, blen, XFSA_FIXUP_CNT_OK))) | |
896 | goto error0; | |
897 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
898 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); | |
899 | TRACE_ALLOC("first", args); | |
900 | return 0; | |
901 | } | |
902 | /* | |
903 | * Second algorithm. | |
904 | * Search in the by-bno tree to the left and to the right | |
905 | * simultaneously, until in each case we find a space big enough, | |
906 | * or run into the edge of the tree. When we run into the edge, | |
907 | * we deallocate that cursor. | |
908 | * If both searches succeed, we compare the two spaces and pick | |
909 | * the better one. | |
910 | * With alignment, it's possible for both to fail; the upper | |
911 | * level algorithm that picks allocation groups for allocations | |
912 | * is not supposed to do this. | |
913 | */ | |
914 | /* | |
915 | * Allocate and initialize the cursor for the leftward search. | |
916 | */ | |
561f7d17 CH |
917 | bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
918 | args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
919 | /* |
920 | * Lookup <= bno to find the leftward search's starting point. | |
921 | */ | |
922 | if ((error = xfs_alloc_lookup_le(bno_cur_lt, args->agbno, args->maxlen, &i))) | |
923 | goto error0; | |
924 | if (!i) { | |
925 | /* | |
926 | * Didn't find anything; use this cursor for the rightward | |
927 | * search. | |
928 | */ | |
929 | bno_cur_gt = bno_cur_lt; | |
930 | bno_cur_lt = NULL; | |
931 | } | |
932 | /* | |
933 | * Found something. Duplicate the cursor for the rightward search. | |
934 | */ | |
935 | else if ((error = xfs_btree_dup_cursor(bno_cur_lt, &bno_cur_gt))) | |
936 | goto error0; | |
937 | /* | |
938 | * Increment the cursor, so we will point at the entry just right | |
939 | * of the leftward entry if any, or to the leftmost entry. | |
940 | */ | |
637aa50f | 941 | if ((error = xfs_btree_increment(bno_cur_gt, 0, &i))) |
1da177e4 LT |
942 | goto error0; |
943 | if (!i) { | |
944 | /* | |
945 | * It failed, there are no rightward entries. | |
946 | */ | |
947 | xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_NOERROR); | |
948 | bno_cur_gt = NULL; | |
949 | } | |
950 | /* | |
951 | * Loop going left with the leftward cursor, right with the | |
952 | * rightward cursor, until either both directions give up or | |
953 | * we find an entry at least as big as minlen. | |
954 | */ | |
955 | do { | |
956 | if (bno_cur_lt) { | |
957 | if ((error = xfs_alloc_get_rec(bno_cur_lt, <bno, <len, &i))) | |
958 | goto error0; | |
959 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
12375c82 DC |
960 | xfs_alloc_compute_aligned(ltbno, ltlen, args->alignment, |
961 | args->minlen, <bnoa, <lena); | |
962 | if (ltlena >= args->minlen) | |
1da177e4 | 963 | break; |
8df4da4a | 964 | if ((error = xfs_btree_decrement(bno_cur_lt, 0, &i))) |
1da177e4 LT |
965 | goto error0; |
966 | if (!i) { | |
967 | xfs_btree_del_cursor(bno_cur_lt, | |
968 | XFS_BTREE_NOERROR); | |
969 | bno_cur_lt = NULL; | |
970 | } | |
971 | } | |
972 | if (bno_cur_gt) { | |
973 | if ((error = xfs_alloc_get_rec(bno_cur_gt, >bno, >len, &i))) | |
974 | goto error0; | |
975 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
12375c82 DC |
976 | xfs_alloc_compute_aligned(gtbno, gtlen, args->alignment, |
977 | args->minlen, >bnoa, >lena); | |
978 | if (gtlena >= args->minlen) | |
1da177e4 | 979 | break; |
637aa50f | 980 | if ((error = xfs_btree_increment(bno_cur_gt, 0, &i))) |
1da177e4 LT |
981 | goto error0; |
982 | if (!i) { | |
983 | xfs_btree_del_cursor(bno_cur_gt, | |
984 | XFS_BTREE_NOERROR); | |
985 | bno_cur_gt = NULL; | |
986 | } | |
987 | } | |
988 | } while (bno_cur_lt || bno_cur_gt); | |
989 | /* | |
990 | * Got both cursors still active, need to find better entry. | |
991 | */ | |
992 | if (bno_cur_lt && bno_cur_gt) { | |
993 | /* | |
994 | * Left side is long enough, look for a right side entry. | |
995 | */ | |
996 | if (ltlena >= args->minlen) { | |
997 | /* | |
998 | * Fix up the length. | |
999 | */ | |
1000 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); | |
1001 | xfs_alloc_fix_len(args); | |
1002 | rlen = args->len; | |
1003 | ltdiff = xfs_alloc_compute_diff(args->agbno, rlen, | |
1004 | args->alignment, ltbno, ltlen, <new); | |
1005 | /* | |
1006 | * Not perfect. | |
1007 | */ | |
1008 | if (ltdiff) { | |
1009 | /* | |
1010 | * Look until we find a better one, run out of | |
1011 | * space, or run off the end. | |
1012 | */ | |
1013 | while (bno_cur_lt && bno_cur_gt) { | |
1014 | if ((error = xfs_alloc_get_rec( | |
1015 | bno_cur_gt, >bno, | |
1016 | >len, &i))) | |
1017 | goto error0; | |
1018 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1019 | xfs_alloc_compute_aligned(gtbno, gtlen, | |
1020 | args->alignment, args->minlen, | |
1021 | >bnoa, >lena); | |
1022 | /* | |
1023 | * The left one is clearly better. | |
1024 | */ | |
1025 | if (gtbnoa >= args->agbno + ltdiff) { | |
1026 | xfs_btree_del_cursor( | |
1027 | bno_cur_gt, | |
1028 | XFS_BTREE_NOERROR); | |
1029 | bno_cur_gt = NULL; | |
1030 | break; | |
1031 | } | |
1032 | /* | |
1033 | * If we reach a big enough entry, | |
1034 | * compare the two and pick the best. | |
1035 | */ | |
1036 | if (gtlena >= args->minlen) { | |
1037 | args->len = | |
1038 | XFS_EXTLEN_MIN(gtlena, | |
1039 | args->maxlen); | |
1040 | xfs_alloc_fix_len(args); | |
1041 | rlen = args->len; | |
1042 | gtdiff = xfs_alloc_compute_diff( | |
1043 | args->agbno, rlen, | |
1044 | args->alignment, | |
1045 | gtbno, gtlen, >new); | |
1046 | /* | |
1047 | * Right side is better. | |
1048 | */ | |
1049 | if (gtdiff < ltdiff) { | |
1050 | xfs_btree_del_cursor( | |
1051 | bno_cur_lt, | |
1052 | XFS_BTREE_NOERROR); | |
1053 | bno_cur_lt = NULL; | |
1054 | } | |
1055 | /* | |
1056 | * Left side is better. | |
1057 | */ | |
1058 | else { | |
1059 | xfs_btree_del_cursor( | |
1060 | bno_cur_gt, | |
1061 | XFS_BTREE_NOERROR); | |
1062 | bno_cur_gt = NULL; | |
1063 | } | |
1064 | break; | |
1065 | } | |
1066 | /* | |
1067 | * Fell off the right end. | |
1068 | */ | |
637aa50f | 1069 | if ((error = xfs_btree_increment( |
1da177e4 LT |
1070 | bno_cur_gt, 0, &i))) |
1071 | goto error0; | |
1072 | if (!i) { | |
1073 | xfs_btree_del_cursor( | |
1074 | bno_cur_gt, | |
1075 | XFS_BTREE_NOERROR); | |
1076 | bno_cur_gt = NULL; | |
1077 | break; | |
1078 | } | |
1079 | } | |
1080 | } | |
1081 | /* | |
1082 | * The left side is perfect, trash the right side. | |
1083 | */ | |
1084 | else { | |
1085 | xfs_btree_del_cursor(bno_cur_gt, | |
1086 | XFS_BTREE_NOERROR); | |
1087 | bno_cur_gt = NULL; | |
1088 | } | |
1089 | } | |
1090 | /* | |
1091 | * It's the right side that was found first, look left. | |
1092 | */ | |
1093 | else { | |
1094 | /* | |
1095 | * Fix up the length. | |
1096 | */ | |
1097 | args->len = XFS_EXTLEN_MIN(gtlena, args->maxlen); | |
1098 | xfs_alloc_fix_len(args); | |
1099 | rlen = args->len; | |
1100 | gtdiff = xfs_alloc_compute_diff(args->agbno, rlen, | |
1101 | args->alignment, gtbno, gtlen, >new); | |
1102 | /* | |
1103 | * Right side entry isn't perfect. | |
1104 | */ | |
1105 | if (gtdiff) { | |
1106 | /* | |
1107 | * Look until we find a better one, run out of | |
1108 | * space, or run off the end. | |
1109 | */ | |
1110 | while (bno_cur_lt && bno_cur_gt) { | |
1111 | if ((error = xfs_alloc_get_rec( | |
1112 | bno_cur_lt, <bno, | |
1113 | <len, &i))) | |
1114 | goto error0; | |
1115 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1116 | xfs_alloc_compute_aligned(ltbno, ltlen, | |
1117 | args->alignment, args->minlen, | |
1118 | <bnoa, <lena); | |
1119 | /* | |
1120 | * The right one is clearly better. | |
1121 | */ | |
1122 | if (ltbnoa <= args->agbno - gtdiff) { | |
1123 | xfs_btree_del_cursor( | |
1124 | bno_cur_lt, | |
1125 | XFS_BTREE_NOERROR); | |
1126 | bno_cur_lt = NULL; | |
1127 | break; | |
1128 | } | |
1129 | /* | |
1130 | * If we reach a big enough entry, | |
1131 | * compare the two and pick the best. | |
1132 | */ | |
1133 | if (ltlena >= args->minlen) { | |
1134 | args->len = XFS_EXTLEN_MIN( | |
1135 | ltlena, args->maxlen); | |
1136 | xfs_alloc_fix_len(args); | |
1137 | rlen = args->len; | |
1138 | ltdiff = xfs_alloc_compute_diff( | |
1139 | args->agbno, rlen, | |
1140 | args->alignment, | |
1141 | ltbno, ltlen, <new); | |
1142 | /* | |
1143 | * Left side is better. | |
1144 | */ | |
1145 | if (ltdiff < gtdiff) { | |
1146 | xfs_btree_del_cursor( | |
1147 | bno_cur_gt, | |
1148 | XFS_BTREE_NOERROR); | |
1149 | bno_cur_gt = NULL; | |
1150 | } | |
1151 | /* | |
1152 | * Right side is better. | |
1153 | */ | |
1154 | else { | |
1155 | xfs_btree_del_cursor( | |
1156 | bno_cur_lt, | |
1157 | XFS_BTREE_NOERROR); | |
1158 | bno_cur_lt = NULL; | |
1159 | } | |
1160 | break; | |
1161 | } | |
1162 | /* | |
1163 | * Fell off the left end. | |
1164 | */ | |
8df4da4a | 1165 | if ((error = xfs_btree_decrement( |
1da177e4 LT |
1166 | bno_cur_lt, 0, &i))) |
1167 | goto error0; | |
1168 | if (!i) { | |
1169 | xfs_btree_del_cursor(bno_cur_lt, | |
1170 | XFS_BTREE_NOERROR); | |
1171 | bno_cur_lt = NULL; | |
1172 | break; | |
1173 | } | |
1174 | } | |
1175 | } | |
1176 | /* | |
1177 | * The right side is perfect, trash the left side. | |
1178 | */ | |
1179 | else { | |
1180 | xfs_btree_del_cursor(bno_cur_lt, | |
1181 | XFS_BTREE_NOERROR); | |
1182 | bno_cur_lt = NULL; | |
1183 | } | |
1184 | } | |
1185 | } | |
1186 | /* | |
1187 | * If we couldn't get anything, give up. | |
1188 | */ | |
1189 | if (bno_cur_lt == NULL && bno_cur_gt == NULL) { | |
1190 | TRACE_ALLOC("neither", args); | |
1191 | args->agbno = NULLAGBLOCK; | |
1192 | return 0; | |
1193 | } | |
1194 | /* | |
1195 | * At this point we have selected a freespace entry, either to the | |
1196 | * left or to the right. If it's on the right, copy all the | |
1197 | * useful variables to the "left" set so we only have one | |
1198 | * copy of this code. | |
1199 | */ | |
1200 | if (bno_cur_gt) { | |
1201 | bno_cur_lt = bno_cur_gt; | |
1202 | bno_cur_gt = NULL; | |
1203 | ltbno = gtbno; | |
1204 | ltbnoa = gtbnoa; | |
1205 | ltlen = gtlen; | |
1206 | ltlena = gtlena; | |
1207 | j = 1; | |
1208 | } else | |
1209 | j = 0; | |
1210 | /* | |
1211 | * Fix up the length and compute the useful address. | |
1212 | */ | |
1213 | ltend = ltbno + ltlen; | |
1214 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); | |
1215 | xfs_alloc_fix_len(args); | |
1216 | if (!xfs_alloc_fix_minleft(args)) { | |
1217 | TRACE_ALLOC("nominleft", args); | |
1218 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); | |
1219 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1220 | return 0; | |
1221 | } | |
1222 | rlen = args->len; | |
1223 | (void)xfs_alloc_compute_diff(args->agbno, rlen, args->alignment, ltbno, | |
1224 | ltlen, <new); | |
1225 | ASSERT(ltnew >= ltbno); | |
1226 | ASSERT(ltnew + rlen <= ltend); | |
16259e7d | 1227 | ASSERT(ltnew + rlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
1da177e4 LT |
1228 | args->agbno = ltnew; |
1229 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, ltlen, | |
1230 | ltnew, rlen, XFSA_FIXUP_BNO_OK))) | |
1231 | goto error0; | |
1232 | TRACE_ALLOC(j ? "gt" : "lt", args); | |
1233 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1234 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); | |
1235 | return 0; | |
1236 | ||
1237 | error0: | |
1238 | TRACE_ALLOC("error", args); | |
1239 | if (cnt_cur != NULL) | |
1240 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1241 | if (bno_cur_lt != NULL) | |
1242 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_ERROR); | |
1243 | if (bno_cur_gt != NULL) | |
1244 | xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_ERROR); | |
1245 | return error; | |
1246 | } | |
1247 | ||
1248 | /* | |
1249 | * Allocate a variable extent anywhere in the allocation group agno. | |
1250 | * Extent's length (returned in len) will be between minlen and maxlen, | |
1251 | * and of the form k * prod + mod unless there's nothing that large. | |
1252 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
1253 | */ | |
1254 | STATIC int /* error */ | |
1255 | xfs_alloc_ag_vextent_size( | |
1256 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
1257 | { | |
1258 | xfs_btree_cur_t *bno_cur; /* cursor for bno btree */ | |
1259 | xfs_btree_cur_t *cnt_cur; /* cursor for cnt btree */ | |
1260 | int error; /* error result */ | |
1261 | xfs_agblock_t fbno; /* start of found freespace */ | |
1262 | xfs_extlen_t flen; /* length of found freespace */ | |
1da177e4 LT |
1263 | int i; /* temp status variable */ |
1264 | xfs_agblock_t rbno; /* returned block number */ | |
1265 | xfs_extlen_t rlen; /* length of returned extent */ | |
1266 | ||
1267 | /* | |
1268 | * Allocate and initialize a cursor for the by-size btree. | |
1269 | */ | |
561f7d17 CH |
1270 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1271 | args->agno, XFS_BTNUM_CNT); | |
1da177e4 LT |
1272 | bno_cur = NULL; |
1273 | /* | |
1274 | * Look for an entry >= maxlen+alignment-1 blocks. | |
1275 | */ | |
1276 | if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, | |
1277 | args->maxlen + args->alignment - 1, &i))) | |
1278 | goto error0; | |
1279 | /* | |
1280 | * If none, then pick up the last entry in the tree unless the | |
1281 | * tree is empty. | |
1282 | */ | |
1283 | if (!i) { | |
1284 | if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, &fbno, | |
1285 | &flen, &i))) | |
1286 | goto error0; | |
1287 | if (i == 0 || flen == 0) { | |
1288 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1289 | TRACE_ALLOC("noentry", args); | |
1290 | return 0; | |
1291 | } | |
1292 | ASSERT(i == 1); | |
1293 | } | |
1294 | /* | |
1295 | * There's a freespace as big as maxlen+alignment-1, get it. | |
1296 | */ | |
1297 | else { | |
1298 | if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i))) | |
1299 | goto error0; | |
1300 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1301 | } | |
1302 | /* | |
1303 | * In the first case above, we got the last entry in the | |
1304 | * by-size btree. Now we check to see if the space hits maxlen | |
1305 | * once aligned; if not, we search left for something better. | |
1306 | * This can't happen in the second case above. | |
1307 | */ | |
1308 | xfs_alloc_compute_aligned(fbno, flen, args->alignment, args->minlen, | |
1309 | &rbno, &rlen); | |
1310 | rlen = XFS_EXTLEN_MIN(args->maxlen, rlen); | |
1311 | XFS_WANT_CORRUPTED_GOTO(rlen == 0 || | |
1312 | (rlen <= flen && rbno + rlen <= fbno + flen), error0); | |
1313 | if (rlen < args->maxlen) { | |
1314 | xfs_agblock_t bestfbno; | |
1315 | xfs_extlen_t bestflen; | |
1316 | xfs_agblock_t bestrbno; | |
1317 | xfs_extlen_t bestrlen; | |
1318 | ||
1319 | bestrlen = rlen; | |
1320 | bestrbno = rbno; | |
1321 | bestflen = flen; | |
1322 | bestfbno = fbno; | |
1323 | for (;;) { | |
8df4da4a | 1324 | if ((error = xfs_btree_decrement(cnt_cur, 0, &i))) |
1da177e4 LT |
1325 | goto error0; |
1326 | if (i == 0) | |
1327 | break; | |
1328 | if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, | |
1329 | &i))) | |
1330 | goto error0; | |
1331 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1332 | if (flen < bestrlen) | |
1333 | break; | |
1334 | xfs_alloc_compute_aligned(fbno, flen, args->alignment, | |
1335 | args->minlen, &rbno, &rlen); | |
1336 | rlen = XFS_EXTLEN_MIN(args->maxlen, rlen); | |
1337 | XFS_WANT_CORRUPTED_GOTO(rlen == 0 || | |
1338 | (rlen <= flen && rbno + rlen <= fbno + flen), | |
1339 | error0); | |
1340 | if (rlen > bestrlen) { | |
1341 | bestrlen = rlen; | |
1342 | bestrbno = rbno; | |
1343 | bestflen = flen; | |
1344 | bestfbno = fbno; | |
1345 | if (rlen == args->maxlen) | |
1346 | break; | |
1347 | } | |
1348 | } | |
1349 | if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen, | |
1350 | &i))) | |
1351 | goto error0; | |
1352 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1353 | rlen = bestrlen; | |
1354 | rbno = bestrbno; | |
1355 | flen = bestflen; | |
1356 | fbno = bestfbno; | |
1357 | } | |
1358 | args->wasfromfl = 0; | |
1359 | /* | |
1360 | * Fix up the length. | |
1361 | */ | |
1362 | args->len = rlen; | |
1363 | xfs_alloc_fix_len(args); | |
1364 | if (rlen < args->minlen || !xfs_alloc_fix_minleft(args)) { | |
1365 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1366 | TRACE_ALLOC("nominleft", args); | |
1367 | args->agbno = NULLAGBLOCK; | |
1368 | return 0; | |
1369 | } | |
1370 | rlen = args->len; | |
1371 | XFS_WANT_CORRUPTED_GOTO(rlen <= flen, error0); | |
1372 | /* | |
1373 | * Allocate and initialize a cursor for the by-block tree. | |
1374 | */ | |
561f7d17 CH |
1375 | bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1376 | args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
1377 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, |
1378 | rbno, rlen, XFSA_FIXUP_CNT_OK))) | |
1379 | goto error0; | |
1380 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1381 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
1382 | cnt_cur = bno_cur = NULL; | |
1383 | args->len = rlen; | |
1384 | args->agbno = rbno; | |
1385 | XFS_WANT_CORRUPTED_GOTO( | |
1386 | args->agbno + args->len <= | |
16259e7d | 1387 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length), |
1da177e4 LT |
1388 | error0); |
1389 | TRACE_ALLOC("normal", args); | |
1390 | return 0; | |
1391 | ||
1392 | error0: | |
1393 | TRACE_ALLOC("error", args); | |
1394 | if (cnt_cur) | |
1395 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1396 | if (bno_cur) | |
1397 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
1398 | return error; | |
1399 | } | |
1400 | ||
1401 | /* | |
1402 | * Deal with the case where only small freespaces remain. | |
1403 | * Either return the contents of the last freespace record, | |
1404 | * or allocate space from the freelist if there is nothing in the tree. | |
1405 | */ | |
1406 | STATIC int /* error */ | |
1407 | xfs_alloc_ag_vextent_small( | |
1408 | xfs_alloc_arg_t *args, /* allocation argument structure */ | |
1409 | xfs_btree_cur_t *ccur, /* by-size cursor */ | |
1410 | xfs_agblock_t *fbnop, /* result block number */ | |
1411 | xfs_extlen_t *flenp, /* result length */ | |
1412 | int *stat) /* status: 0-freelist, 1-normal/none */ | |
1413 | { | |
1414 | int error; | |
1415 | xfs_agblock_t fbno; | |
1416 | xfs_extlen_t flen; | |
1da177e4 LT |
1417 | int i; |
1418 | ||
8df4da4a | 1419 | if ((error = xfs_btree_decrement(ccur, 0, &i))) |
1da177e4 LT |
1420 | goto error0; |
1421 | if (i) { | |
1422 | if ((error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i))) | |
1423 | goto error0; | |
1424 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1425 | } | |
1426 | /* | |
1427 | * Nothing in the btree, try the freelist. Make sure | |
1428 | * to respect minleft even when pulling from the | |
1429 | * freelist. | |
1430 | */ | |
1431 | else if (args->minlen == 1 && args->alignment == 1 && !args->isfl && | |
16259e7d CH |
1432 | (be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_flcount) |
1433 | > args->minleft)) { | |
92821e2b DC |
1434 | error = xfs_alloc_get_freelist(args->tp, args->agbp, &fbno, 0); |
1435 | if (error) | |
1da177e4 LT |
1436 | goto error0; |
1437 | if (fbno != NULLAGBLOCK) { | |
1438 | if (args->userdata) { | |
1439 | xfs_buf_t *bp; | |
1440 | ||
1441 | bp = xfs_btree_get_bufs(args->mp, args->tp, | |
1442 | args->agno, fbno, 0); | |
1443 | xfs_trans_binval(args->tp, bp); | |
1444 | } | |
1445 | args->len = 1; | |
1446 | args->agbno = fbno; | |
1447 | XFS_WANT_CORRUPTED_GOTO( | |
1448 | args->agbno + args->len <= | |
16259e7d | 1449 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length), |
1da177e4 LT |
1450 | error0); |
1451 | args->wasfromfl = 1; | |
1452 | TRACE_ALLOC("freelist", args); | |
1453 | *stat = 0; | |
1454 | return 0; | |
1455 | } | |
1456 | /* | |
1457 | * Nothing in the freelist. | |
1458 | */ | |
1459 | else | |
1460 | flen = 0; | |
1461 | } | |
1462 | /* | |
1463 | * Can't allocate from the freelist for some reason. | |
1464 | */ | |
d432c80e NS |
1465 | else { |
1466 | fbno = NULLAGBLOCK; | |
1da177e4 | 1467 | flen = 0; |
d432c80e | 1468 | } |
1da177e4 LT |
1469 | /* |
1470 | * Can't do the allocation, give up. | |
1471 | */ | |
1472 | if (flen < args->minlen) { | |
1473 | args->agbno = NULLAGBLOCK; | |
1474 | TRACE_ALLOC("notenough", args); | |
1475 | flen = 0; | |
1476 | } | |
1477 | *fbnop = fbno; | |
1478 | *flenp = flen; | |
1479 | *stat = 1; | |
1480 | TRACE_ALLOC("normal", args); | |
1481 | return 0; | |
1482 | ||
1483 | error0: | |
1484 | TRACE_ALLOC("error", args); | |
1485 | return error; | |
1486 | } | |
1487 | ||
1488 | /* | |
1489 | * Free the extent starting at agno/bno for length. | |
1490 | */ | |
1491 | STATIC int /* error */ | |
1492 | xfs_free_ag_extent( | |
1493 | xfs_trans_t *tp, /* transaction pointer */ | |
1494 | xfs_buf_t *agbp, /* buffer for a.g. freelist header */ | |
1495 | xfs_agnumber_t agno, /* allocation group number */ | |
1496 | xfs_agblock_t bno, /* starting block number */ | |
1497 | xfs_extlen_t len, /* length of extent */ | |
1498 | int isfl) /* set if is freelist blocks - no sb acctg */ | |
1499 | { | |
1500 | xfs_btree_cur_t *bno_cur; /* cursor for by-block btree */ | |
1501 | xfs_btree_cur_t *cnt_cur; /* cursor for by-size btree */ | |
1502 | int error; /* error return value */ | |
1da177e4 LT |
1503 | xfs_agblock_t gtbno; /* start of right neighbor block */ |
1504 | xfs_extlen_t gtlen; /* length of right neighbor block */ | |
1505 | int haveleft; /* have a left neighbor block */ | |
1506 | int haveright; /* have a right neighbor block */ | |
1507 | int i; /* temp, result code */ | |
1508 | xfs_agblock_t ltbno; /* start of left neighbor block */ | |
1509 | xfs_extlen_t ltlen; /* length of left neighbor block */ | |
1510 | xfs_mount_t *mp; /* mount point struct for filesystem */ | |
1511 | xfs_agblock_t nbno; /* new starting block of freespace */ | |
1512 | xfs_extlen_t nlen; /* new length of freespace */ | |
1513 | ||
1514 | mp = tp->t_mountp; | |
1515 | /* | |
1516 | * Allocate and initialize a cursor for the by-block btree. | |
1517 | */ | |
561f7d17 | 1518 | bno_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_BNO); |
1da177e4 LT |
1519 | cnt_cur = NULL; |
1520 | /* | |
1521 | * Look for a neighboring block on the left (lower block numbers) | |
1522 | * that is contiguous with this space. | |
1523 | */ | |
1524 | if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft))) | |
1525 | goto error0; | |
1526 | if (haveleft) { | |
1527 | /* | |
1528 | * There is a block to our left. | |
1529 | */ | |
1530 | if ((error = xfs_alloc_get_rec(bno_cur, <bno, <len, &i))) | |
1531 | goto error0; | |
1532 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1533 | /* | |
1534 | * It's not contiguous, though. | |
1535 | */ | |
1536 | if (ltbno + ltlen < bno) | |
1537 | haveleft = 0; | |
1538 | else { | |
1539 | /* | |
1540 | * If this failure happens the request to free this | |
1541 | * space was invalid, it's (partly) already free. | |
1542 | * Very bad. | |
1543 | */ | |
1544 | XFS_WANT_CORRUPTED_GOTO(ltbno + ltlen <= bno, error0); | |
1545 | } | |
1546 | } | |
1547 | /* | |
1548 | * Look for a neighboring block on the right (higher block numbers) | |
1549 | * that is contiguous with this space. | |
1550 | */ | |
637aa50f | 1551 | if ((error = xfs_btree_increment(bno_cur, 0, &haveright))) |
1da177e4 LT |
1552 | goto error0; |
1553 | if (haveright) { | |
1554 | /* | |
1555 | * There is a block to our right. | |
1556 | */ | |
1557 | if ((error = xfs_alloc_get_rec(bno_cur, >bno, >len, &i))) | |
1558 | goto error0; | |
1559 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1560 | /* | |
1561 | * It's not contiguous, though. | |
1562 | */ | |
1563 | if (bno + len < gtbno) | |
1564 | haveright = 0; | |
1565 | else { | |
1566 | /* | |
1567 | * If this failure happens the request to free this | |
1568 | * space was invalid, it's (partly) already free. | |
1569 | * Very bad. | |
1570 | */ | |
1571 | XFS_WANT_CORRUPTED_GOTO(gtbno >= bno + len, error0); | |
1572 | } | |
1573 | } | |
1574 | /* | |
1575 | * Now allocate and initialize a cursor for the by-size tree. | |
1576 | */ | |
561f7d17 | 1577 | cnt_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_CNT); |
1da177e4 LT |
1578 | /* |
1579 | * Have both left and right contiguous neighbors. | |
1580 | * Merge all three into a single free block. | |
1581 | */ | |
1582 | if (haveleft && haveright) { | |
1583 | /* | |
1584 | * Delete the old by-size entry on the left. | |
1585 | */ | |
1586 | if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i))) | |
1587 | goto error0; | |
1588 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1589 | if ((error = xfs_alloc_delete(cnt_cur, &i))) | |
1590 | goto error0; | |
1591 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1592 | /* | |
1593 | * Delete the old by-size entry on the right. | |
1594 | */ | |
1595 | if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i))) | |
1596 | goto error0; | |
1597 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1598 | if ((error = xfs_alloc_delete(cnt_cur, &i))) | |
1599 | goto error0; | |
1600 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1601 | /* | |
1602 | * Delete the old by-block entry for the right block. | |
1603 | */ | |
1604 | if ((error = xfs_alloc_delete(bno_cur, &i))) | |
1605 | goto error0; | |
1606 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1607 | /* | |
1608 | * Move the by-block cursor back to the left neighbor. | |
1609 | */ | |
8df4da4a | 1610 | if ((error = xfs_btree_decrement(bno_cur, 0, &i))) |
1da177e4 LT |
1611 | goto error0; |
1612 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1613 | #ifdef DEBUG | |
1614 | /* | |
1615 | * Check that this is the right record: delete didn't | |
1616 | * mangle the cursor. | |
1617 | */ | |
1618 | { | |
1619 | xfs_agblock_t xxbno; | |
1620 | xfs_extlen_t xxlen; | |
1621 | ||
1622 | if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen, | |
1623 | &i))) | |
1624 | goto error0; | |
1625 | XFS_WANT_CORRUPTED_GOTO( | |
1626 | i == 1 && xxbno == ltbno && xxlen == ltlen, | |
1627 | error0); | |
1628 | } | |
1629 | #endif | |
1630 | /* | |
1631 | * Update remaining by-block entry to the new, joined block. | |
1632 | */ | |
1633 | nbno = ltbno; | |
1634 | nlen = len + ltlen + gtlen; | |
1635 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
1636 | goto error0; | |
1637 | } | |
1638 | /* | |
1639 | * Have only a left contiguous neighbor. | |
1640 | * Merge it together with the new freespace. | |
1641 | */ | |
1642 | else if (haveleft) { | |
1643 | /* | |
1644 | * Delete the old by-size entry on the left. | |
1645 | */ | |
1646 | if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i))) | |
1647 | goto error0; | |
1648 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1649 | if ((error = xfs_alloc_delete(cnt_cur, &i))) | |
1650 | goto error0; | |
1651 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1652 | /* | |
1653 | * Back up the by-block cursor to the left neighbor, and | |
1654 | * update its length. | |
1655 | */ | |
8df4da4a | 1656 | if ((error = xfs_btree_decrement(bno_cur, 0, &i))) |
1da177e4 LT |
1657 | goto error0; |
1658 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1659 | nbno = ltbno; | |
1660 | nlen = len + ltlen; | |
1661 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
1662 | goto error0; | |
1663 | } | |
1664 | /* | |
1665 | * Have only a right contiguous neighbor. | |
1666 | * Merge it together with the new freespace. | |
1667 | */ | |
1668 | else if (haveright) { | |
1669 | /* | |
1670 | * Delete the old by-size entry on the right. | |
1671 | */ | |
1672 | if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i))) | |
1673 | goto error0; | |
1674 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1675 | if ((error = xfs_alloc_delete(cnt_cur, &i))) | |
1676 | goto error0; | |
1677 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1678 | /* | |
1679 | * Update the starting block and length of the right | |
1680 | * neighbor in the by-block tree. | |
1681 | */ | |
1682 | nbno = bno; | |
1683 | nlen = len + gtlen; | |
1684 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
1685 | goto error0; | |
1686 | } | |
1687 | /* | |
1688 | * No contiguous neighbors. | |
1689 | * Insert the new freespace into the by-block tree. | |
1690 | */ | |
1691 | else { | |
1692 | nbno = bno; | |
1693 | nlen = len; | |
1694 | if ((error = xfs_alloc_insert(bno_cur, &i))) | |
1695 | goto error0; | |
1696 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1697 | } | |
1698 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
1699 | bno_cur = NULL; | |
1700 | /* | |
1701 | * In all cases we need to insert the new freespace in the by-size tree. | |
1702 | */ | |
1703 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i))) | |
1704 | goto error0; | |
1705 | XFS_WANT_CORRUPTED_GOTO(i == 0, error0); | |
1706 | if ((error = xfs_alloc_insert(cnt_cur, &i))) | |
1707 | goto error0; | |
1708 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1709 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1710 | cnt_cur = NULL; | |
1711 | /* | |
1712 | * Update the freespace totals in the ag and superblock. | |
1713 | */ | |
1714 | { | |
1715 | xfs_agf_t *agf; | |
1716 | xfs_perag_t *pag; /* per allocation group data */ | |
1717 | ||
1718 | agf = XFS_BUF_TO_AGF(agbp); | |
1719 | pag = &mp->m_perag[agno]; | |
413d57c9 | 1720 | be32_add_cpu(&agf->agf_freeblks, len); |
1da177e4 LT |
1721 | xfs_trans_agblocks_delta(tp, len); |
1722 | pag->pagf_freeblks += len; | |
1723 | XFS_WANT_CORRUPTED_GOTO( | |
16259e7d CH |
1724 | be32_to_cpu(agf->agf_freeblks) <= |
1725 | be32_to_cpu(agf->agf_length), | |
1da177e4 LT |
1726 | error0); |
1727 | TRACE_MODAGF(NULL, agf, XFS_AGF_FREEBLKS); | |
1728 | xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS); | |
1729 | if (!isfl) | |
1730 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, (long)len); | |
1731 | XFS_STATS_INC(xs_freex); | |
1732 | XFS_STATS_ADD(xs_freeb, len); | |
1733 | } | |
1734 | TRACE_FREE(haveleft ? | |
1735 | (haveright ? "both" : "left") : | |
1736 | (haveright ? "right" : "none"), | |
1737 | agno, bno, len, isfl); | |
1738 | ||
1739 | /* | |
1740 | * Since blocks move to the free list without the coordination | |
1741 | * used in xfs_bmap_finish, we can't allow block to be available | |
1742 | * for reallocation and non-transaction writing (user data) | |
1743 | * until we know that the transaction that moved it to the free | |
1744 | * list is permanently on disk. We track the blocks by declaring | |
1745 | * these blocks as "busy"; the busy list is maintained on a per-ag | |
1746 | * basis and each transaction records which entries should be removed | |
1747 | * when the iclog commits to disk. If a busy block is allocated, | |
1748 | * the iclog is pushed up to the LSN that freed the block. | |
1749 | */ | |
1750 | xfs_alloc_mark_busy(tp, agno, bno, len); | |
1751 | return 0; | |
1752 | ||
1753 | error0: | |
1754 | TRACE_FREE("error", agno, bno, len, isfl); | |
1755 | if (bno_cur) | |
1756 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
1757 | if (cnt_cur) | |
1758 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1759 | return error; | |
1760 | } | |
1761 | ||
1762 | /* | |
1763 | * Visible (exported) allocation/free functions. | |
1764 | * Some of these are used just by xfs_alloc_btree.c and this file. | |
1765 | */ | |
1766 | ||
1767 | /* | |
1768 | * Compute and fill in value of m_ag_maxlevels. | |
1769 | */ | |
1770 | void | |
1771 | xfs_alloc_compute_maxlevels( | |
1772 | xfs_mount_t *mp) /* file system mount structure */ | |
1773 | { | |
1774 | int level; | |
1775 | uint maxblocks; | |
1776 | uint maxleafents; | |
1777 | int minleafrecs; | |
1778 | int minnoderecs; | |
1779 | ||
1780 | maxleafents = (mp->m_sb.sb_agblocks + 1) / 2; | |
1781 | minleafrecs = mp->m_alloc_mnr[0]; | |
1782 | minnoderecs = mp->m_alloc_mnr[1]; | |
1783 | maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs; | |
1784 | for (level = 1; maxblocks > 1; level++) | |
1785 | maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs; | |
1786 | mp->m_ag_maxlevels = level; | |
1787 | } | |
1788 | ||
1789 | /* | |
1790 | * Decide whether to use this allocation group for this allocation. | |
1791 | * If so, fix up the btree freelist's size. | |
1792 | */ | |
1793 | STATIC int /* error */ | |
1794 | xfs_alloc_fix_freelist( | |
1795 | xfs_alloc_arg_t *args, /* allocation argument structure */ | |
1796 | int flags) /* XFS_ALLOC_FLAG_... */ | |
1797 | { | |
1798 | xfs_buf_t *agbp; /* agf buffer pointer */ | |
1799 | xfs_agf_t *agf; /* a.g. freespace structure pointer */ | |
1800 | xfs_buf_t *agflbp;/* agfl buffer pointer */ | |
1801 | xfs_agblock_t bno; /* freelist block */ | |
1802 | xfs_extlen_t delta; /* new blocks needed in freelist */ | |
1803 | int error; /* error result code */ | |
1804 | xfs_extlen_t longest;/* longest extent in allocation group */ | |
1805 | xfs_mount_t *mp; /* file system mount point structure */ | |
1806 | xfs_extlen_t need; /* total blocks needed in freelist */ | |
1807 | xfs_perag_t *pag; /* per-ag information structure */ | |
1808 | xfs_alloc_arg_t targs; /* local allocation arguments */ | |
1809 | xfs_trans_t *tp; /* transaction pointer */ | |
1810 | ||
1811 | mp = args->mp; | |
1812 | ||
1813 | pag = args->pag; | |
1814 | tp = args->tp; | |
1815 | if (!pag->pagf_init) { | |
1816 | if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags, | |
1817 | &agbp))) | |
1818 | return error; | |
1819 | if (!pag->pagf_init) { | |
0e1edbd9 NS |
1820 | ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK); |
1821 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
1da177e4 LT |
1822 | args->agbp = NULL; |
1823 | return 0; | |
1824 | } | |
1825 | } else | |
1826 | agbp = NULL; | |
1827 | ||
0e1edbd9 NS |
1828 | /* |
1829 | * If this is a metadata preferred pag and we are user data | |
1da177e4 LT |
1830 | * then try somewhere else if we are not being asked to |
1831 | * try harder at this point | |
1832 | */ | |
0e1edbd9 NS |
1833 | if (pag->pagf_metadata && args->userdata && |
1834 | (flags & XFS_ALLOC_FLAG_TRYLOCK)) { | |
1835 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
1da177e4 LT |
1836 | args->agbp = NULL; |
1837 | return 0; | |
1838 | } | |
1839 | ||
0e1edbd9 NS |
1840 | if (!(flags & XFS_ALLOC_FLAG_FREEING)) { |
1841 | need = XFS_MIN_FREELIST_PAG(pag, mp); | |
1842 | delta = need > pag->pagf_flcount ? need - pag->pagf_flcount : 0; | |
1843 | /* | |
1844 | * If it looks like there isn't a long enough extent, or enough | |
1845 | * total blocks, reject it. | |
1846 | */ | |
1847 | longest = (pag->pagf_longest > delta) ? | |
1848 | (pag->pagf_longest - delta) : | |
1849 | (pag->pagf_flcount > 0 || pag->pagf_longest > 0); | |
1850 | if ((args->minlen + args->alignment + args->minalignslop - 1) > | |
1851 | longest || | |
1852 | ((int)(pag->pagf_freeblks + pag->pagf_flcount - | |
1853 | need - args->total) < (int)args->minleft)) { | |
1854 | if (agbp) | |
1855 | xfs_trans_brelse(tp, agbp); | |
1856 | args->agbp = NULL; | |
1857 | return 0; | |
1858 | } | |
1da177e4 | 1859 | } |
0e1edbd9 | 1860 | |
1da177e4 LT |
1861 | /* |
1862 | * Get the a.g. freespace buffer. | |
1863 | * Can fail if we're not blocking on locks, and it's held. | |
1864 | */ | |
1865 | if (agbp == NULL) { | |
1866 | if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags, | |
1867 | &agbp))) | |
1868 | return error; | |
1869 | if (agbp == NULL) { | |
0e1edbd9 NS |
1870 | ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK); |
1871 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
1da177e4 LT |
1872 | args->agbp = NULL; |
1873 | return 0; | |
1874 | } | |
1875 | } | |
1876 | /* | |
1877 | * Figure out how many blocks we should have in the freelist. | |
1878 | */ | |
1879 | agf = XFS_BUF_TO_AGF(agbp); | |
1880 | need = XFS_MIN_FREELIST(agf, mp); | |
1da177e4 LT |
1881 | /* |
1882 | * If there isn't enough total or single-extent, reject it. | |
1883 | */ | |
0e1edbd9 NS |
1884 | if (!(flags & XFS_ALLOC_FLAG_FREEING)) { |
1885 | delta = need > be32_to_cpu(agf->agf_flcount) ? | |
1886 | (need - be32_to_cpu(agf->agf_flcount)) : 0; | |
1887 | longest = be32_to_cpu(agf->agf_longest); | |
1888 | longest = (longest > delta) ? (longest - delta) : | |
1889 | (be32_to_cpu(agf->agf_flcount) > 0 || longest > 0); | |
1890 | if ((args->minlen + args->alignment + args->minalignslop - 1) > | |
1891 | longest || | |
1892 | ((int)(be32_to_cpu(agf->agf_freeblks) + | |
1893 | be32_to_cpu(agf->agf_flcount) - need - args->total) < | |
1894 | (int)args->minleft)) { | |
1895 | xfs_trans_brelse(tp, agbp); | |
1896 | args->agbp = NULL; | |
1897 | return 0; | |
1898 | } | |
1da177e4 LT |
1899 | } |
1900 | /* | |
1901 | * Make the freelist shorter if it's too long. | |
1902 | */ | |
16259e7d | 1903 | while (be32_to_cpu(agf->agf_flcount) > need) { |
1da177e4 LT |
1904 | xfs_buf_t *bp; |
1905 | ||
92821e2b DC |
1906 | error = xfs_alloc_get_freelist(tp, agbp, &bno, 0); |
1907 | if (error) | |
1da177e4 LT |
1908 | return error; |
1909 | if ((error = xfs_free_ag_extent(tp, agbp, args->agno, bno, 1, 1))) | |
1910 | return error; | |
1911 | bp = xfs_btree_get_bufs(mp, tp, args->agno, bno, 0); | |
1912 | xfs_trans_binval(tp, bp); | |
1913 | } | |
1914 | /* | |
1915 | * Initialize the args structure. | |
1916 | */ | |
1917 | targs.tp = tp; | |
1918 | targs.mp = mp; | |
1919 | targs.agbp = agbp; | |
1920 | targs.agno = args->agno; | |
1921 | targs.mod = targs.minleft = targs.wasdel = targs.userdata = | |
1922 | targs.minalignslop = 0; | |
1923 | targs.alignment = targs.minlen = targs.prod = targs.isfl = 1; | |
1924 | targs.type = XFS_ALLOCTYPE_THIS_AG; | |
1925 | targs.pag = pag; | |
1926 | if ((error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp))) | |
1927 | return error; | |
1928 | /* | |
1929 | * Make the freelist longer if it's too short. | |
1930 | */ | |
16259e7d | 1931 | while (be32_to_cpu(agf->agf_flcount) < need) { |
1da177e4 | 1932 | targs.agbno = 0; |
16259e7d | 1933 | targs.maxlen = need - be32_to_cpu(agf->agf_flcount); |
1da177e4 LT |
1934 | /* |
1935 | * Allocate as many blocks as possible at once. | |
1936 | */ | |
e63a3690 NS |
1937 | if ((error = xfs_alloc_ag_vextent(&targs))) { |
1938 | xfs_trans_brelse(tp, agflbp); | |
1da177e4 | 1939 | return error; |
e63a3690 | 1940 | } |
1da177e4 LT |
1941 | /* |
1942 | * Stop if we run out. Won't happen if callers are obeying | |
1943 | * the restrictions correctly. Can happen for free calls | |
1944 | * on a completely full ag. | |
1945 | */ | |
d210a28c | 1946 | if (targs.agbno == NULLAGBLOCK) { |
0e1edbd9 NS |
1947 | if (flags & XFS_ALLOC_FLAG_FREEING) |
1948 | break; | |
1949 | xfs_trans_brelse(tp, agflbp); | |
1950 | args->agbp = NULL; | |
1951 | return 0; | |
d210a28c | 1952 | } |
1da177e4 LT |
1953 | /* |
1954 | * Put each allocated block on the list. | |
1955 | */ | |
1956 | for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) { | |
92821e2b DC |
1957 | error = xfs_alloc_put_freelist(tp, agbp, |
1958 | agflbp, bno, 0); | |
1959 | if (error) | |
1da177e4 LT |
1960 | return error; |
1961 | } | |
1962 | } | |
e63a3690 | 1963 | xfs_trans_brelse(tp, agflbp); |
1da177e4 LT |
1964 | args->agbp = agbp; |
1965 | return 0; | |
1966 | } | |
1967 | ||
1968 | /* | |
1969 | * Get a block from the freelist. | |
1970 | * Returns with the buffer for the block gotten. | |
1971 | */ | |
1972 | int /* error */ | |
1973 | xfs_alloc_get_freelist( | |
1974 | xfs_trans_t *tp, /* transaction pointer */ | |
1975 | xfs_buf_t *agbp, /* buffer containing the agf structure */ | |
92821e2b DC |
1976 | xfs_agblock_t *bnop, /* block address retrieved from freelist */ |
1977 | int btreeblk) /* destination is a AGF btree */ | |
1da177e4 LT |
1978 | { |
1979 | xfs_agf_t *agf; /* a.g. freespace structure */ | |
1980 | xfs_agfl_t *agfl; /* a.g. freelist structure */ | |
1981 | xfs_buf_t *agflbp;/* buffer for a.g. freelist structure */ | |
1982 | xfs_agblock_t bno; /* block number returned */ | |
1983 | int error; | |
92821e2b | 1984 | int logflags; |
1da177e4 LT |
1985 | xfs_mount_t *mp; /* mount structure */ |
1986 | xfs_perag_t *pag; /* per allocation group data */ | |
1987 | ||
1988 | agf = XFS_BUF_TO_AGF(agbp); | |
1989 | /* | |
1990 | * Freelist is empty, give up. | |
1991 | */ | |
1992 | if (!agf->agf_flcount) { | |
1993 | *bnop = NULLAGBLOCK; | |
1994 | return 0; | |
1995 | } | |
1996 | /* | |
1997 | * Read the array of free blocks. | |
1998 | */ | |
1999 | mp = tp->t_mountp; | |
2000 | if ((error = xfs_alloc_read_agfl(mp, tp, | |
16259e7d | 2001 | be32_to_cpu(agf->agf_seqno), &agflbp))) |
1da177e4 LT |
2002 | return error; |
2003 | agfl = XFS_BUF_TO_AGFL(agflbp); | |
2004 | /* | |
2005 | * Get the block number and update the data structures. | |
2006 | */ | |
e2101005 | 2007 | bno = be32_to_cpu(agfl->agfl_bno[be32_to_cpu(agf->agf_flfirst)]); |
413d57c9 | 2008 | be32_add_cpu(&agf->agf_flfirst, 1); |
1da177e4 | 2009 | xfs_trans_brelse(tp, agflbp); |
16259e7d | 2010 | if (be32_to_cpu(agf->agf_flfirst) == XFS_AGFL_SIZE(mp)) |
1da177e4 | 2011 | agf->agf_flfirst = 0; |
16259e7d | 2012 | pag = &mp->m_perag[be32_to_cpu(agf->agf_seqno)]; |
413d57c9 | 2013 | be32_add_cpu(&agf->agf_flcount, -1); |
1da177e4 LT |
2014 | xfs_trans_agflist_delta(tp, -1); |
2015 | pag->pagf_flcount--; | |
92821e2b DC |
2016 | |
2017 | logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT; | |
2018 | if (btreeblk) { | |
413d57c9 | 2019 | be32_add_cpu(&agf->agf_btreeblks, 1); |
92821e2b DC |
2020 | pag->pagf_btreeblks++; |
2021 | logflags |= XFS_AGF_BTREEBLKS; | |
2022 | } | |
2023 | ||
2024 | TRACE_MODAGF(NULL, agf, logflags); | |
2025 | xfs_alloc_log_agf(tp, agbp, logflags); | |
1da177e4 LT |
2026 | *bnop = bno; |
2027 | ||
2028 | /* | |
2029 | * As blocks are freed, they are added to the per-ag busy list | |
2030 | * and remain there until the freeing transaction is committed to | |
2031 | * disk. Now that we have allocated blocks, this list must be | |
2032 | * searched to see if a block is being reused. If one is, then | |
2033 | * the freeing transaction must be pushed to disk NOW by forcing | |
2034 | * to disk all iclogs up that transaction's LSN. | |
2035 | */ | |
16259e7d | 2036 | xfs_alloc_search_busy(tp, be32_to_cpu(agf->agf_seqno), bno, 1); |
1da177e4 LT |
2037 | return 0; |
2038 | } | |
2039 | ||
2040 | /* | |
2041 | * Log the given fields from the agf structure. | |
2042 | */ | |
2043 | void | |
2044 | xfs_alloc_log_agf( | |
2045 | xfs_trans_t *tp, /* transaction pointer */ | |
2046 | xfs_buf_t *bp, /* buffer for a.g. freelist header */ | |
2047 | int fields) /* mask of fields to be logged (XFS_AGF_...) */ | |
2048 | { | |
2049 | int first; /* first byte offset */ | |
2050 | int last; /* last byte offset */ | |
2051 | static const short offsets[] = { | |
2052 | offsetof(xfs_agf_t, agf_magicnum), | |
2053 | offsetof(xfs_agf_t, agf_versionnum), | |
2054 | offsetof(xfs_agf_t, agf_seqno), | |
2055 | offsetof(xfs_agf_t, agf_length), | |
2056 | offsetof(xfs_agf_t, agf_roots[0]), | |
2057 | offsetof(xfs_agf_t, agf_levels[0]), | |
2058 | offsetof(xfs_agf_t, agf_flfirst), | |
2059 | offsetof(xfs_agf_t, agf_fllast), | |
2060 | offsetof(xfs_agf_t, agf_flcount), | |
2061 | offsetof(xfs_agf_t, agf_freeblks), | |
2062 | offsetof(xfs_agf_t, agf_longest), | |
92821e2b | 2063 | offsetof(xfs_agf_t, agf_btreeblks), |
1da177e4 LT |
2064 | sizeof(xfs_agf_t) |
2065 | }; | |
2066 | ||
2067 | xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last); | |
2068 | xfs_trans_log_buf(tp, bp, (uint)first, (uint)last); | |
2069 | } | |
2070 | ||
2071 | /* | |
2072 | * Interface for inode allocation to force the pag data to be initialized. | |
2073 | */ | |
2074 | int /* error */ | |
2075 | xfs_alloc_pagf_init( | |
2076 | xfs_mount_t *mp, /* file system mount structure */ | |
2077 | xfs_trans_t *tp, /* transaction pointer */ | |
2078 | xfs_agnumber_t agno, /* allocation group number */ | |
2079 | int flags) /* XFS_ALLOC_FLAGS_... */ | |
2080 | { | |
2081 | xfs_buf_t *bp; | |
2082 | int error; | |
2083 | ||
2084 | if ((error = xfs_alloc_read_agf(mp, tp, agno, flags, &bp))) | |
2085 | return error; | |
2086 | if (bp) | |
2087 | xfs_trans_brelse(tp, bp); | |
2088 | return 0; | |
2089 | } | |
2090 | ||
2091 | /* | |
2092 | * Put the block on the freelist for the allocation group. | |
2093 | */ | |
2094 | int /* error */ | |
2095 | xfs_alloc_put_freelist( | |
2096 | xfs_trans_t *tp, /* transaction pointer */ | |
2097 | xfs_buf_t *agbp, /* buffer for a.g. freelist header */ | |
2098 | xfs_buf_t *agflbp,/* buffer for a.g. free block array */ | |
92821e2b DC |
2099 | xfs_agblock_t bno, /* block being freed */ |
2100 | int btreeblk) /* block came from a AGF btree */ | |
1da177e4 LT |
2101 | { |
2102 | xfs_agf_t *agf; /* a.g. freespace structure */ | |
2103 | xfs_agfl_t *agfl; /* a.g. free block array */ | |
e2101005 | 2104 | __be32 *blockp;/* pointer to array entry */ |
1da177e4 | 2105 | int error; |
92821e2b | 2106 | int logflags; |
1da177e4 LT |
2107 | xfs_mount_t *mp; /* mount structure */ |
2108 | xfs_perag_t *pag; /* per allocation group data */ | |
2109 | ||
2110 | agf = XFS_BUF_TO_AGF(agbp); | |
2111 | mp = tp->t_mountp; | |
2112 | ||
2113 | if (!agflbp && (error = xfs_alloc_read_agfl(mp, tp, | |
16259e7d | 2114 | be32_to_cpu(agf->agf_seqno), &agflbp))) |
1da177e4 LT |
2115 | return error; |
2116 | agfl = XFS_BUF_TO_AGFL(agflbp); | |
413d57c9 | 2117 | be32_add_cpu(&agf->agf_fllast, 1); |
16259e7d | 2118 | if (be32_to_cpu(agf->agf_fllast) == XFS_AGFL_SIZE(mp)) |
1da177e4 | 2119 | agf->agf_fllast = 0; |
16259e7d | 2120 | pag = &mp->m_perag[be32_to_cpu(agf->agf_seqno)]; |
413d57c9 | 2121 | be32_add_cpu(&agf->agf_flcount, 1); |
1da177e4 LT |
2122 | xfs_trans_agflist_delta(tp, 1); |
2123 | pag->pagf_flcount++; | |
92821e2b DC |
2124 | |
2125 | logflags = XFS_AGF_FLLAST | XFS_AGF_FLCOUNT; | |
2126 | if (btreeblk) { | |
413d57c9 | 2127 | be32_add_cpu(&agf->agf_btreeblks, -1); |
92821e2b DC |
2128 | pag->pagf_btreeblks--; |
2129 | logflags |= XFS_AGF_BTREEBLKS; | |
2130 | } | |
2131 | ||
2132 | TRACE_MODAGF(NULL, agf, logflags); | |
2133 | xfs_alloc_log_agf(tp, agbp, logflags); | |
2134 | ||
16259e7d CH |
2135 | ASSERT(be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp)); |
2136 | blockp = &agfl->agfl_bno[be32_to_cpu(agf->agf_fllast)]; | |
e2101005 | 2137 | *blockp = cpu_to_be32(bno); |
92821e2b DC |
2138 | TRACE_MODAGF(NULL, agf, logflags); |
2139 | xfs_alloc_log_agf(tp, agbp, logflags); | |
1da177e4 LT |
2140 | xfs_trans_log_buf(tp, agflbp, |
2141 | (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl), | |
2142 | (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl + | |
2143 | sizeof(xfs_agblock_t) - 1)); | |
2144 | return 0; | |
2145 | } | |
2146 | ||
2147 | /* | |
2148 | * Read in the allocation group header (free/alloc section). | |
2149 | */ | |
2150 | int /* error */ | |
2151 | xfs_alloc_read_agf( | |
2152 | xfs_mount_t *mp, /* mount point structure */ | |
2153 | xfs_trans_t *tp, /* transaction pointer */ | |
2154 | xfs_agnumber_t agno, /* allocation group number */ | |
2155 | int flags, /* XFS_ALLOC_FLAG_... */ | |
2156 | xfs_buf_t **bpp) /* buffer for the ag freelist header */ | |
2157 | { | |
2158 | xfs_agf_t *agf; /* ag freelist header */ | |
2159 | int agf_ok; /* set if agf is consistent */ | |
2160 | xfs_buf_t *bp; /* return value */ | |
2161 | xfs_perag_t *pag; /* per allocation group data */ | |
2162 | int error; | |
2163 | ||
2164 | ASSERT(agno != NULLAGNUMBER); | |
2165 | error = xfs_trans_read_buf( | |
2166 | mp, tp, mp->m_ddev_targp, | |
2167 | XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)), | |
2168 | XFS_FSS_TO_BB(mp, 1), | |
2169 | (flags & XFS_ALLOC_FLAG_TRYLOCK) ? XFS_BUF_TRYLOCK : 0U, | |
2170 | &bp); | |
2171 | if (error) | |
2172 | return error; | |
2173 | ASSERT(!bp || !XFS_BUF_GETERROR(bp)); | |
2174 | if (!bp) { | |
2175 | *bpp = NULL; | |
2176 | return 0; | |
2177 | } | |
2178 | /* | |
2179 | * Validate the magic number of the agf block. | |
2180 | */ | |
2181 | agf = XFS_BUF_TO_AGF(bp); | |
2182 | agf_ok = | |
16259e7d CH |
2183 | be32_to_cpu(agf->agf_magicnum) == XFS_AGF_MAGIC && |
2184 | XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) && | |
2185 | be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) && | |
2186 | be32_to_cpu(agf->agf_flfirst) < XFS_AGFL_SIZE(mp) && | |
2187 | be32_to_cpu(agf->agf_fllast) < XFS_AGFL_SIZE(mp) && | |
2188 | be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp); | |
1da177e4 LT |
2189 | if (unlikely(XFS_TEST_ERROR(!agf_ok, mp, XFS_ERRTAG_ALLOC_READ_AGF, |
2190 | XFS_RANDOM_ALLOC_READ_AGF))) { | |
2191 | XFS_CORRUPTION_ERROR("xfs_alloc_read_agf", | |
2192 | XFS_ERRLEVEL_LOW, mp, agf); | |
2193 | xfs_trans_brelse(tp, bp); | |
2194 | return XFS_ERROR(EFSCORRUPTED); | |
2195 | } | |
2196 | pag = &mp->m_perag[agno]; | |
2197 | if (!pag->pagf_init) { | |
16259e7d | 2198 | pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks); |
92821e2b | 2199 | pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks); |
16259e7d CH |
2200 | pag->pagf_flcount = be32_to_cpu(agf->agf_flcount); |
2201 | pag->pagf_longest = be32_to_cpu(agf->agf_longest); | |
1da177e4 | 2202 | pag->pagf_levels[XFS_BTNUM_BNOi] = |
16259e7d | 2203 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]); |
1da177e4 | 2204 | pag->pagf_levels[XFS_BTNUM_CNTi] = |
16259e7d | 2205 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]); |
007c61c6 | 2206 | spin_lock_init(&pag->pagb_lock); |
1da177e4 LT |
2207 | pag->pagb_list = kmem_zalloc(XFS_PAGB_NUM_SLOTS * |
2208 | sizeof(xfs_perag_busy_t), KM_SLEEP); | |
2209 | pag->pagf_init = 1; | |
2210 | } | |
2211 | #ifdef DEBUG | |
2212 | else if (!XFS_FORCED_SHUTDOWN(mp)) { | |
16259e7d CH |
2213 | ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks)); |
2214 | ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount)); | |
2215 | ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest)); | |
1da177e4 | 2216 | ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] == |
16259e7d | 2217 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi])); |
1da177e4 | 2218 | ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] == |
16259e7d | 2219 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi])); |
1da177e4 LT |
2220 | } |
2221 | #endif | |
2222 | XFS_BUF_SET_VTYPE_REF(bp, B_FS_AGF, XFS_AGF_REF); | |
2223 | *bpp = bp; | |
2224 | return 0; | |
2225 | } | |
2226 | ||
2227 | /* | |
2228 | * Allocate an extent (variable-size). | |
2229 | * Depending on the allocation type, we either look in a single allocation | |
2230 | * group or loop over the allocation groups to find the result. | |
2231 | */ | |
2232 | int /* error */ | |
2233 | xfs_alloc_vextent( | |
2234 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
2235 | { | |
2236 | xfs_agblock_t agsize; /* allocation group size */ | |
2237 | int error; | |
2238 | int flags; /* XFS_ALLOC_FLAG_... locking flags */ | |
1da177e4 LT |
2239 | xfs_extlen_t minleft;/* minimum left value, temp copy */ |
2240 | xfs_mount_t *mp; /* mount structure pointer */ | |
2241 | xfs_agnumber_t sagno; /* starting allocation group number */ | |
2242 | xfs_alloctype_t type; /* input allocation type */ | |
2243 | int bump_rotor = 0; | |
2244 | int no_min = 0; | |
2245 | xfs_agnumber_t rotorstep = xfs_rotorstep; /* inode32 agf stepper */ | |
2246 | ||
2247 | mp = args->mp; | |
2248 | type = args->otype = args->type; | |
2249 | args->agbno = NULLAGBLOCK; | |
2250 | /* | |
2251 | * Just fix this up, for the case where the last a.g. is shorter | |
2252 | * (or there's only one a.g.) and the caller couldn't easily figure | |
2253 | * that out (xfs_bmap_alloc). | |
2254 | */ | |
2255 | agsize = mp->m_sb.sb_agblocks; | |
2256 | if (args->maxlen > agsize) | |
2257 | args->maxlen = agsize; | |
2258 | if (args->alignment == 0) | |
2259 | args->alignment = 1; | |
2260 | ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount); | |
2261 | ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize); | |
2262 | ASSERT(args->minlen <= args->maxlen); | |
2263 | ASSERT(args->minlen <= agsize); | |
2264 | ASSERT(args->mod < args->prod); | |
2265 | if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount || | |
2266 | XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize || | |
2267 | args->minlen > args->maxlen || args->minlen > agsize || | |
2268 | args->mod >= args->prod) { | |
2269 | args->fsbno = NULLFSBLOCK; | |
2270 | TRACE_ALLOC("badargs", args); | |
2271 | return 0; | |
2272 | } | |
2273 | minleft = args->minleft; | |
2274 | ||
2275 | switch (type) { | |
2276 | case XFS_ALLOCTYPE_THIS_AG: | |
2277 | case XFS_ALLOCTYPE_NEAR_BNO: | |
2278 | case XFS_ALLOCTYPE_THIS_BNO: | |
2279 | /* | |
2280 | * These three force us into a single a.g. | |
2281 | */ | |
2282 | args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
2283 | down_read(&mp->m_peraglock); | |
2284 | args->pag = &mp->m_perag[args->agno]; | |
2285 | args->minleft = 0; | |
2286 | error = xfs_alloc_fix_freelist(args, 0); | |
2287 | args->minleft = minleft; | |
2288 | if (error) { | |
2289 | TRACE_ALLOC("nofix", args); | |
2290 | goto error0; | |
2291 | } | |
2292 | if (!args->agbp) { | |
2293 | up_read(&mp->m_peraglock); | |
2294 | TRACE_ALLOC("noagbp", args); | |
2295 | break; | |
2296 | } | |
2297 | args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno); | |
2298 | if ((error = xfs_alloc_ag_vextent(args))) | |
2299 | goto error0; | |
2300 | up_read(&mp->m_peraglock); | |
2301 | break; | |
2302 | case XFS_ALLOCTYPE_START_BNO: | |
2303 | /* | |
2304 | * Try near allocation first, then anywhere-in-ag after | |
2305 | * the first a.g. fails. | |
2306 | */ | |
2307 | if ((args->userdata == XFS_ALLOC_INITIAL_USER_DATA) && | |
2308 | (mp->m_flags & XFS_MOUNT_32BITINODES)) { | |
2309 | args->fsbno = XFS_AGB_TO_FSB(mp, | |
2310 | ((mp->m_agfrotor / rotorstep) % | |
2311 | mp->m_sb.sb_agcount), 0); | |
2312 | bump_rotor = 1; | |
2313 | } | |
2314 | args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno); | |
2315 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
2316 | /* FALLTHROUGH */ | |
2317 | case XFS_ALLOCTYPE_ANY_AG: | |
2318 | case XFS_ALLOCTYPE_START_AG: | |
2319 | case XFS_ALLOCTYPE_FIRST_AG: | |
2320 | /* | |
2321 | * Rotate through the allocation groups looking for a winner. | |
2322 | */ | |
2323 | if (type == XFS_ALLOCTYPE_ANY_AG) { | |
2324 | /* | |
2325 | * Start with the last place we left off. | |
2326 | */ | |
2327 | args->agno = sagno = (mp->m_agfrotor / rotorstep) % | |
2328 | mp->m_sb.sb_agcount; | |
2329 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2330 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
2331 | } else if (type == XFS_ALLOCTYPE_FIRST_AG) { | |
2332 | /* | |
2333 | * Start with allocation group given by bno. | |
2334 | */ | |
2335 | args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
2336 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2337 | sagno = 0; | |
2338 | flags = 0; | |
2339 | } else { | |
2340 | if (type == XFS_ALLOCTYPE_START_AG) | |
2341 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2342 | /* | |
2343 | * Start with the given allocation group. | |
2344 | */ | |
2345 | args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
2346 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
2347 | } | |
2348 | /* | |
2349 | * Loop over allocation groups twice; first time with | |
2350 | * trylock set, second time without. | |
2351 | */ | |
2352 | down_read(&mp->m_peraglock); | |
2353 | for (;;) { | |
2354 | args->pag = &mp->m_perag[args->agno]; | |
2355 | if (no_min) args->minleft = 0; | |
2356 | error = xfs_alloc_fix_freelist(args, flags); | |
2357 | args->minleft = minleft; | |
2358 | if (error) { | |
2359 | TRACE_ALLOC("nofix", args); | |
2360 | goto error0; | |
2361 | } | |
2362 | /* | |
2363 | * If we get a buffer back then the allocation will fly. | |
2364 | */ | |
2365 | if (args->agbp) { | |
2366 | if ((error = xfs_alloc_ag_vextent(args))) | |
2367 | goto error0; | |
2368 | break; | |
2369 | } | |
2370 | TRACE_ALLOC("loopfailed", args); | |
2371 | /* | |
2372 | * Didn't work, figure out the next iteration. | |
2373 | */ | |
2374 | if (args->agno == sagno && | |
2375 | type == XFS_ALLOCTYPE_START_BNO) | |
2376 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
d210a28c YL |
2377 | /* |
2378 | * For the first allocation, we can try any AG to get | |
2379 | * space. However, if we already have allocated a | |
2380 | * block, we don't want to try AGs whose number is below | |
2381 | * sagno. Otherwise, we may end up with out-of-order | |
2382 | * locking of AGF, which might cause deadlock. | |
2383 | */ | |
2384 | if (++(args->agno) == mp->m_sb.sb_agcount) { | |
2385 | if (args->firstblock != NULLFSBLOCK) | |
2386 | args->agno = sagno; | |
2387 | else | |
2388 | args->agno = 0; | |
2389 | } | |
1da177e4 LT |
2390 | /* |
2391 | * Reached the starting a.g., must either be done | |
2392 | * or switch to non-trylock mode. | |
2393 | */ | |
2394 | if (args->agno == sagno) { | |
2395 | if (no_min == 1) { | |
2396 | args->agbno = NULLAGBLOCK; | |
2397 | TRACE_ALLOC("allfailed", args); | |
2398 | break; | |
2399 | } | |
2400 | if (flags == 0) { | |
2401 | no_min = 1; | |
2402 | } else { | |
2403 | flags = 0; | |
2404 | if (type == XFS_ALLOCTYPE_START_BNO) { | |
2405 | args->agbno = XFS_FSB_TO_AGBNO(mp, | |
2406 | args->fsbno); | |
2407 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
2408 | } | |
2409 | } | |
2410 | } | |
2411 | } | |
2412 | up_read(&mp->m_peraglock); | |
2413 | if (bump_rotor || (type == XFS_ALLOCTYPE_ANY_AG)) { | |
2414 | if (args->agno == sagno) | |
2415 | mp->m_agfrotor = (mp->m_agfrotor + 1) % | |
2416 | (mp->m_sb.sb_agcount * rotorstep); | |
2417 | else | |
2418 | mp->m_agfrotor = (args->agno * rotorstep + 1) % | |
2419 | (mp->m_sb.sb_agcount * rotorstep); | |
2420 | } | |
2421 | break; | |
2422 | default: | |
2423 | ASSERT(0); | |
2424 | /* NOTREACHED */ | |
2425 | } | |
2426 | if (args->agbno == NULLAGBLOCK) | |
2427 | args->fsbno = NULLFSBLOCK; | |
2428 | else { | |
2429 | args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno); | |
2430 | #ifdef DEBUG | |
2431 | ASSERT(args->len >= args->minlen); | |
2432 | ASSERT(args->len <= args->maxlen); | |
2433 | ASSERT(args->agbno % args->alignment == 0); | |
2434 | XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno), | |
2435 | args->len); | |
2436 | #endif | |
2437 | } | |
2438 | return 0; | |
2439 | error0: | |
2440 | up_read(&mp->m_peraglock); | |
2441 | return error; | |
2442 | } | |
2443 | ||
2444 | /* | |
2445 | * Free an extent. | |
2446 | * Just break up the extent address and hand off to xfs_free_ag_extent | |
2447 | * after fixing up the freelist. | |
2448 | */ | |
2449 | int /* error */ | |
2450 | xfs_free_extent( | |
2451 | xfs_trans_t *tp, /* transaction pointer */ | |
2452 | xfs_fsblock_t bno, /* starting block number of extent */ | |
2453 | xfs_extlen_t len) /* length of extent */ | |
2454 | { | |
0e1edbd9 | 2455 | xfs_alloc_arg_t args; |
1da177e4 LT |
2456 | int error; |
2457 | ||
2458 | ASSERT(len != 0); | |
0e1edbd9 | 2459 | memset(&args, 0, sizeof(xfs_alloc_arg_t)); |
1da177e4 LT |
2460 | args.tp = tp; |
2461 | args.mp = tp->t_mountp; | |
2462 | args.agno = XFS_FSB_TO_AGNO(args.mp, bno); | |
2463 | ASSERT(args.agno < args.mp->m_sb.sb_agcount); | |
2464 | args.agbno = XFS_FSB_TO_AGBNO(args.mp, bno); | |
1da177e4 LT |
2465 | down_read(&args.mp->m_peraglock); |
2466 | args.pag = &args.mp->m_perag[args.agno]; | |
d210a28c | 2467 | if ((error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING))) |
1da177e4 LT |
2468 | goto error0; |
2469 | #ifdef DEBUG | |
2470 | ASSERT(args.agbp != NULL); | |
0e1edbd9 NS |
2471 | ASSERT((args.agbno + len) <= |
2472 | be32_to_cpu(XFS_BUF_TO_AGF(args.agbp)->agf_length)); | |
1da177e4 | 2473 | #endif |
0e1edbd9 | 2474 | error = xfs_free_ag_extent(tp, args.agbp, args.agno, args.agbno, len, 0); |
1da177e4 LT |
2475 | error0: |
2476 | up_read(&args.mp->m_peraglock); | |
2477 | return error; | |
2478 | } | |
2479 | ||
2480 | ||
2481 | /* | |
2482 | * AG Busy list management | |
2483 | * The busy list contains block ranges that have been freed but whose | |
c41564b5 | 2484 | * transactions have not yet hit disk. If any block listed in a busy |
1da177e4 LT |
2485 | * list is reused, the transaction that freed it must be forced to disk |
2486 | * before continuing to use the block. | |
2487 | * | |
2488 | * xfs_alloc_mark_busy - add to the per-ag busy list | |
2489 | * xfs_alloc_clear_busy - remove an item from the per-ag busy list | |
2490 | */ | |
2491 | void | |
2492 | xfs_alloc_mark_busy(xfs_trans_t *tp, | |
2493 | xfs_agnumber_t agno, | |
2494 | xfs_agblock_t bno, | |
2495 | xfs_extlen_t len) | |
2496 | { | |
2497 | xfs_mount_t *mp; | |
2498 | xfs_perag_busy_t *bsy; | |
2499 | int n; | |
1da177e4 LT |
2500 | |
2501 | mp = tp->t_mountp; | |
64137e56 | 2502 | spin_lock(&mp->m_perag[agno].pagb_lock); |
1da177e4 LT |
2503 | |
2504 | /* search pagb_list for an open slot */ | |
2505 | for (bsy = mp->m_perag[agno].pagb_list, n = 0; | |
2506 | n < XFS_PAGB_NUM_SLOTS; | |
2507 | bsy++, n++) { | |
2508 | if (bsy->busy_tp == NULL) { | |
2509 | break; | |
2510 | } | |
2511 | } | |
2512 | ||
2513 | if (n < XFS_PAGB_NUM_SLOTS) { | |
2514 | bsy = &mp->m_perag[agno].pagb_list[n]; | |
2515 | mp->m_perag[agno].pagb_count++; | |
2516 | TRACE_BUSY("xfs_alloc_mark_busy", "got", agno, bno, len, n, tp); | |
2517 | bsy->busy_start = bno; | |
2518 | bsy->busy_length = len; | |
2519 | bsy->busy_tp = tp; | |
2520 | xfs_trans_add_busy(tp, agno, n); | |
2521 | } else { | |
2522 | TRACE_BUSY("xfs_alloc_mark_busy", "FULL", agno, bno, len, -1, tp); | |
2523 | /* | |
2524 | * The busy list is full! Since it is now not possible to | |
2525 | * track the free block, make this a synchronous transaction | |
2526 | * to insure that the block is not reused before this | |
2527 | * transaction commits. | |
2528 | */ | |
2529 | xfs_trans_set_sync(tp); | |
2530 | } | |
2531 | ||
64137e56 | 2532 | spin_unlock(&mp->m_perag[agno].pagb_lock); |
1da177e4 LT |
2533 | } |
2534 | ||
2535 | void | |
2536 | xfs_alloc_clear_busy(xfs_trans_t *tp, | |
2537 | xfs_agnumber_t agno, | |
2538 | int idx) | |
2539 | { | |
2540 | xfs_mount_t *mp; | |
2541 | xfs_perag_busy_t *list; | |
1da177e4 LT |
2542 | |
2543 | mp = tp->t_mountp; | |
2544 | ||
64137e56 | 2545 | spin_lock(&mp->m_perag[agno].pagb_lock); |
1da177e4 LT |
2546 | list = mp->m_perag[agno].pagb_list; |
2547 | ||
2548 | ASSERT(idx < XFS_PAGB_NUM_SLOTS); | |
2549 | if (list[idx].busy_tp == tp) { | |
2550 | TRACE_UNBUSY("xfs_alloc_clear_busy", "found", agno, idx, tp); | |
2551 | list[idx].busy_tp = NULL; | |
2552 | mp->m_perag[agno].pagb_count--; | |
2553 | } else { | |
2554 | TRACE_UNBUSY("xfs_alloc_clear_busy", "missing", agno, idx, tp); | |
2555 | } | |
2556 | ||
64137e56 | 2557 | spin_unlock(&mp->m_perag[agno].pagb_lock); |
1da177e4 LT |
2558 | } |
2559 | ||
2560 | ||
2561 | /* | |
f4586e40 DC |
2562 | * If we find the extent in the busy list, force the log out to get the |
2563 | * extent out of the busy list so the caller can use it straight away. | |
1da177e4 | 2564 | */ |
f4586e40 | 2565 | STATIC void |
1da177e4 LT |
2566 | xfs_alloc_search_busy(xfs_trans_t *tp, |
2567 | xfs_agnumber_t agno, | |
2568 | xfs_agblock_t bno, | |
2569 | xfs_extlen_t len) | |
2570 | { | |
2571 | xfs_mount_t *mp; | |
2572 | xfs_perag_busy_t *bsy; | |
1da177e4 LT |
2573 | xfs_agblock_t uend, bend; |
2574 | xfs_lsn_t lsn; | |
2575 | int cnt; | |
1da177e4 LT |
2576 | |
2577 | mp = tp->t_mountp; | |
2578 | ||
64137e56 | 2579 | spin_lock(&mp->m_perag[agno].pagb_lock); |
1da177e4 LT |
2580 | cnt = mp->m_perag[agno].pagb_count; |
2581 | ||
2582 | uend = bno + len - 1; | |
2583 | ||
2584 | /* search pagb_list for this slot, skipping open slots */ | |
f4586e40 | 2585 | for (bsy = mp->m_perag[agno].pagb_list; cnt; bsy++) { |
1da177e4 LT |
2586 | |
2587 | /* | |
2588 | * (start1,length1) within (start2, length2) | |
2589 | */ | |
2590 | if (bsy->busy_tp != NULL) { | |
2591 | bend = bsy->busy_start + bsy->busy_length - 1; | |
f4586e40 | 2592 | if ((bno > bend) || (uend < bsy->busy_start)) { |
1da177e4 LT |
2593 | cnt--; |
2594 | } else { | |
2595 | TRACE_BUSYSEARCH("xfs_alloc_search_busy", | |
f4586e40 | 2596 | "found1", agno, bno, len, tp); |
1da177e4 LT |
2597 | break; |
2598 | } | |
2599 | } | |
2600 | } | |
2601 | ||
2602 | /* | |
2603 | * If a block was found, force the log through the LSN of the | |
2604 | * transaction that freed the block | |
2605 | */ | |
2606 | if (cnt) { | |
f4586e40 | 2607 | TRACE_BUSYSEARCH("xfs_alloc_search_busy", "found", agno, bno, len, tp); |
1da177e4 | 2608 | lsn = bsy->busy_tp->t_commit_lsn; |
64137e56 | 2609 | spin_unlock(&mp->m_perag[agno].pagb_lock); |
1da177e4 LT |
2610 | xfs_log_force(mp, lsn, XFS_LOG_FORCE|XFS_LOG_SYNC); |
2611 | } else { | |
f4586e40 | 2612 | TRACE_BUSYSEARCH("xfs_alloc_search_busy", "not-found", agno, bno, len, tp); |
64137e56 | 2613 | spin_unlock(&mp->m_perag[agno].pagb_lock); |
1da177e4 | 2614 | } |
1da177e4 | 2615 | } |