]>
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 | ||
fe033cc8 CH |
92 | /* |
93 | * Lookup the record equal to [bno, len] in the btree given by cur. | |
94 | */ | |
95 | STATIC int /* error */ | |
96 | xfs_alloc_lookup_eq( | |
97 | struct xfs_btree_cur *cur, /* btree cursor */ | |
98 | xfs_agblock_t bno, /* starting block of extent */ | |
99 | xfs_extlen_t len, /* length of extent */ | |
100 | int *stat) /* success/failure */ | |
101 | { | |
102 | cur->bc_rec.a.ar_startblock = bno; | |
103 | cur->bc_rec.a.ar_blockcount = len; | |
104 | return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat); | |
105 | } | |
106 | ||
107 | /* | |
108 | * Lookup the first record greater than or equal to [bno, len] | |
109 | * in the btree given by cur. | |
110 | */ | |
111 | STATIC int /* error */ | |
112 | xfs_alloc_lookup_ge( | |
113 | struct xfs_btree_cur *cur, /* btree cursor */ | |
114 | xfs_agblock_t bno, /* starting block of extent */ | |
115 | xfs_extlen_t len, /* length of extent */ | |
116 | int *stat) /* success/failure */ | |
117 | { | |
118 | cur->bc_rec.a.ar_startblock = bno; | |
119 | cur->bc_rec.a.ar_blockcount = len; | |
120 | return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat); | |
121 | } | |
122 | ||
123 | /* | |
124 | * Lookup the first record less than or equal to [bno, len] | |
125 | * in the btree given by cur. | |
126 | */ | |
127 | STATIC int /* error */ | |
128 | xfs_alloc_lookup_le( | |
129 | struct xfs_btree_cur *cur, /* btree cursor */ | |
130 | xfs_agblock_t bno, /* starting block of extent */ | |
131 | xfs_extlen_t len, /* length of extent */ | |
132 | int *stat) /* success/failure */ | |
133 | { | |
134 | cur->bc_rec.a.ar_startblock = bno; | |
135 | cur->bc_rec.a.ar_blockcount = len; | |
136 | return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat); | |
137 | } | |
138 | ||
278d0ca1 CH |
139 | /* |
140 | * Update the record referred to by cur to the value given | |
141 | * by [bno, len]. | |
142 | * This either works (return 0) or gets an EFSCORRUPTED error. | |
143 | */ | |
144 | STATIC int /* error */ | |
145 | xfs_alloc_update( | |
146 | struct xfs_btree_cur *cur, /* btree cursor */ | |
147 | xfs_agblock_t bno, /* starting block of extent */ | |
148 | xfs_extlen_t len) /* length of extent */ | |
149 | { | |
150 | union xfs_btree_rec rec; | |
151 | ||
152 | rec.alloc.ar_startblock = cpu_to_be32(bno); | |
153 | rec.alloc.ar_blockcount = cpu_to_be32(len); | |
154 | return xfs_btree_update(cur, &rec); | |
155 | } | |
fe033cc8 | 156 | |
8cc938fe CH |
157 | /* |
158 | * Get the data from the pointed-to record. | |
159 | */ | |
160 | STATIC int /* error */ | |
161 | xfs_alloc_get_rec( | |
162 | struct xfs_btree_cur *cur, /* btree cursor */ | |
163 | xfs_agblock_t *bno, /* output: starting block of extent */ | |
164 | xfs_extlen_t *len, /* output: length of extent */ | |
165 | int *stat) /* output: success/failure */ | |
166 | { | |
167 | union xfs_btree_rec *rec; | |
168 | int error; | |
169 | ||
170 | error = xfs_btree_get_rec(cur, &rec, stat); | |
171 | if (!error && *stat == 1) { | |
172 | *bno = be32_to_cpu(rec->alloc.ar_startblock); | |
173 | *len = be32_to_cpu(rec->alloc.ar_blockcount); | |
174 | } | |
175 | return error; | |
176 | } | |
177 | ||
1da177e4 LT |
178 | /* |
179 | * Compute aligned version of the found extent. | |
180 | * Takes alignment and min length into account. | |
181 | */ | |
12375c82 | 182 | STATIC void |
1da177e4 LT |
183 | xfs_alloc_compute_aligned( |
184 | xfs_agblock_t foundbno, /* starting block in found extent */ | |
185 | xfs_extlen_t foundlen, /* length in found extent */ | |
186 | xfs_extlen_t alignment, /* alignment for allocation */ | |
187 | xfs_extlen_t minlen, /* minimum length for allocation */ | |
188 | xfs_agblock_t *resbno, /* result block number */ | |
189 | xfs_extlen_t *reslen) /* result length */ | |
190 | { | |
191 | xfs_agblock_t bno; | |
192 | xfs_extlen_t diff; | |
193 | xfs_extlen_t len; | |
194 | ||
195 | if (alignment > 1 && foundlen >= minlen) { | |
196 | bno = roundup(foundbno, alignment); | |
197 | diff = bno - foundbno; | |
198 | len = diff >= foundlen ? 0 : foundlen - diff; | |
199 | } else { | |
200 | bno = foundbno; | |
201 | len = foundlen; | |
202 | } | |
203 | *resbno = bno; | |
204 | *reslen = len; | |
1da177e4 LT |
205 | } |
206 | ||
207 | /* | |
208 | * Compute best start block and diff for "near" allocations. | |
209 | * freelen >= wantlen already checked by caller. | |
210 | */ | |
211 | STATIC xfs_extlen_t /* difference value (absolute) */ | |
212 | xfs_alloc_compute_diff( | |
213 | xfs_agblock_t wantbno, /* target starting block */ | |
214 | xfs_extlen_t wantlen, /* target length */ | |
215 | xfs_extlen_t alignment, /* target alignment */ | |
216 | xfs_agblock_t freebno, /* freespace's starting block */ | |
217 | xfs_extlen_t freelen, /* freespace's length */ | |
218 | xfs_agblock_t *newbnop) /* result: best start block from free */ | |
219 | { | |
220 | xfs_agblock_t freeend; /* end of freespace extent */ | |
221 | xfs_agblock_t newbno1; /* return block number */ | |
222 | xfs_agblock_t newbno2; /* other new block number */ | |
223 | xfs_extlen_t newlen1=0; /* length with newbno1 */ | |
224 | xfs_extlen_t newlen2=0; /* length with newbno2 */ | |
225 | xfs_agblock_t wantend; /* end of target extent */ | |
226 | ||
227 | ASSERT(freelen >= wantlen); | |
228 | freeend = freebno + freelen; | |
229 | wantend = wantbno + wantlen; | |
230 | if (freebno >= wantbno) { | |
231 | if ((newbno1 = roundup(freebno, alignment)) >= freeend) | |
232 | newbno1 = NULLAGBLOCK; | |
233 | } else if (freeend >= wantend && alignment > 1) { | |
234 | newbno1 = roundup(wantbno, alignment); | |
235 | newbno2 = newbno1 - alignment; | |
236 | if (newbno1 >= freeend) | |
237 | newbno1 = NULLAGBLOCK; | |
238 | else | |
239 | newlen1 = XFS_EXTLEN_MIN(wantlen, freeend - newbno1); | |
240 | if (newbno2 < freebno) | |
241 | newbno2 = NULLAGBLOCK; | |
242 | else | |
243 | newlen2 = XFS_EXTLEN_MIN(wantlen, freeend - newbno2); | |
244 | if (newbno1 != NULLAGBLOCK && newbno2 != NULLAGBLOCK) { | |
245 | if (newlen1 < newlen2 || | |
246 | (newlen1 == newlen2 && | |
247 | XFS_ABSDIFF(newbno1, wantbno) > | |
248 | XFS_ABSDIFF(newbno2, wantbno))) | |
249 | newbno1 = newbno2; | |
250 | } else if (newbno2 != NULLAGBLOCK) | |
251 | newbno1 = newbno2; | |
252 | } else if (freeend >= wantend) { | |
253 | newbno1 = wantbno; | |
254 | } else if (alignment > 1) { | |
255 | newbno1 = roundup(freeend - wantlen, alignment); | |
256 | if (newbno1 > freeend - wantlen && | |
257 | newbno1 - alignment >= freebno) | |
258 | newbno1 -= alignment; | |
259 | else if (newbno1 >= freeend) | |
260 | newbno1 = NULLAGBLOCK; | |
261 | } else | |
262 | newbno1 = freeend - wantlen; | |
263 | *newbnop = newbno1; | |
264 | return newbno1 == NULLAGBLOCK ? 0 : XFS_ABSDIFF(newbno1, wantbno); | |
265 | } | |
266 | ||
267 | /* | |
268 | * Fix up the length, based on mod and prod. | |
269 | * len should be k * prod + mod for some k. | |
270 | * If len is too small it is returned unchanged. | |
271 | * If len hits maxlen it is left alone. | |
272 | */ | |
273 | STATIC void | |
274 | xfs_alloc_fix_len( | |
275 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
276 | { | |
277 | xfs_extlen_t k; | |
278 | xfs_extlen_t rlen; | |
279 | ||
280 | ASSERT(args->mod < args->prod); | |
281 | rlen = args->len; | |
282 | ASSERT(rlen >= args->minlen); | |
283 | ASSERT(rlen <= args->maxlen); | |
284 | if (args->prod <= 1 || rlen < args->mod || rlen == args->maxlen || | |
285 | (args->mod == 0 && rlen < args->prod)) | |
286 | return; | |
287 | k = rlen % args->prod; | |
288 | if (k == args->mod) | |
289 | return; | |
290 | if (k > args->mod) { | |
291 | if ((int)(rlen = rlen - k - args->mod) < (int)args->minlen) | |
292 | return; | |
293 | } else { | |
294 | if ((int)(rlen = rlen - args->prod - (args->mod - k)) < | |
295 | (int)args->minlen) | |
296 | return; | |
297 | } | |
298 | ASSERT(rlen >= args->minlen); | |
299 | ASSERT(rlen <= args->maxlen); | |
300 | args->len = rlen; | |
301 | } | |
302 | ||
303 | /* | |
304 | * Fix up length if there is too little space left in the a.g. | |
305 | * Return 1 if ok, 0 if too little, should give up. | |
306 | */ | |
307 | STATIC int | |
308 | xfs_alloc_fix_minleft( | |
309 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
310 | { | |
311 | xfs_agf_t *agf; /* a.g. freelist header */ | |
312 | int diff; /* free space difference */ | |
313 | ||
314 | if (args->minleft == 0) | |
315 | return 1; | |
316 | agf = XFS_BUF_TO_AGF(args->agbp); | |
16259e7d CH |
317 | diff = be32_to_cpu(agf->agf_freeblks) |
318 | + be32_to_cpu(agf->agf_flcount) | |
1da177e4 LT |
319 | - args->len - args->minleft; |
320 | if (diff >= 0) | |
321 | return 1; | |
322 | args->len += diff; /* shrink the allocated space */ | |
323 | if (args->len >= args->minlen) | |
324 | return 1; | |
325 | args->agbno = NULLAGBLOCK; | |
326 | return 0; | |
327 | } | |
328 | ||
329 | /* | |
330 | * Update the two btrees, logically removing from freespace the extent | |
331 | * starting at rbno, rlen blocks. The extent is contained within the | |
332 | * actual (current) free extent fbno for flen blocks. | |
333 | * Flags are passed in indicating whether the cursors are set to the | |
334 | * relevant records. | |
335 | */ | |
336 | STATIC int /* error code */ | |
337 | xfs_alloc_fixup_trees( | |
338 | xfs_btree_cur_t *cnt_cur, /* cursor for by-size btree */ | |
339 | xfs_btree_cur_t *bno_cur, /* cursor for by-block btree */ | |
340 | xfs_agblock_t fbno, /* starting block of free extent */ | |
341 | xfs_extlen_t flen, /* length of free extent */ | |
342 | xfs_agblock_t rbno, /* starting block of returned extent */ | |
343 | xfs_extlen_t rlen, /* length of returned extent */ | |
344 | int flags) /* flags, XFSA_FIXUP_... */ | |
345 | { | |
346 | int error; /* error code */ | |
347 | int i; /* operation results */ | |
348 | xfs_agblock_t nfbno1; /* first new free startblock */ | |
349 | xfs_agblock_t nfbno2; /* second new free startblock */ | |
350 | xfs_extlen_t nflen1=0; /* first new free length */ | |
351 | xfs_extlen_t nflen2=0; /* second new free length */ | |
352 | ||
353 | /* | |
354 | * Look up the record in the by-size tree if necessary. | |
355 | */ | |
356 | if (flags & XFSA_FIXUP_CNT_OK) { | |
357 | #ifdef DEBUG | |
358 | if ((error = xfs_alloc_get_rec(cnt_cur, &nfbno1, &nflen1, &i))) | |
359 | return error; | |
360 | XFS_WANT_CORRUPTED_RETURN( | |
361 | i == 1 && nfbno1 == fbno && nflen1 == flen); | |
362 | #endif | |
363 | } else { | |
364 | if ((error = xfs_alloc_lookup_eq(cnt_cur, fbno, flen, &i))) | |
365 | return error; | |
366 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
367 | } | |
368 | /* | |
369 | * Look up the record in the by-block tree if necessary. | |
370 | */ | |
371 | if (flags & XFSA_FIXUP_BNO_OK) { | |
372 | #ifdef DEBUG | |
373 | if ((error = xfs_alloc_get_rec(bno_cur, &nfbno1, &nflen1, &i))) | |
374 | return error; | |
375 | XFS_WANT_CORRUPTED_RETURN( | |
376 | i == 1 && nfbno1 == fbno && nflen1 == flen); | |
377 | #endif | |
378 | } else { | |
379 | if ((error = xfs_alloc_lookup_eq(bno_cur, fbno, flen, &i))) | |
380 | return error; | |
381 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
382 | } | |
383 | #ifdef DEBUG | |
384 | { | |
385 | xfs_alloc_block_t *bnoblock; | |
386 | xfs_alloc_block_t *cntblock; | |
387 | ||
388 | if (bno_cur->bc_nlevels == 1 && | |
389 | cnt_cur->bc_nlevels == 1) { | |
390 | bnoblock = XFS_BUF_TO_ALLOC_BLOCK(bno_cur->bc_bufs[0]); | |
391 | cntblock = XFS_BUF_TO_ALLOC_BLOCK(cnt_cur->bc_bufs[0]); | |
392 | XFS_WANT_CORRUPTED_RETURN( | |
16259e7d CH |
393 | be16_to_cpu(bnoblock->bb_numrecs) == |
394 | be16_to_cpu(cntblock->bb_numrecs)); | |
1da177e4 LT |
395 | } |
396 | } | |
397 | #endif | |
398 | /* | |
399 | * Deal with all four cases: the allocated record is contained | |
400 | * within the freespace record, so we can have new freespace | |
401 | * at either (or both) end, or no freespace remaining. | |
402 | */ | |
403 | if (rbno == fbno && rlen == flen) | |
404 | nfbno1 = nfbno2 = NULLAGBLOCK; | |
405 | else if (rbno == fbno) { | |
406 | nfbno1 = rbno + rlen; | |
407 | nflen1 = flen - rlen; | |
408 | nfbno2 = NULLAGBLOCK; | |
409 | } else if (rbno + rlen == fbno + flen) { | |
410 | nfbno1 = fbno; | |
411 | nflen1 = flen - rlen; | |
412 | nfbno2 = NULLAGBLOCK; | |
413 | } else { | |
414 | nfbno1 = fbno; | |
415 | nflen1 = rbno - fbno; | |
416 | nfbno2 = rbno + rlen; | |
417 | nflen2 = (fbno + flen) - nfbno2; | |
418 | } | |
419 | /* | |
420 | * Delete the entry from the by-size btree. | |
421 | */ | |
91cca5df | 422 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 LT |
423 | return error; |
424 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
425 | /* | |
426 | * Add new by-size btree entry(s). | |
427 | */ | |
428 | if (nfbno1 != NULLAGBLOCK) { | |
429 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno1, nflen1, &i))) | |
430 | return error; | |
431 | XFS_WANT_CORRUPTED_RETURN(i == 0); | |
4b22a571 | 432 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
1da177e4 LT |
433 | return error; |
434 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
435 | } | |
436 | if (nfbno2 != NULLAGBLOCK) { | |
437 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno2, nflen2, &i))) | |
438 | return error; | |
439 | XFS_WANT_CORRUPTED_RETURN(i == 0); | |
4b22a571 | 440 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
1da177e4 LT |
441 | return error; |
442 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
443 | } | |
444 | /* | |
445 | * Fix up the by-block btree entry(s). | |
446 | */ | |
447 | if (nfbno1 == NULLAGBLOCK) { | |
448 | /* | |
449 | * No remaining freespace, just delete the by-block tree entry. | |
450 | */ | |
91cca5df | 451 | if ((error = xfs_btree_delete(bno_cur, &i))) |
1da177e4 LT |
452 | return error; |
453 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
454 | } else { | |
455 | /* | |
456 | * Update the by-block entry to start later|be shorter. | |
457 | */ | |
458 | if ((error = xfs_alloc_update(bno_cur, nfbno1, nflen1))) | |
459 | return error; | |
460 | } | |
461 | if (nfbno2 != NULLAGBLOCK) { | |
462 | /* | |
463 | * 2 resulting free entries, need to add one. | |
464 | */ | |
465 | if ((error = xfs_alloc_lookup_eq(bno_cur, nfbno2, nflen2, &i))) | |
466 | return error; | |
467 | XFS_WANT_CORRUPTED_RETURN(i == 0); | |
4b22a571 | 468 | if ((error = xfs_btree_insert(bno_cur, &i))) |
1da177e4 LT |
469 | return error; |
470 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
471 | } | |
472 | return 0; | |
473 | } | |
474 | ||
475 | /* | |
476 | * Read in the allocation group free block array. | |
477 | */ | |
478 | STATIC int /* error */ | |
479 | xfs_alloc_read_agfl( | |
480 | xfs_mount_t *mp, /* mount point structure */ | |
481 | xfs_trans_t *tp, /* transaction pointer */ | |
482 | xfs_agnumber_t agno, /* allocation group number */ | |
483 | xfs_buf_t **bpp) /* buffer for the ag free block array */ | |
484 | { | |
485 | xfs_buf_t *bp; /* return value */ | |
486 | int error; | |
487 | ||
488 | ASSERT(agno != NULLAGNUMBER); | |
489 | error = xfs_trans_read_buf( | |
490 | mp, tp, mp->m_ddev_targp, | |
491 | XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)), | |
492 | XFS_FSS_TO_BB(mp, 1), 0, &bp); | |
493 | if (error) | |
494 | return error; | |
495 | ASSERT(bp); | |
496 | ASSERT(!XFS_BUF_GETERROR(bp)); | |
497 | XFS_BUF_SET_VTYPE_REF(bp, B_FS_AGFL, XFS_AGFL_REF); | |
498 | *bpp = bp; | |
499 | return 0; | |
500 | } | |
501 | ||
502 | #if defined(XFS_ALLOC_TRACE) | |
503 | /* | |
504 | * Add an allocation trace entry for an alloc call. | |
505 | */ | |
506 | STATIC void | |
507 | xfs_alloc_trace_alloc( | |
3a59c94c | 508 | const char *name, /* function tag string */ |
1da177e4 LT |
509 | char *str, /* additional string */ |
510 | xfs_alloc_arg_t *args, /* allocation argument structure */ | |
511 | int line) /* source line number */ | |
512 | { | |
513 | ktrace_enter(xfs_alloc_trace_buf, | |
514 | (void *)(__psint_t)(XFS_ALLOC_KTRACE_ALLOC | (line << 16)), | |
515 | (void *)name, | |
516 | (void *)str, | |
517 | (void *)args->mp, | |
518 | (void *)(__psunsigned_t)args->agno, | |
519 | (void *)(__psunsigned_t)args->agbno, | |
520 | (void *)(__psunsigned_t)args->minlen, | |
521 | (void *)(__psunsigned_t)args->maxlen, | |
522 | (void *)(__psunsigned_t)args->mod, | |
523 | (void *)(__psunsigned_t)args->prod, | |
524 | (void *)(__psunsigned_t)args->minleft, | |
525 | (void *)(__psunsigned_t)args->total, | |
526 | (void *)(__psunsigned_t)args->alignment, | |
527 | (void *)(__psunsigned_t)args->len, | |
528 | (void *)((((__psint_t)args->type) << 16) | | |
529 | (__psint_t)args->otype), | |
530 | (void *)(__psint_t)((args->wasdel << 3) | | |
531 | (args->wasfromfl << 2) | | |
532 | (args->isfl << 1) | | |
533 | (args->userdata << 0))); | |
534 | } | |
535 | ||
536 | /* | |
537 | * Add an allocation trace entry for a free call. | |
538 | */ | |
539 | STATIC void | |
540 | xfs_alloc_trace_free( | |
3a59c94c | 541 | const char *name, /* function tag string */ |
1da177e4 LT |
542 | char *str, /* additional string */ |
543 | xfs_mount_t *mp, /* file system mount point */ | |
544 | xfs_agnumber_t agno, /* allocation group number */ | |
545 | xfs_agblock_t agbno, /* a.g. relative block number */ | |
546 | xfs_extlen_t len, /* length of extent */ | |
547 | int isfl, /* set if is freelist allocation/free */ | |
548 | int line) /* source line number */ | |
549 | { | |
550 | ktrace_enter(xfs_alloc_trace_buf, | |
551 | (void *)(__psint_t)(XFS_ALLOC_KTRACE_FREE | (line << 16)), | |
552 | (void *)name, | |
553 | (void *)str, | |
554 | (void *)mp, | |
555 | (void *)(__psunsigned_t)agno, | |
556 | (void *)(__psunsigned_t)agbno, | |
557 | (void *)(__psunsigned_t)len, | |
558 | (void *)(__psint_t)isfl, | |
559 | NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL); | |
560 | } | |
561 | ||
562 | /* | |
563 | * Add an allocation trace entry for modifying an agf. | |
564 | */ | |
565 | STATIC void | |
566 | xfs_alloc_trace_modagf( | |
3a59c94c | 567 | const char *name, /* function tag string */ |
1da177e4 LT |
568 | char *str, /* additional string */ |
569 | xfs_mount_t *mp, /* file system mount point */ | |
570 | xfs_agf_t *agf, /* new agf value */ | |
571 | int flags, /* logging flags for agf */ | |
572 | int line) /* source line number */ | |
573 | { | |
574 | ktrace_enter(xfs_alloc_trace_buf, | |
575 | (void *)(__psint_t)(XFS_ALLOC_KTRACE_MODAGF | (line << 16)), | |
576 | (void *)name, | |
577 | (void *)str, | |
578 | (void *)mp, | |
579 | (void *)(__psint_t)flags, | |
16259e7d CH |
580 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_seqno), |
581 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_length), | |
582 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_roots[XFS_BTNUM_BNO]), | |
583 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_roots[XFS_BTNUM_CNT]), | |
584 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]), | |
585 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]), | |
586 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_flfirst), | |
587 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_fllast), | |
588 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_flcount), | |
589 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_freeblks), | |
590 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_longest)); | |
1da177e4 LT |
591 | } |
592 | ||
593 | STATIC void | |
594 | xfs_alloc_trace_busy( | |
3a59c94c | 595 | const char *name, /* function tag string */ |
1da177e4 | 596 | char *str, /* additional string */ |
c41564b5 | 597 | xfs_mount_t *mp, /* file system mount point */ |
1da177e4 LT |
598 | xfs_agnumber_t agno, /* allocation group number */ |
599 | xfs_agblock_t agbno, /* a.g. relative block number */ | |
600 | xfs_extlen_t len, /* length of extent */ | |
601 | int slot, /* perag Busy slot */ | |
602 | xfs_trans_t *tp, | |
603 | int trtype, /* type: add, delete, search */ | |
604 | int line) /* source line number */ | |
605 | { | |
606 | ktrace_enter(xfs_alloc_trace_buf, | |
607 | (void *)(__psint_t)(trtype | (line << 16)), | |
608 | (void *)name, | |
609 | (void *)str, | |
610 | (void *)mp, | |
611 | (void *)(__psunsigned_t)agno, | |
612 | (void *)(__psunsigned_t)agbno, | |
613 | (void *)(__psunsigned_t)len, | |
614 | (void *)(__psint_t)slot, | |
615 | (void *)tp, | |
616 | NULL, NULL, NULL, NULL, NULL, NULL, NULL); | |
617 | } | |
618 | #endif /* XFS_ALLOC_TRACE */ | |
619 | ||
620 | /* | |
621 | * Allocation group level functions. | |
622 | */ | |
623 | ||
624 | /* | |
625 | * Allocate a variable extent in the allocation group agno. | |
626 | * Type and bno are used to determine where in the allocation group the | |
627 | * extent will start. | |
628 | * Extent's length (returned in *len) will be between minlen and maxlen, | |
629 | * and of the form k * prod + mod unless there's nothing that large. | |
630 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
631 | */ | |
632 | STATIC int /* error */ | |
633 | xfs_alloc_ag_vextent( | |
634 | xfs_alloc_arg_t *args) /* argument structure for allocation */ | |
635 | { | |
636 | int error=0; | |
1da177e4 LT |
637 | |
638 | ASSERT(args->minlen > 0); | |
639 | ASSERT(args->maxlen > 0); | |
640 | ASSERT(args->minlen <= args->maxlen); | |
641 | ASSERT(args->mod < args->prod); | |
642 | ASSERT(args->alignment > 0); | |
643 | /* | |
644 | * Branch to correct routine based on the type. | |
645 | */ | |
646 | args->wasfromfl = 0; | |
647 | switch (args->type) { | |
648 | case XFS_ALLOCTYPE_THIS_AG: | |
649 | error = xfs_alloc_ag_vextent_size(args); | |
650 | break; | |
651 | case XFS_ALLOCTYPE_NEAR_BNO: | |
652 | error = xfs_alloc_ag_vextent_near(args); | |
653 | break; | |
654 | case XFS_ALLOCTYPE_THIS_BNO: | |
655 | error = xfs_alloc_ag_vextent_exact(args); | |
656 | break; | |
657 | default: | |
658 | ASSERT(0); | |
659 | /* NOTREACHED */ | |
660 | } | |
661 | if (error) | |
662 | return error; | |
663 | /* | |
664 | * If the allocation worked, need to change the agf structure | |
665 | * (and log it), and the superblock. | |
666 | */ | |
667 | if (args->agbno != NULLAGBLOCK) { | |
668 | xfs_agf_t *agf; /* allocation group freelist header */ | |
669 | #ifdef XFS_ALLOC_TRACE | |
670 | xfs_mount_t *mp = args->mp; | |
671 | #endif | |
672 | long slen = (long)args->len; | |
673 | ||
674 | ASSERT(args->len >= args->minlen && args->len <= args->maxlen); | |
675 | ASSERT(!(args->wasfromfl) || !args->isfl); | |
676 | ASSERT(args->agbno % args->alignment == 0); | |
677 | if (!(args->wasfromfl)) { | |
678 | ||
679 | agf = XFS_BUF_TO_AGF(args->agbp); | |
413d57c9 | 680 | be32_add_cpu(&agf->agf_freeblks, -(args->len)); |
1da177e4 LT |
681 | xfs_trans_agblocks_delta(args->tp, |
682 | -((long)(args->len))); | |
683 | args->pag->pagf_freeblks -= args->len; | |
16259e7d CH |
684 | ASSERT(be32_to_cpu(agf->agf_freeblks) <= |
685 | be32_to_cpu(agf->agf_length)); | |
1da177e4 LT |
686 | TRACE_MODAGF(NULL, agf, XFS_AGF_FREEBLKS); |
687 | xfs_alloc_log_agf(args->tp, args->agbp, | |
688 | XFS_AGF_FREEBLKS); | |
689 | /* search the busylist for these blocks */ | |
690 | xfs_alloc_search_busy(args->tp, args->agno, | |
691 | args->agbno, args->len); | |
692 | } | |
693 | if (!args->isfl) | |
694 | xfs_trans_mod_sb(args->tp, | |
695 | args->wasdel ? XFS_TRANS_SB_RES_FDBLOCKS : | |
696 | XFS_TRANS_SB_FDBLOCKS, -slen); | |
697 | XFS_STATS_INC(xs_allocx); | |
698 | XFS_STATS_ADD(xs_allocb, args->len); | |
699 | } | |
700 | return 0; | |
701 | } | |
702 | ||
703 | /* | |
704 | * Allocate a variable extent at exactly agno/bno. | |
705 | * Extent's length (returned in *len) will be between minlen and maxlen, | |
706 | * and of the form k * prod + mod unless there's nothing that large. | |
707 | * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it. | |
708 | */ | |
709 | STATIC int /* error */ | |
710 | xfs_alloc_ag_vextent_exact( | |
711 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
712 | { | |
713 | xfs_btree_cur_t *bno_cur;/* by block-number btree cursor */ | |
714 | xfs_btree_cur_t *cnt_cur;/* by count btree cursor */ | |
715 | xfs_agblock_t end; /* end of allocated extent */ | |
716 | int error; | |
717 | xfs_agblock_t fbno; /* start block of found extent */ | |
718 | xfs_agblock_t fend; /* end block of found extent */ | |
719 | xfs_extlen_t flen; /* length of found extent */ | |
1da177e4 LT |
720 | int i; /* success/failure of operation */ |
721 | xfs_agblock_t maxend; /* end of maximal extent */ | |
722 | xfs_agblock_t minend; /* end of minimal extent */ | |
723 | xfs_extlen_t rlen; /* length of returned extent */ | |
724 | ||
725 | ASSERT(args->alignment == 1); | |
726 | /* | |
727 | * Allocate/initialize a cursor for the by-number freespace btree. | |
728 | */ | |
561f7d17 CH |
729 | bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
730 | args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
731 | /* |
732 | * Lookup bno and minlen in the btree (minlen is irrelevant, really). | |
733 | * Look for the closest free block <= bno, it must contain bno | |
734 | * if any free block does. | |
735 | */ | |
736 | if ((error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i))) | |
737 | goto error0; | |
738 | if (!i) { | |
739 | /* | |
740 | * Didn't find it, return null. | |
741 | */ | |
742 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
743 | args->agbno = NULLAGBLOCK; | |
744 | return 0; | |
745 | } | |
746 | /* | |
747 | * Grab the freespace record. | |
748 | */ | |
749 | if ((error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i))) | |
750 | goto error0; | |
751 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
752 | ASSERT(fbno <= args->agbno); | |
753 | minend = args->agbno + args->minlen; | |
754 | maxend = args->agbno + args->maxlen; | |
755 | fend = fbno + flen; | |
756 | /* | |
757 | * Give up if the freespace isn't long enough for the minimum request. | |
758 | */ | |
759 | if (fend < minend) { | |
760 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
761 | args->agbno = NULLAGBLOCK; | |
762 | return 0; | |
763 | } | |
764 | /* | |
765 | * End of extent will be smaller of the freespace end and the | |
766 | * maximal requested end. | |
767 | */ | |
768 | end = XFS_AGBLOCK_MIN(fend, maxend); | |
769 | /* | |
770 | * Fix the length according to mod and prod if given. | |
771 | */ | |
772 | args->len = end - args->agbno; | |
773 | xfs_alloc_fix_len(args); | |
774 | if (!xfs_alloc_fix_minleft(args)) { | |
775 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
776 | return 0; | |
777 | } | |
778 | rlen = args->len; | |
779 | ASSERT(args->agbno + rlen <= fend); | |
780 | end = args->agbno + rlen; | |
781 | /* | |
782 | * We are allocating agbno for rlen [agbno .. end] | |
783 | * Allocate/initialize a cursor for the by-size btree. | |
784 | */ | |
561f7d17 CH |
785 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
786 | args->agno, XFS_BTNUM_CNT); | |
1da177e4 | 787 | ASSERT(args->agbno + args->len <= |
16259e7d | 788 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
1da177e4 LT |
789 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, |
790 | args->agbno, args->len, XFSA_FIXUP_BNO_OK))) { | |
791 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
792 | goto error0; | |
793 | } | |
794 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
795 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
796 | TRACE_ALLOC("normal", args); | |
797 | args->wasfromfl = 0; | |
798 | return 0; | |
799 | ||
800 | error0: | |
801 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
802 | TRACE_ALLOC("error", args); | |
803 | return error; | |
804 | } | |
805 | ||
806 | /* | |
807 | * Allocate a variable extent near bno in the allocation group agno. | |
808 | * Extent's length (returned in len) will be between minlen and maxlen, | |
809 | * and of the form k * prod + mod unless there's nothing that large. | |
810 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
811 | */ | |
812 | STATIC int /* error */ | |
813 | xfs_alloc_ag_vextent_near( | |
814 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
815 | { | |
816 | xfs_btree_cur_t *bno_cur_gt; /* cursor for bno btree, right side */ | |
817 | xfs_btree_cur_t *bno_cur_lt; /* cursor for bno btree, left side */ | |
818 | xfs_btree_cur_t *cnt_cur; /* cursor for count btree */ | |
1da177e4 LT |
819 | xfs_agblock_t gtbno; /* start bno of right side entry */ |
820 | xfs_agblock_t gtbnoa; /* aligned ... */ | |
821 | xfs_extlen_t gtdiff; /* difference to right side entry */ | |
822 | xfs_extlen_t gtlen; /* length of right side entry */ | |
823 | xfs_extlen_t gtlena; /* aligned ... */ | |
824 | xfs_agblock_t gtnew; /* useful start bno of right side */ | |
825 | int error; /* error code */ | |
826 | int i; /* result code, temporary */ | |
827 | int j; /* result code, temporary */ | |
828 | xfs_agblock_t ltbno; /* start bno of left side entry */ | |
829 | xfs_agblock_t ltbnoa; /* aligned ... */ | |
830 | xfs_extlen_t ltdiff; /* difference to left side entry */ | |
831 | /*REFERENCED*/ | |
832 | xfs_agblock_t ltend; /* end bno of left side entry */ | |
833 | xfs_extlen_t ltlen; /* length of left side entry */ | |
834 | xfs_extlen_t ltlena; /* aligned ... */ | |
835 | xfs_agblock_t ltnew; /* useful start bno of left side */ | |
836 | xfs_extlen_t rlen; /* length of returned extent */ | |
837 | #if defined(DEBUG) && defined(__KERNEL__) | |
838 | /* | |
839 | * Randomly don't execute the first algorithm. | |
840 | */ | |
841 | int dofirst; /* set to do first algorithm */ | |
842 | ||
e7a23a9b | 843 | dofirst = random32() & 1; |
1da177e4 LT |
844 | #endif |
845 | /* | |
846 | * Get a cursor for the by-size btree. | |
847 | */ | |
561f7d17 CH |
848 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
849 | args->agno, XFS_BTNUM_CNT); | |
1da177e4 LT |
850 | ltlen = 0; |
851 | bno_cur_lt = bno_cur_gt = NULL; | |
852 | /* | |
853 | * See if there are any free extents as big as maxlen. | |
854 | */ | |
855 | if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, args->maxlen, &i))) | |
856 | goto error0; | |
857 | /* | |
858 | * If none, then pick up the last entry in the tree unless the | |
859 | * tree is empty. | |
860 | */ | |
861 | if (!i) { | |
862 | if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, <bno, | |
863 | <len, &i))) | |
864 | goto error0; | |
865 | if (i == 0 || ltlen == 0) { | |
866 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
867 | return 0; | |
868 | } | |
869 | ASSERT(i == 1); | |
870 | } | |
871 | args->wasfromfl = 0; | |
872 | /* | |
873 | * First algorithm. | |
874 | * If the requested extent is large wrt the freespaces available | |
875 | * in this a.g., then the cursor will be pointing to a btree entry | |
876 | * near the right edge of the tree. If it's in the last btree leaf | |
877 | * block, then we just examine all the entries in that block | |
878 | * that are big enough, and pick the best one. | |
879 | * This is written as a while loop so we can break out of it, | |
880 | * but we never loop back to the top. | |
881 | */ | |
882 | while (xfs_btree_islastblock(cnt_cur, 0)) { | |
883 | xfs_extlen_t bdiff; | |
884 | int besti=0; | |
885 | xfs_extlen_t blen=0; | |
886 | xfs_agblock_t bnew=0; | |
887 | ||
888 | #if defined(DEBUG) && defined(__KERNEL__) | |
889 | if (!dofirst) | |
890 | break; | |
891 | #endif | |
892 | /* | |
893 | * Start from the entry that lookup found, sequence through | |
894 | * all larger free blocks. If we're actually pointing at a | |
895 | * record smaller than maxlen, go to the start of this block, | |
896 | * and skip all those smaller than minlen. | |
897 | */ | |
898 | if (ltlen || args->alignment > 1) { | |
899 | cnt_cur->bc_ptrs[0] = 1; | |
900 | do { | |
901 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, | |
902 | <len, &i))) | |
903 | goto error0; | |
904 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
905 | if (ltlen >= args->minlen) | |
906 | break; | |
637aa50f | 907 | if ((error = xfs_btree_increment(cnt_cur, 0, &i))) |
1da177e4 LT |
908 | goto error0; |
909 | } while (i); | |
910 | ASSERT(ltlen >= args->minlen); | |
911 | if (!i) | |
912 | break; | |
913 | } | |
914 | i = cnt_cur->bc_ptrs[0]; | |
915 | for (j = 1, blen = 0, bdiff = 0; | |
916 | !error && j && (blen < args->maxlen || bdiff > 0); | |
637aa50f | 917 | error = xfs_btree_increment(cnt_cur, 0, &j)) { |
1da177e4 LT |
918 | /* |
919 | * For each entry, decide if it's better than | |
920 | * the previous best entry. | |
921 | */ | |
922 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i))) | |
923 | goto error0; | |
924 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
12375c82 DC |
925 | xfs_alloc_compute_aligned(ltbno, ltlen, args->alignment, |
926 | args->minlen, <bnoa, <lena); | |
e6430037 | 927 | if (ltlena < args->minlen) |
1da177e4 LT |
928 | continue; |
929 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); | |
930 | xfs_alloc_fix_len(args); | |
931 | ASSERT(args->len >= args->minlen); | |
932 | if (args->len < blen) | |
933 | continue; | |
934 | ltdiff = xfs_alloc_compute_diff(args->agbno, args->len, | |
935 | args->alignment, ltbno, ltlen, <new); | |
936 | if (ltnew != NULLAGBLOCK && | |
937 | (args->len > blen || ltdiff < bdiff)) { | |
938 | bdiff = ltdiff; | |
939 | bnew = ltnew; | |
940 | blen = args->len; | |
941 | besti = cnt_cur->bc_ptrs[0]; | |
942 | } | |
943 | } | |
944 | /* | |
945 | * It didn't work. We COULD be in a case where | |
946 | * there's a good record somewhere, so try again. | |
947 | */ | |
948 | if (blen == 0) | |
949 | break; | |
950 | /* | |
951 | * Point at the best entry, and retrieve it again. | |
952 | */ | |
953 | cnt_cur->bc_ptrs[0] = besti; | |
954 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i))) | |
955 | goto error0; | |
956 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
957 | ltend = ltbno + ltlen; | |
16259e7d | 958 | ASSERT(ltend <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
1da177e4 LT |
959 | args->len = blen; |
960 | if (!xfs_alloc_fix_minleft(args)) { | |
961 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
962 | TRACE_ALLOC("nominleft", args); | |
963 | return 0; | |
964 | } | |
965 | blen = args->len; | |
966 | /* | |
967 | * We are allocating starting at bnew for blen blocks. | |
968 | */ | |
969 | args->agbno = bnew; | |
970 | ASSERT(bnew >= ltbno); | |
971 | ASSERT(bnew + blen <= ltend); | |
972 | /* | |
973 | * Set up a cursor for the by-bno tree. | |
974 | */ | |
561f7d17 CH |
975 | bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, |
976 | args->agbp, args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
977 | /* |
978 | * Fix up the btree entries. | |
979 | */ | |
980 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, | |
981 | ltlen, bnew, blen, XFSA_FIXUP_CNT_OK))) | |
982 | goto error0; | |
983 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
984 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); | |
985 | TRACE_ALLOC("first", args); | |
986 | return 0; | |
987 | } | |
988 | /* | |
989 | * Second algorithm. | |
990 | * Search in the by-bno tree to the left and to the right | |
991 | * simultaneously, until in each case we find a space big enough, | |
992 | * or run into the edge of the tree. When we run into the edge, | |
993 | * we deallocate that cursor. | |
994 | * If both searches succeed, we compare the two spaces and pick | |
995 | * the better one. | |
996 | * With alignment, it's possible for both to fail; the upper | |
997 | * level algorithm that picks allocation groups for allocations | |
998 | * is not supposed to do this. | |
999 | */ | |
1000 | /* | |
1001 | * Allocate and initialize the cursor for the leftward search. | |
1002 | */ | |
561f7d17 CH |
1003 | bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1004 | args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
1005 | /* |
1006 | * Lookup <= bno to find the leftward search's starting point. | |
1007 | */ | |
1008 | if ((error = xfs_alloc_lookup_le(bno_cur_lt, args->agbno, args->maxlen, &i))) | |
1009 | goto error0; | |
1010 | if (!i) { | |
1011 | /* | |
1012 | * Didn't find anything; use this cursor for the rightward | |
1013 | * search. | |
1014 | */ | |
1015 | bno_cur_gt = bno_cur_lt; | |
1016 | bno_cur_lt = NULL; | |
1017 | } | |
1018 | /* | |
1019 | * Found something. Duplicate the cursor for the rightward search. | |
1020 | */ | |
1021 | else if ((error = xfs_btree_dup_cursor(bno_cur_lt, &bno_cur_gt))) | |
1022 | goto error0; | |
1023 | /* | |
1024 | * Increment the cursor, so we will point at the entry just right | |
1025 | * of the leftward entry if any, or to the leftmost entry. | |
1026 | */ | |
637aa50f | 1027 | if ((error = xfs_btree_increment(bno_cur_gt, 0, &i))) |
1da177e4 LT |
1028 | goto error0; |
1029 | if (!i) { | |
1030 | /* | |
1031 | * It failed, there are no rightward entries. | |
1032 | */ | |
1033 | xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_NOERROR); | |
1034 | bno_cur_gt = NULL; | |
1035 | } | |
1036 | /* | |
1037 | * Loop going left with the leftward cursor, right with the | |
1038 | * rightward cursor, until either both directions give up or | |
1039 | * we find an entry at least as big as minlen. | |
1040 | */ | |
1041 | do { | |
1042 | if (bno_cur_lt) { | |
1043 | if ((error = xfs_alloc_get_rec(bno_cur_lt, <bno, <len, &i))) | |
1044 | goto error0; | |
1045 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
12375c82 DC |
1046 | xfs_alloc_compute_aligned(ltbno, ltlen, args->alignment, |
1047 | args->minlen, <bnoa, <lena); | |
1048 | if (ltlena >= args->minlen) | |
1da177e4 | 1049 | break; |
8df4da4a | 1050 | if ((error = xfs_btree_decrement(bno_cur_lt, 0, &i))) |
1da177e4 LT |
1051 | goto error0; |
1052 | if (!i) { | |
1053 | xfs_btree_del_cursor(bno_cur_lt, | |
1054 | XFS_BTREE_NOERROR); | |
1055 | bno_cur_lt = NULL; | |
1056 | } | |
1057 | } | |
1058 | if (bno_cur_gt) { | |
1059 | if ((error = xfs_alloc_get_rec(bno_cur_gt, >bno, >len, &i))) | |
1060 | goto error0; | |
1061 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
12375c82 DC |
1062 | xfs_alloc_compute_aligned(gtbno, gtlen, args->alignment, |
1063 | args->minlen, >bnoa, >lena); | |
1064 | if (gtlena >= args->minlen) | |
1da177e4 | 1065 | break; |
637aa50f | 1066 | if ((error = xfs_btree_increment(bno_cur_gt, 0, &i))) |
1da177e4 LT |
1067 | goto error0; |
1068 | if (!i) { | |
1069 | xfs_btree_del_cursor(bno_cur_gt, | |
1070 | XFS_BTREE_NOERROR); | |
1071 | bno_cur_gt = NULL; | |
1072 | } | |
1073 | } | |
1074 | } while (bno_cur_lt || bno_cur_gt); | |
1075 | /* | |
1076 | * Got both cursors still active, need to find better entry. | |
1077 | */ | |
1078 | if (bno_cur_lt && bno_cur_gt) { | |
1079 | /* | |
1080 | * Left side is long enough, look for a right side entry. | |
1081 | */ | |
1082 | if (ltlena >= args->minlen) { | |
1083 | /* | |
1084 | * Fix up the length. | |
1085 | */ | |
1086 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); | |
1087 | xfs_alloc_fix_len(args); | |
1088 | rlen = args->len; | |
1089 | ltdiff = xfs_alloc_compute_diff(args->agbno, rlen, | |
1090 | args->alignment, ltbno, ltlen, <new); | |
1091 | /* | |
1092 | * Not perfect. | |
1093 | */ | |
1094 | if (ltdiff) { | |
1095 | /* | |
1096 | * Look until we find a better one, run out of | |
1097 | * space, or run off the end. | |
1098 | */ | |
1099 | while (bno_cur_lt && bno_cur_gt) { | |
1100 | if ((error = xfs_alloc_get_rec( | |
1101 | bno_cur_gt, >bno, | |
1102 | >len, &i))) | |
1103 | goto error0; | |
1104 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1105 | xfs_alloc_compute_aligned(gtbno, gtlen, | |
1106 | args->alignment, args->minlen, | |
1107 | >bnoa, >lena); | |
1108 | /* | |
1109 | * The left one is clearly better. | |
1110 | */ | |
1111 | if (gtbnoa >= args->agbno + ltdiff) { | |
1112 | xfs_btree_del_cursor( | |
1113 | bno_cur_gt, | |
1114 | XFS_BTREE_NOERROR); | |
1115 | bno_cur_gt = NULL; | |
1116 | break; | |
1117 | } | |
1118 | /* | |
1119 | * If we reach a big enough entry, | |
1120 | * compare the two and pick the best. | |
1121 | */ | |
1122 | if (gtlena >= args->minlen) { | |
1123 | args->len = | |
1124 | XFS_EXTLEN_MIN(gtlena, | |
1125 | args->maxlen); | |
1126 | xfs_alloc_fix_len(args); | |
1127 | rlen = args->len; | |
1128 | gtdiff = xfs_alloc_compute_diff( | |
1129 | args->agbno, rlen, | |
1130 | args->alignment, | |
1131 | gtbno, gtlen, >new); | |
1132 | /* | |
1133 | * Right side is better. | |
1134 | */ | |
1135 | if (gtdiff < ltdiff) { | |
1136 | xfs_btree_del_cursor( | |
1137 | bno_cur_lt, | |
1138 | XFS_BTREE_NOERROR); | |
1139 | bno_cur_lt = NULL; | |
1140 | } | |
1141 | /* | |
1142 | * Left side is better. | |
1143 | */ | |
1144 | else { | |
1145 | xfs_btree_del_cursor( | |
1146 | bno_cur_gt, | |
1147 | XFS_BTREE_NOERROR); | |
1148 | bno_cur_gt = NULL; | |
1149 | } | |
1150 | break; | |
1151 | } | |
1152 | /* | |
1153 | * Fell off the right end. | |
1154 | */ | |
637aa50f | 1155 | if ((error = xfs_btree_increment( |
1da177e4 LT |
1156 | bno_cur_gt, 0, &i))) |
1157 | goto error0; | |
1158 | if (!i) { | |
1159 | xfs_btree_del_cursor( | |
1160 | bno_cur_gt, | |
1161 | XFS_BTREE_NOERROR); | |
1162 | bno_cur_gt = NULL; | |
1163 | break; | |
1164 | } | |
1165 | } | |
1166 | } | |
1167 | /* | |
1168 | * The left side is perfect, trash the right side. | |
1169 | */ | |
1170 | else { | |
1171 | xfs_btree_del_cursor(bno_cur_gt, | |
1172 | XFS_BTREE_NOERROR); | |
1173 | bno_cur_gt = NULL; | |
1174 | } | |
1175 | } | |
1176 | /* | |
1177 | * It's the right side that was found first, look left. | |
1178 | */ | |
1179 | else { | |
1180 | /* | |
1181 | * Fix up the length. | |
1182 | */ | |
1183 | args->len = XFS_EXTLEN_MIN(gtlena, args->maxlen); | |
1184 | xfs_alloc_fix_len(args); | |
1185 | rlen = args->len; | |
1186 | gtdiff = xfs_alloc_compute_diff(args->agbno, rlen, | |
1187 | args->alignment, gtbno, gtlen, >new); | |
1188 | /* | |
1189 | * Right side entry isn't perfect. | |
1190 | */ | |
1191 | if (gtdiff) { | |
1192 | /* | |
1193 | * Look until we find a better one, run out of | |
1194 | * space, or run off the end. | |
1195 | */ | |
1196 | while (bno_cur_lt && bno_cur_gt) { | |
1197 | if ((error = xfs_alloc_get_rec( | |
1198 | bno_cur_lt, <bno, | |
1199 | <len, &i))) | |
1200 | goto error0; | |
1201 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1202 | xfs_alloc_compute_aligned(ltbno, ltlen, | |
1203 | args->alignment, args->minlen, | |
1204 | <bnoa, <lena); | |
1205 | /* | |
1206 | * The right one is clearly better. | |
1207 | */ | |
1208 | if (ltbnoa <= args->agbno - gtdiff) { | |
1209 | xfs_btree_del_cursor( | |
1210 | bno_cur_lt, | |
1211 | XFS_BTREE_NOERROR); | |
1212 | bno_cur_lt = NULL; | |
1213 | break; | |
1214 | } | |
1215 | /* | |
1216 | * If we reach a big enough entry, | |
1217 | * compare the two and pick the best. | |
1218 | */ | |
1219 | if (ltlena >= args->minlen) { | |
1220 | args->len = XFS_EXTLEN_MIN( | |
1221 | ltlena, args->maxlen); | |
1222 | xfs_alloc_fix_len(args); | |
1223 | rlen = args->len; | |
1224 | ltdiff = xfs_alloc_compute_diff( | |
1225 | args->agbno, rlen, | |
1226 | args->alignment, | |
1227 | ltbno, ltlen, <new); | |
1228 | /* | |
1229 | * Left side is better. | |
1230 | */ | |
1231 | if (ltdiff < gtdiff) { | |
1232 | xfs_btree_del_cursor( | |
1233 | bno_cur_gt, | |
1234 | XFS_BTREE_NOERROR); | |
1235 | bno_cur_gt = NULL; | |
1236 | } | |
1237 | /* | |
1238 | * Right side is better. | |
1239 | */ | |
1240 | else { | |
1241 | xfs_btree_del_cursor( | |
1242 | bno_cur_lt, | |
1243 | XFS_BTREE_NOERROR); | |
1244 | bno_cur_lt = NULL; | |
1245 | } | |
1246 | break; | |
1247 | } | |
1248 | /* | |
1249 | * Fell off the left end. | |
1250 | */ | |
8df4da4a | 1251 | if ((error = xfs_btree_decrement( |
1da177e4 LT |
1252 | bno_cur_lt, 0, &i))) |
1253 | goto error0; | |
1254 | if (!i) { | |
1255 | xfs_btree_del_cursor(bno_cur_lt, | |
1256 | XFS_BTREE_NOERROR); | |
1257 | bno_cur_lt = NULL; | |
1258 | break; | |
1259 | } | |
1260 | } | |
1261 | } | |
1262 | /* | |
1263 | * The right side is perfect, trash the left side. | |
1264 | */ | |
1265 | else { | |
1266 | xfs_btree_del_cursor(bno_cur_lt, | |
1267 | XFS_BTREE_NOERROR); | |
1268 | bno_cur_lt = NULL; | |
1269 | } | |
1270 | } | |
1271 | } | |
1272 | /* | |
1273 | * If we couldn't get anything, give up. | |
1274 | */ | |
1275 | if (bno_cur_lt == NULL && bno_cur_gt == NULL) { | |
1276 | TRACE_ALLOC("neither", args); | |
1277 | args->agbno = NULLAGBLOCK; | |
1278 | return 0; | |
1279 | } | |
1280 | /* | |
1281 | * At this point we have selected a freespace entry, either to the | |
1282 | * left or to the right. If it's on the right, copy all the | |
1283 | * useful variables to the "left" set so we only have one | |
1284 | * copy of this code. | |
1285 | */ | |
1286 | if (bno_cur_gt) { | |
1287 | bno_cur_lt = bno_cur_gt; | |
1288 | bno_cur_gt = NULL; | |
1289 | ltbno = gtbno; | |
1290 | ltbnoa = gtbnoa; | |
1291 | ltlen = gtlen; | |
1292 | ltlena = gtlena; | |
1293 | j = 1; | |
1294 | } else | |
1295 | j = 0; | |
1296 | /* | |
1297 | * Fix up the length and compute the useful address. | |
1298 | */ | |
1299 | ltend = ltbno + ltlen; | |
1300 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); | |
1301 | xfs_alloc_fix_len(args); | |
1302 | if (!xfs_alloc_fix_minleft(args)) { | |
1303 | TRACE_ALLOC("nominleft", args); | |
1304 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); | |
1305 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1306 | return 0; | |
1307 | } | |
1308 | rlen = args->len; | |
1309 | (void)xfs_alloc_compute_diff(args->agbno, rlen, args->alignment, ltbno, | |
1310 | ltlen, <new); | |
1311 | ASSERT(ltnew >= ltbno); | |
1312 | ASSERT(ltnew + rlen <= ltend); | |
16259e7d | 1313 | ASSERT(ltnew + rlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
1da177e4 LT |
1314 | args->agbno = ltnew; |
1315 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, ltlen, | |
1316 | ltnew, rlen, XFSA_FIXUP_BNO_OK))) | |
1317 | goto error0; | |
1318 | TRACE_ALLOC(j ? "gt" : "lt", args); | |
1319 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1320 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); | |
1321 | return 0; | |
1322 | ||
1323 | error0: | |
1324 | TRACE_ALLOC("error", args); | |
1325 | if (cnt_cur != NULL) | |
1326 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1327 | if (bno_cur_lt != NULL) | |
1328 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_ERROR); | |
1329 | if (bno_cur_gt != NULL) | |
1330 | xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_ERROR); | |
1331 | return error; | |
1332 | } | |
1333 | ||
1334 | /* | |
1335 | * Allocate a variable extent anywhere in the allocation group agno. | |
1336 | * Extent's length (returned in len) will be between minlen and maxlen, | |
1337 | * and of the form k * prod + mod unless there's nothing that large. | |
1338 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
1339 | */ | |
1340 | STATIC int /* error */ | |
1341 | xfs_alloc_ag_vextent_size( | |
1342 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
1343 | { | |
1344 | xfs_btree_cur_t *bno_cur; /* cursor for bno btree */ | |
1345 | xfs_btree_cur_t *cnt_cur; /* cursor for cnt btree */ | |
1346 | int error; /* error result */ | |
1347 | xfs_agblock_t fbno; /* start of found freespace */ | |
1348 | xfs_extlen_t flen; /* length of found freespace */ | |
1da177e4 LT |
1349 | int i; /* temp status variable */ |
1350 | xfs_agblock_t rbno; /* returned block number */ | |
1351 | xfs_extlen_t rlen; /* length of returned extent */ | |
1352 | ||
1353 | /* | |
1354 | * Allocate and initialize a cursor for the by-size btree. | |
1355 | */ | |
561f7d17 CH |
1356 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1357 | args->agno, XFS_BTNUM_CNT); | |
1da177e4 LT |
1358 | bno_cur = NULL; |
1359 | /* | |
1360 | * Look for an entry >= maxlen+alignment-1 blocks. | |
1361 | */ | |
1362 | if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, | |
1363 | args->maxlen + args->alignment - 1, &i))) | |
1364 | goto error0; | |
1365 | /* | |
1366 | * If none, then pick up the last entry in the tree unless the | |
1367 | * tree is empty. | |
1368 | */ | |
1369 | if (!i) { | |
1370 | if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, &fbno, | |
1371 | &flen, &i))) | |
1372 | goto error0; | |
1373 | if (i == 0 || flen == 0) { | |
1374 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1375 | TRACE_ALLOC("noentry", args); | |
1376 | return 0; | |
1377 | } | |
1378 | ASSERT(i == 1); | |
1379 | } | |
1380 | /* | |
1381 | * There's a freespace as big as maxlen+alignment-1, get it. | |
1382 | */ | |
1383 | else { | |
1384 | if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i))) | |
1385 | goto error0; | |
1386 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1387 | } | |
1388 | /* | |
1389 | * In the first case above, we got the last entry in the | |
1390 | * by-size btree. Now we check to see if the space hits maxlen | |
1391 | * once aligned; if not, we search left for something better. | |
1392 | * This can't happen in the second case above. | |
1393 | */ | |
1394 | xfs_alloc_compute_aligned(fbno, flen, args->alignment, args->minlen, | |
1395 | &rbno, &rlen); | |
1396 | rlen = XFS_EXTLEN_MIN(args->maxlen, rlen); | |
1397 | XFS_WANT_CORRUPTED_GOTO(rlen == 0 || | |
1398 | (rlen <= flen && rbno + rlen <= fbno + flen), error0); | |
1399 | if (rlen < args->maxlen) { | |
1400 | xfs_agblock_t bestfbno; | |
1401 | xfs_extlen_t bestflen; | |
1402 | xfs_agblock_t bestrbno; | |
1403 | xfs_extlen_t bestrlen; | |
1404 | ||
1405 | bestrlen = rlen; | |
1406 | bestrbno = rbno; | |
1407 | bestflen = flen; | |
1408 | bestfbno = fbno; | |
1409 | for (;;) { | |
8df4da4a | 1410 | if ((error = xfs_btree_decrement(cnt_cur, 0, &i))) |
1da177e4 LT |
1411 | goto error0; |
1412 | if (i == 0) | |
1413 | break; | |
1414 | if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, | |
1415 | &i))) | |
1416 | goto error0; | |
1417 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1418 | if (flen < bestrlen) | |
1419 | break; | |
1420 | xfs_alloc_compute_aligned(fbno, flen, args->alignment, | |
1421 | args->minlen, &rbno, &rlen); | |
1422 | rlen = XFS_EXTLEN_MIN(args->maxlen, rlen); | |
1423 | XFS_WANT_CORRUPTED_GOTO(rlen == 0 || | |
1424 | (rlen <= flen && rbno + rlen <= fbno + flen), | |
1425 | error0); | |
1426 | if (rlen > bestrlen) { | |
1427 | bestrlen = rlen; | |
1428 | bestrbno = rbno; | |
1429 | bestflen = flen; | |
1430 | bestfbno = fbno; | |
1431 | if (rlen == args->maxlen) | |
1432 | break; | |
1433 | } | |
1434 | } | |
1435 | if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen, | |
1436 | &i))) | |
1437 | goto error0; | |
1438 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1439 | rlen = bestrlen; | |
1440 | rbno = bestrbno; | |
1441 | flen = bestflen; | |
1442 | fbno = bestfbno; | |
1443 | } | |
1444 | args->wasfromfl = 0; | |
1445 | /* | |
1446 | * Fix up the length. | |
1447 | */ | |
1448 | args->len = rlen; | |
1449 | xfs_alloc_fix_len(args); | |
1450 | if (rlen < args->minlen || !xfs_alloc_fix_minleft(args)) { | |
1451 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1452 | TRACE_ALLOC("nominleft", args); | |
1453 | args->agbno = NULLAGBLOCK; | |
1454 | return 0; | |
1455 | } | |
1456 | rlen = args->len; | |
1457 | XFS_WANT_CORRUPTED_GOTO(rlen <= flen, error0); | |
1458 | /* | |
1459 | * Allocate and initialize a cursor for the by-block tree. | |
1460 | */ | |
561f7d17 CH |
1461 | bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1462 | args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
1463 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, |
1464 | rbno, rlen, XFSA_FIXUP_CNT_OK))) | |
1465 | goto error0; | |
1466 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1467 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
1468 | cnt_cur = bno_cur = NULL; | |
1469 | args->len = rlen; | |
1470 | args->agbno = rbno; | |
1471 | XFS_WANT_CORRUPTED_GOTO( | |
1472 | args->agbno + args->len <= | |
16259e7d | 1473 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length), |
1da177e4 LT |
1474 | error0); |
1475 | TRACE_ALLOC("normal", args); | |
1476 | return 0; | |
1477 | ||
1478 | error0: | |
1479 | TRACE_ALLOC("error", args); | |
1480 | if (cnt_cur) | |
1481 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1482 | if (bno_cur) | |
1483 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
1484 | return error; | |
1485 | } | |
1486 | ||
1487 | /* | |
1488 | * Deal with the case where only small freespaces remain. | |
1489 | * Either return the contents of the last freespace record, | |
1490 | * or allocate space from the freelist if there is nothing in the tree. | |
1491 | */ | |
1492 | STATIC int /* error */ | |
1493 | xfs_alloc_ag_vextent_small( | |
1494 | xfs_alloc_arg_t *args, /* allocation argument structure */ | |
1495 | xfs_btree_cur_t *ccur, /* by-size cursor */ | |
1496 | xfs_agblock_t *fbnop, /* result block number */ | |
1497 | xfs_extlen_t *flenp, /* result length */ | |
1498 | int *stat) /* status: 0-freelist, 1-normal/none */ | |
1499 | { | |
1500 | int error; | |
1501 | xfs_agblock_t fbno; | |
1502 | xfs_extlen_t flen; | |
1da177e4 LT |
1503 | int i; |
1504 | ||
8df4da4a | 1505 | if ((error = xfs_btree_decrement(ccur, 0, &i))) |
1da177e4 LT |
1506 | goto error0; |
1507 | if (i) { | |
1508 | if ((error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i))) | |
1509 | goto error0; | |
1510 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1511 | } | |
1512 | /* | |
1513 | * Nothing in the btree, try the freelist. Make sure | |
1514 | * to respect minleft even when pulling from the | |
1515 | * freelist. | |
1516 | */ | |
1517 | else if (args->minlen == 1 && args->alignment == 1 && !args->isfl && | |
16259e7d CH |
1518 | (be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_flcount) |
1519 | > args->minleft)) { | |
92821e2b DC |
1520 | error = xfs_alloc_get_freelist(args->tp, args->agbp, &fbno, 0); |
1521 | if (error) | |
1da177e4 LT |
1522 | goto error0; |
1523 | if (fbno != NULLAGBLOCK) { | |
1524 | if (args->userdata) { | |
1525 | xfs_buf_t *bp; | |
1526 | ||
1527 | bp = xfs_btree_get_bufs(args->mp, args->tp, | |
1528 | args->agno, fbno, 0); | |
1529 | xfs_trans_binval(args->tp, bp); | |
1530 | } | |
1531 | args->len = 1; | |
1532 | args->agbno = fbno; | |
1533 | XFS_WANT_CORRUPTED_GOTO( | |
1534 | args->agbno + args->len <= | |
16259e7d | 1535 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length), |
1da177e4 LT |
1536 | error0); |
1537 | args->wasfromfl = 1; | |
1538 | TRACE_ALLOC("freelist", args); | |
1539 | *stat = 0; | |
1540 | return 0; | |
1541 | } | |
1542 | /* | |
1543 | * Nothing in the freelist. | |
1544 | */ | |
1545 | else | |
1546 | flen = 0; | |
1547 | } | |
1548 | /* | |
1549 | * Can't allocate from the freelist for some reason. | |
1550 | */ | |
d432c80e NS |
1551 | else { |
1552 | fbno = NULLAGBLOCK; | |
1da177e4 | 1553 | flen = 0; |
d432c80e | 1554 | } |
1da177e4 LT |
1555 | /* |
1556 | * Can't do the allocation, give up. | |
1557 | */ | |
1558 | if (flen < args->minlen) { | |
1559 | args->agbno = NULLAGBLOCK; | |
1560 | TRACE_ALLOC("notenough", args); | |
1561 | flen = 0; | |
1562 | } | |
1563 | *fbnop = fbno; | |
1564 | *flenp = flen; | |
1565 | *stat = 1; | |
1566 | TRACE_ALLOC("normal", args); | |
1567 | return 0; | |
1568 | ||
1569 | error0: | |
1570 | TRACE_ALLOC("error", args); | |
1571 | return error; | |
1572 | } | |
1573 | ||
1574 | /* | |
1575 | * Free the extent starting at agno/bno for length. | |
1576 | */ | |
1577 | STATIC int /* error */ | |
1578 | xfs_free_ag_extent( | |
1579 | xfs_trans_t *tp, /* transaction pointer */ | |
1580 | xfs_buf_t *agbp, /* buffer for a.g. freelist header */ | |
1581 | xfs_agnumber_t agno, /* allocation group number */ | |
1582 | xfs_agblock_t bno, /* starting block number */ | |
1583 | xfs_extlen_t len, /* length of extent */ | |
1584 | int isfl) /* set if is freelist blocks - no sb acctg */ | |
1585 | { | |
1586 | xfs_btree_cur_t *bno_cur; /* cursor for by-block btree */ | |
1587 | xfs_btree_cur_t *cnt_cur; /* cursor for by-size btree */ | |
1588 | int error; /* error return value */ | |
1da177e4 LT |
1589 | xfs_agblock_t gtbno; /* start of right neighbor block */ |
1590 | xfs_extlen_t gtlen; /* length of right neighbor block */ | |
1591 | int haveleft; /* have a left neighbor block */ | |
1592 | int haveright; /* have a right neighbor block */ | |
1593 | int i; /* temp, result code */ | |
1594 | xfs_agblock_t ltbno; /* start of left neighbor block */ | |
1595 | xfs_extlen_t ltlen; /* length of left neighbor block */ | |
1596 | xfs_mount_t *mp; /* mount point struct for filesystem */ | |
1597 | xfs_agblock_t nbno; /* new starting block of freespace */ | |
1598 | xfs_extlen_t nlen; /* new length of freespace */ | |
1599 | ||
1600 | mp = tp->t_mountp; | |
1601 | /* | |
1602 | * Allocate and initialize a cursor for the by-block btree. | |
1603 | */ | |
561f7d17 | 1604 | bno_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_BNO); |
1da177e4 LT |
1605 | cnt_cur = NULL; |
1606 | /* | |
1607 | * Look for a neighboring block on the left (lower block numbers) | |
1608 | * that is contiguous with this space. | |
1609 | */ | |
1610 | if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft))) | |
1611 | goto error0; | |
1612 | if (haveleft) { | |
1613 | /* | |
1614 | * There is a block to our left. | |
1615 | */ | |
1616 | if ((error = xfs_alloc_get_rec(bno_cur, <bno, <len, &i))) | |
1617 | goto error0; | |
1618 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1619 | /* | |
1620 | * It's not contiguous, though. | |
1621 | */ | |
1622 | if (ltbno + ltlen < bno) | |
1623 | haveleft = 0; | |
1624 | else { | |
1625 | /* | |
1626 | * If this failure happens the request to free this | |
1627 | * space was invalid, it's (partly) already free. | |
1628 | * Very bad. | |
1629 | */ | |
1630 | XFS_WANT_CORRUPTED_GOTO(ltbno + ltlen <= bno, error0); | |
1631 | } | |
1632 | } | |
1633 | /* | |
1634 | * Look for a neighboring block on the right (higher block numbers) | |
1635 | * that is contiguous with this space. | |
1636 | */ | |
637aa50f | 1637 | if ((error = xfs_btree_increment(bno_cur, 0, &haveright))) |
1da177e4 LT |
1638 | goto error0; |
1639 | if (haveright) { | |
1640 | /* | |
1641 | * There is a block to our right. | |
1642 | */ | |
1643 | if ((error = xfs_alloc_get_rec(bno_cur, >bno, >len, &i))) | |
1644 | goto error0; | |
1645 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1646 | /* | |
1647 | * It's not contiguous, though. | |
1648 | */ | |
1649 | if (bno + len < gtbno) | |
1650 | haveright = 0; | |
1651 | else { | |
1652 | /* | |
1653 | * If this failure happens the request to free this | |
1654 | * space was invalid, it's (partly) already free. | |
1655 | * Very bad. | |
1656 | */ | |
1657 | XFS_WANT_CORRUPTED_GOTO(gtbno >= bno + len, error0); | |
1658 | } | |
1659 | } | |
1660 | /* | |
1661 | * Now allocate and initialize a cursor for the by-size tree. | |
1662 | */ | |
561f7d17 | 1663 | cnt_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_CNT); |
1da177e4 LT |
1664 | /* |
1665 | * Have both left and right contiguous neighbors. | |
1666 | * Merge all three into a single free block. | |
1667 | */ | |
1668 | if (haveleft && haveright) { | |
1669 | /* | |
1670 | * Delete the old by-size entry on the left. | |
1671 | */ | |
1672 | if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i))) | |
1673 | goto error0; | |
1674 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
91cca5df | 1675 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 LT |
1676 | goto error0; |
1677 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1678 | /* | |
1679 | * Delete the old by-size entry on the right. | |
1680 | */ | |
1681 | if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i))) | |
1682 | goto error0; | |
1683 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
91cca5df | 1684 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 LT |
1685 | goto error0; |
1686 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1687 | /* | |
1688 | * Delete the old by-block entry for the right block. | |
1689 | */ | |
91cca5df | 1690 | if ((error = xfs_btree_delete(bno_cur, &i))) |
1da177e4 LT |
1691 | goto error0; |
1692 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1693 | /* | |
1694 | * Move the by-block cursor back to the left neighbor. | |
1695 | */ | |
8df4da4a | 1696 | if ((error = xfs_btree_decrement(bno_cur, 0, &i))) |
1da177e4 LT |
1697 | goto error0; |
1698 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1699 | #ifdef DEBUG | |
1700 | /* | |
1701 | * Check that this is the right record: delete didn't | |
1702 | * mangle the cursor. | |
1703 | */ | |
1704 | { | |
1705 | xfs_agblock_t xxbno; | |
1706 | xfs_extlen_t xxlen; | |
1707 | ||
1708 | if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen, | |
1709 | &i))) | |
1710 | goto error0; | |
1711 | XFS_WANT_CORRUPTED_GOTO( | |
1712 | i == 1 && xxbno == ltbno && xxlen == ltlen, | |
1713 | error0); | |
1714 | } | |
1715 | #endif | |
1716 | /* | |
1717 | * Update remaining by-block entry to the new, joined block. | |
1718 | */ | |
1719 | nbno = ltbno; | |
1720 | nlen = len + ltlen + gtlen; | |
1721 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
1722 | goto error0; | |
1723 | } | |
1724 | /* | |
1725 | * Have only a left contiguous neighbor. | |
1726 | * Merge it together with the new freespace. | |
1727 | */ | |
1728 | else if (haveleft) { | |
1729 | /* | |
1730 | * Delete the old by-size entry on the left. | |
1731 | */ | |
1732 | if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i))) | |
1733 | goto error0; | |
1734 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
91cca5df | 1735 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 LT |
1736 | goto error0; |
1737 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1738 | /* | |
1739 | * Back up the by-block cursor to the left neighbor, and | |
1740 | * update its length. | |
1741 | */ | |
8df4da4a | 1742 | if ((error = xfs_btree_decrement(bno_cur, 0, &i))) |
1da177e4 LT |
1743 | goto error0; |
1744 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1745 | nbno = ltbno; | |
1746 | nlen = len + ltlen; | |
1747 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
1748 | goto error0; | |
1749 | } | |
1750 | /* | |
1751 | * Have only a right contiguous neighbor. | |
1752 | * Merge it together with the new freespace. | |
1753 | */ | |
1754 | else if (haveright) { | |
1755 | /* | |
1756 | * Delete the old by-size entry on the right. | |
1757 | */ | |
1758 | if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i))) | |
1759 | goto error0; | |
1760 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
91cca5df | 1761 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 LT |
1762 | goto error0; |
1763 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1764 | /* | |
1765 | * Update the starting block and length of the right | |
1766 | * neighbor in the by-block tree. | |
1767 | */ | |
1768 | nbno = bno; | |
1769 | nlen = len + gtlen; | |
1770 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
1771 | goto error0; | |
1772 | } | |
1773 | /* | |
1774 | * No contiguous neighbors. | |
1775 | * Insert the new freespace into the by-block tree. | |
1776 | */ | |
1777 | else { | |
1778 | nbno = bno; | |
1779 | nlen = len; | |
4b22a571 | 1780 | if ((error = xfs_btree_insert(bno_cur, &i))) |
1da177e4 LT |
1781 | goto error0; |
1782 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1783 | } | |
1784 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
1785 | bno_cur = NULL; | |
1786 | /* | |
1787 | * In all cases we need to insert the new freespace in the by-size tree. | |
1788 | */ | |
1789 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i))) | |
1790 | goto error0; | |
1791 | XFS_WANT_CORRUPTED_GOTO(i == 0, error0); | |
4b22a571 | 1792 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
1da177e4 LT |
1793 | goto error0; |
1794 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1795 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1796 | cnt_cur = NULL; | |
1797 | /* | |
1798 | * Update the freespace totals in the ag and superblock. | |
1799 | */ | |
1800 | { | |
1801 | xfs_agf_t *agf; | |
1802 | xfs_perag_t *pag; /* per allocation group data */ | |
1803 | ||
1804 | agf = XFS_BUF_TO_AGF(agbp); | |
1805 | pag = &mp->m_perag[agno]; | |
413d57c9 | 1806 | be32_add_cpu(&agf->agf_freeblks, len); |
1da177e4 LT |
1807 | xfs_trans_agblocks_delta(tp, len); |
1808 | pag->pagf_freeblks += len; | |
1809 | XFS_WANT_CORRUPTED_GOTO( | |
16259e7d CH |
1810 | be32_to_cpu(agf->agf_freeblks) <= |
1811 | be32_to_cpu(agf->agf_length), | |
1da177e4 LT |
1812 | error0); |
1813 | TRACE_MODAGF(NULL, agf, XFS_AGF_FREEBLKS); | |
1814 | xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS); | |
1815 | if (!isfl) | |
1816 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, (long)len); | |
1817 | XFS_STATS_INC(xs_freex); | |
1818 | XFS_STATS_ADD(xs_freeb, len); | |
1819 | } | |
1820 | TRACE_FREE(haveleft ? | |
1821 | (haveright ? "both" : "left") : | |
1822 | (haveright ? "right" : "none"), | |
1823 | agno, bno, len, isfl); | |
1824 | ||
1825 | /* | |
1826 | * Since blocks move to the free list without the coordination | |
1827 | * used in xfs_bmap_finish, we can't allow block to be available | |
1828 | * for reallocation and non-transaction writing (user data) | |
1829 | * until we know that the transaction that moved it to the free | |
1830 | * list is permanently on disk. We track the blocks by declaring | |
1831 | * these blocks as "busy"; the busy list is maintained on a per-ag | |
1832 | * basis and each transaction records which entries should be removed | |
1833 | * when the iclog commits to disk. If a busy block is allocated, | |
1834 | * the iclog is pushed up to the LSN that freed the block. | |
1835 | */ | |
1836 | xfs_alloc_mark_busy(tp, agno, bno, len); | |
1837 | return 0; | |
1838 | ||
1839 | error0: | |
1840 | TRACE_FREE("error", agno, bno, len, isfl); | |
1841 | if (bno_cur) | |
1842 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
1843 | if (cnt_cur) | |
1844 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1845 | return error; | |
1846 | } | |
1847 | ||
1848 | /* | |
1849 | * Visible (exported) allocation/free functions. | |
1850 | * Some of these are used just by xfs_alloc_btree.c and this file. | |
1851 | */ | |
1852 | ||
1853 | /* | |
1854 | * Compute and fill in value of m_ag_maxlevels. | |
1855 | */ | |
1856 | void | |
1857 | xfs_alloc_compute_maxlevels( | |
1858 | xfs_mount_t *mp) /* file system mount structure */ | |
1859 | { | |
1860 | int level; | |
1861 | uint maxblocks; | |
1862 | uint maxleafents; | |
1863 | int minleafrecs; | |
1864 | int minnoderecs; | |
1865 | ||
1866 | maxleafents = (mp->m_sb.sb_agblocks + 1) / 2; | |
1867 | minleafrecs = mp->m_alloc_mnr[0]; | |
1868 | minnoderecs = mp->m_alloc_mnr[1]; | |
1869 | maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs; | |
1870 | for (level = 1; maxblocks > 1; level++) | |
1871 | maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs; | |
1872 | mp->m_ag_maxlevels = level; | |
1873 | } | |
1874 | ||
1875 | /* | |
1876 | * Decide whether to use this allocation group for this allocation. | |
1877 | * If so, fix up the btree freelist's size. | |
1878 | */ | |
1879 | STATIC int /* error */ | |
1880 | xfs_alloc_fix_freelist( | |
1881 | xfs_alloc_arg_t *args, /* allocation argument structure */ | |
1882 | int flags) /* XFS_ALLOC_FLAG_... */ | |
1883 | { | |
1884 | xfs_buf_t *agbp; /* agf buffer pointer */ | |
1885 | xfs_agf_t *agf; /* a.g. freespace structure pointer */ | |
1886 | xfs_buf_t *agflbp;/* agfl buffer pointer */ | |
1887 | xfs_agblock_t bno; /* freelist block */ | |
1888 | xfs_extlen_t delta; /* new blocks needed in freelist */ | |
1889 | int error; /* error result code */ | |
1890 | xfs_extlen_t longest;/* longest extent in allocation group */ | |
1891 | xfs_mount_t *mp; /* file system mount point structure */ | |
1892 | xfs_extlen_t need; /* total blocks needed in freelist */ | |
1893 | xfs_perag_t *pag; /* per-ag information structure */ | |
1894 | xfs_alloc_arg_t targs; /* local allocation arguments */ | |
1895 | xfs_trans_t *tp; /* transaction pointer */ | |
1896 | ||
1897 | mp = args->mp; | |
1898 | ||
1899 | pag = args->pag; | |
1900 | tp = args->tp; | |
1901 | if (!pag->pagf_init) { | |
1902 | if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags, | |
1903 | &agbp))) | |
1904 | return error; | |
1905 | if (!pag->pagf_init) { | |
0e1edbd9 NS |
1906 | ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK); |
1907 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
1da177e4 LT |
1908 | args->agbp = NULL; |
1909 | return 0; | |
1910 | } | |
1911 | } else | |
1912 | agbp = NULL; | |
1913 | ||
0e1edbd9 NS |
1914 | /* |
1915 | * If this is a metadata preferred pag and we are user data | |
1da177e4 LT |
1916 | * then try somewhere else if we are not being asked to |
1917 | * try harder at this point | |
1918 | */ | |
0e1edbd9 NS |
1919 | if (pag->pagf_metadata && args->userdata && |
1920 | (flags & XFS_ALLOC_FLAG_TRYLOCK)) { | |
1921 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
1da177e4 LT |
1922 | args->agbp = NULL; |
1923 | return 0; | |
1924 | } | |
1925 | ||
0e1edbd9 NS |
1926 | if (!(flags & XFS_ALLOC_FLAG_FREEING)) { |
1927 | need = XFS_MIN_FREELIST_PAG(pag, mp); | |
1928 | delta = need > pag->pagf_flcount ? need - pag->pagf_flcount : 0; | |
1929 | /* | |
1930 | * If it looks like there isn't a long enough extent, or enough | |
1931 | * total blocks, reject it. | |
1932 | */ | |
1933 | longest = (pag->pagf_longest > delta) ? | |
1934 | (pag->pagf_longest - delta) : | |
1935 | (pag->pagf_flcount > 0 || pag->pagf_longest > 0); | |
1936 | if ((args->minlen + args->alignment + args->minalignslop - 1) > | |
1937 | longest || | |
1938 | ((int)(pag->pagf_freeblks + pag->pagf_flcount - | |
1939 | need - args->total) < (int)args->minleft)) { | |
1940 | if (agbp) | |
1941 | xfs_trans_brelse(tp, agbp); | |
1942 | args->agbp = NULL; | |
1943 | return 0; | |
1944 | } | |
1da177e4 | 1945 | } |
0e1edbd9 | 1946 | |
1da177e4 LT |
1947 | /* |
1948 | * Get the a.g. freespace buffer. | |
1949 | * Can fail if we're not blocking on locks, and it's held. | |
1950 | */ | |
1951 | if (agbp == NULL) { | |
1952 | if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags, | |
1953 | &agbp))) | |
1954 | return error; | |
1955 | if (agbp == NULL) { | |
0e1edbd9 NS |
1956 | ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK); |
1957 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
1da177e4 LT |
1958 | args->agbp = NULL; |
1959 | return 0; | |
1960 | } | |
1961 | } | |
1962 | /* | |
1963 | * Figure out how many blocks we should have in the freelist. | |
1964 | */ | |
1965 | agf = XFS_BUF_TO_AGF(agbp); | |
1966 | need = XFS_MIN_FREELIST(agf, mp); | |
1da177e4 LT |
1967 | /* |
1968 | * If there isn't enough total or single-extent, reject it. | |
1969 | */ | |
0e1edbd9 NS |
1970 | if (!(flags & XFS_ALLOC_FLAG_FREEING)) { |
1971 | delta = need > be32_to_cpu(agf->agf_flcount) ? | |
1972 | (need - be32_to_cpu(agf->agf_flcount)) : 0; | |
1973 | longest = be32_to_cpu(agf->agf_longest); | |
1974 | longest = (longest > delta) ? (longest - delta) : | |
1975 | (be32_to_cpu(agf->agf_flcount) > 0 || longest > 0); | |
1976 | if ((args->minlen + args->alignment + args->minalignslop - 1) > | |
1977 | longest || | |
1978 | ((int)(be32_to_cpu(agf->agf_freeblks) + | |
1979 | be32_to_cpu(agf->agf_flcount) - need - args->total) < | |
1980 | (int)args->minleft)) { | |
1981 | xfs_trans_brelse(tp, agbp); | |
1982 | args->agbp = NULL; | |
1983 | return 0; | |
1984 | } | |
1da177e4 LT |
1985 | } |
1986 | /* | |
1987 | * Make the freelist shorter if it's too long. | |
1988 | */ | |
16259e7d | 1989 | while (be32_to_cpu(agf->agf_flcount) > need) { |
1da177e4 LT |
1990 | xfs_buf_t *bp; |
1991 | ||
92821e2b DC |
1992 | error = xfs_alloc_get_freelist(tp, agbp, &bno, 0); |
1993 | if (error) | |
1da177e4 LT |
1994 | return error; |
1995 | if ((error = xfs_free_ag_extent(tp, agbp, args->agno, bno, 1, 1))) | |
1996 | return error; | |
1997 | bp = xfs_btree_get_bufs(mp, tp, args->agno, bno, 0); | |
1998 | xfs_trans_binval(tp, bp); | |
1999 | } | |
2000 | /* | |
2001 | * Initialize the args structure. | |
2002 | */ | |
2003 | targs.tp = tp; | |
2004 | targs.mp = mp; | |
2005 | targs.agbp = agbp; | |
2006 | targs.agno = args->agno; | |
2007 | targs.mod = targs.minleft = targs.wasdel = targs.userdata = | |
2008 | targs.minalignslop = 0; | |
2009 | targs.alignment = targs.minlen = targs.prod = targs.isfl = 1; | |
2010 | targs.type = XFS_ALLOCTYPE_THIS_AG; | |
2011 | targs.pag = pag; | |
2012 | if ((error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp))) | |
2013 | return error; | |
2014 | /* | |
2015 | * Make the freelist longer if it's too short. | |
2016 | */ | |
16259e7d | 2017 | while (be32_to_cpu(agf->agf_flcount) < need) { |
1da177e4 | 2018 | targs.agbno = 0; |
16259e7d | 2019 | targs.maxlen = need - be32_to_cpu(agf->agf_flcount); |
1da177e4 LT |
2020 | /* |
2021 | * Allocate as many blocks as possible at once. | |
2022 | */ | |
e63a3690 NS |
2023 | if ((error = xfs_alloc_ag_vextent(&targs))) { |
2024 | xfs_trans_brelse(tp, agflbp); | |
1da177e4 | 2025 | return error; |
e63a3690 | 2026 | } |
1da177e4 LT |
2027 | /* |
2028 | * Stop if we run out. Won't happen if callers are obeying | |
2029 | * the restrictions correctly. Can happen for free calls | |
2030 | * on a completely full ag. | |
2031 | */ | |
d210a28c | 2032 | if (targs.agbno == NULLAGBLOCK) { |
0e1edbd9 NS |
2033 | if (flags & XFS_ALLOC_FLAG_FREEING) |
2034 | break; | |
2035 | xfs_trans_brelse(tp, agflbp); | |
2036 | args->agbp = NULL; | |
2037 | return 0; | |
d210a28c | 2038 | } |
1da177e4 LT |
2039 | /* |
2040 | * Put each allocated block on the list. | |
2041 | */ | |
2042 | for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) { | |
92821e2b DC |
2043 | error = xfs_alloc_put_freelist(tp, agbp, |
2044 | agflbp, bno, 0); | |
2045 | if (error) | |
1da177e4 LT |
2046 | return error; |
2047 | } | |
2048 | } | |
e63a3690 | 2049 | xfs_trans_brelse(tp, agflbp); |
1da177e4 LT |
2050 | args->agbp = agbp; |
2051 | return 0; | |
2052 | } | |
2053 | ||
2054 | /* | |
2055 | * Get a block from the freelist. | |
2056 | * Returns with the buffer for the block gotten. | |
2057 | */ | |
2058 | int /* error */ | |
2059 | xfs_alloc_get_freelist( | |
2060 | xfs_trans_t *tp, /* transaction pointer */ | |
2061 | xfs_buf_t *agbp, /* buffer containing the agf structure */ | |
92821e2b DC |
2062 | xfs_agblock_t *bnop, /* block address retrieved from freelist */ |
2063 | int btreeblk) /* destination is a AGF btree */ | |
1da177e4 LT |
2064 | { |
2065 | xfs_agf_t *agf; /* a.g. freespace structure */ | |
2066 | xfs_agfl_t *agfl; /* a.g. freelist structure */ | |
2067 | xfs_buf_t *agflbp;/* buffer for a.g. freelist structure */ | |
2068 | xfs_agblock_t bno; /* block number returned */ | |
2069 | int error; | |
92821e2b | 2070 | int logflags; |
1da177e4 LT |
2071 | xfs_mount_t *mp; /* mount structure */ |
2072 | xfs_perag_t *pag; /* per allocation group data */ | |
2073 | ||
2074 | agf = XFS_BUF_TO_AGF(agbp); | |
2075 | /* | |
2076 | * Freelist is empty, give up. | |
2077 | */ | |
2078 | if (!agf->agf_flcount) { | |
2079 | *bnop = NULLAGBLOCK; | |
2080 | return 0; | |
2081 | } | |
2082 | /* | |
2083 | * Read the array of free blocks. | |
2084 | */ | |
2085 | mp = tp->t_mountp; | |
2086 | if ((error = xfs_alloc_read_agfl(mp, tp, | |
16259e7d | 2087 | be32_to_cpu(agf->agf_seqno), &agflbp))) |
1da177e4 LT |
2088 | return error; |
2089 | agfl = XFS_BUF_TO_AGFL(agflbp); | |
2090 | /* | |
2091 | * Get the block number and update the data structures. | |
2092 | */ | |
e2101005 | 2093 | bno = be32_to_cpu(agfl->agfl_bno[be32_to_cpu(agf->agf_flfirst)]); |
413d57c9 | 2094 | be32_add_cpu(&agf->agf_flfirst, 1); |
1da177e4 | 2095 | xfs_trans_brelse(tp, agflbp); |
16259e7d | 2096 | if (be32_to_cpu(agf->agf_flfirst) == XFS_AGFL_SIZE(mp)) |
1da177e4 | 2097 | agf->agf_flfirst = 0; |
16259e7d | 2098 | pag = &mp->m_perag[be32_to_cpu(agf->agf_seqno)]; |
413d57c9 | 2099 | be32_add_cpu(&agf->agf_flcount, -1); |
1da177e4 LT |
2100 | xfs_trans_agflist_delta(tp, -1); |
2101 | pag->pagf_flcount--; | |
92821e2b DC |
2102 | |
2103 | logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT; | |
2104 | if (btreeblk) { | |
413d57c9 | 2105 | be32_add_cpu(&agf->agf_btreeblks, 1); |
92821e2b DC |
2106 | pag->pagf_btreeblks++; |
2107 | logflags |= XFS_AGF_BTREEBLKS; | |
2108 | } | |
2109 | ||
2110 | TRACE_MODAGF(NULL, agf, logflags); | |
2111 | xfs_alloc_log_agf(tp, agbp, logflags); | |
1da177e4 LT |
2112 | *bnop = bno; |
2113 | ||
2114 | /* | |
2115 | * As blocks are freed, they are added to the per-ag busy list | |
2116 | * and remain there until the freeing transaction is committed to | |
2117 | * disk. Now that we have allocated blocks, this list must be | |
2118 | * searched to see if a block is being reused. If one is, then | |
2119 | * the freeing transaction must be pushed to disk NOW by forcing | |
2120 | * to disk all iclogs up that transaction's LSN. | |
2121 | */ | |
16259e7d | 2122 | xfs_alloc_search_busy(tp, be32_to_cpu(agf->agf_seqno), bno, 1); |
1da177e4 LT |
2123 | return 0; |
2124 | } | |
2125 | ||
2126 | /* | |
2127 | * Log the given fields from the agf structure. | |
2128 | */ | |
2129 | void | |
2130 | xfs_alloc_log_agf( | |
2131 | xfs_trans_t *tp, /* transaction pointer */ | |
2132 | xfs_buf_t *bp, /* buffer for a.g. freelist header */ | |
2133 | int fields) /* mask of fields to be logged (XFS_AGF_...) */ | |
2134 | { | |
2135 | int first; /* first byte offset */ | |
2136 | int last; /* last byte offset */ | |
2137 | static const short offsets[] = { | |
2138 | offsetof(xfs_agf_t, agf_magicnum), | |
2139 | offsetof(xfs_agf_t, agf_versionnum), | |
2140 | offsetof(xfs_agf_t, agf_seqno), | |
2141 | offsetof(xfs_agf_t, agf_length), | |
2142 | offsetof(xfs_agf_t, agf_roots[0]), | |
2143 | offsetof(xfs_agf_t, agf_levels[0]), | |
2144 | offsetof(xfs_agf_t, agf_flfirst), | |
2145 | offsetof(xfs_agf_t, agf_fllast), | |
2146 | offsetof(xfs_agf_t, agf_flcount), | |
2147 | offsetof(xfs_agf_t, agf_freeblks), | |
2148 | offsetof(xfs_agf_t, agf_longest), | |
92821e2b | 2149 | offsetof(xfs_agf_t, agf_btreeblks), |
1da177e4 LT |
2150 | sizeof(xfs_agf_t) |
2151 | }; | |
2152 | ||
2153 | xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last); | |
2154 | xfs_trans_log_buf(tp, bp, (uint)first, (uint)last); | |
2155 | } | |
2156 | ||
2157 | /* | |
2158 | * Interface for inode allocation to force the pag data to be initialized. | |
2159 | */ | |
2160 | int /* error */ | |
2161 | xfs_alloc_pagf_init( | |
2162 | xfs_mount_t *mp, /* file system mount structure */ | |
2163 | xfs_trans_t *tp, /* transaction pointer */ | |
2164 | xfs_agnumber_t agno, /* allocation group number */ | |
2165 | int flags) /* XFS_ALLOC_FLAGS_... */ | |
2166 | { | |
2167 | xfs_buf_t *bp; | |
2168 | int error; | |
2169 | ||
2170 | if ((error = xfs_alloc_read_agf(mp, tp, agno, flags, &bp))) | |
2171 | return error; | |
2172 | if (bp) | |
2173 | xfs_trans_brelse(tp, bp); | |
2174 | return 0; | |
2175 | } | |
2176 | ||
2177 | /* | |
2178 | * Put the block on the freelist for the allocation group. | |
2179 | */ | |
2180 | int /* error */ | |
2181 | xfs_alloc_put_freelist( | |
2182 | xfs_trans_t *tp, /* transaction pointer */ | |
2183 | xfs_buf_t *agbp, /* buffer for a.g. freelist header */ | |
2184 | xfs_buf_t *agflbp,/* buffer for a.g. free block array */ | |
92821e2b DC |
2185 | xfs_agblock_t bno, /* block being freed */ |
2186 | int btreeblk) /* block came from a AGF btree */ | |
1da177e4 LT |
2187 | { |
2188 | xfs_agf_t *agf; /* a.g. freespace structure */ | |
2189 | xfs_agfl_t *agfl; /* a.g. free block array */ | |
e2101005 | 2190 | __be32 *blockp;/* pointer to array entry */ |
1da177e4 | 2191 | int error; |
92821e2b | 2192 | int logflags; |
1da177e4 LT |
2193 | xfs_mount_t *mp; /* mount structure */ |
2194 | xfs_perag_t *pag; /* per allocation group data */ | |
2195 | ||
2196 | agf = XFS_BUF_TO_AGF(agbp); | |
2197 | mp = tp->t_mountp; | |
2198 | ||
2199 | if (!agflbp && (error = xfs_alloc_read_agfl(mp, tp, | |
16259e7d | 2200 | be32_to_cpu(agf->agf_seqno), &agflbp))) |
1da177e4 LT |
2201 | return error; |
2202 | agfl = XFS_BUF_TO_AGFL(agflbp); | |
413d57c9 | 2203 | be32_add_cpu(&agf->agf_fllast, 1); |
16259e7d | 2204 | if (be32_to_cpu(agf->agf_fllast) == XFS_AGFL_SIZE(mp)) |
1da177e4 | 2205 | agf->agf_fllast = 0; |
16259e7d | 2206 | pag = &mp->m_perag[be32_to_cpu(agf->agf_seqno)]; |
413d57c9 | 2207 | be32_add_cpu(&agf->agf_flcount, 1); |
1da177e4 LT |
2208 | xfs_trans_agflist_delta(tp, 1); |
2209 | pag->pagf_flcount++; | |
92821e2b DC |
2210 | |
2211 | logflags = XFS_AGF_FLLAST | XFS_AGF_FLCOUNT; | |
2212 | if (btreeblk) { | |
413d57c9 | 2213 | be32_add_cpu(&agf->agf_btreeblks, -1); |
92821e2b DC |
2214 | pag->pagf_btreeblks--; |
2215 | logflags |= XFS_AGF_BTREEBLKS; | |
2216 | } | |
2217 | ||
2218 | TRACE_MODAGF(NULL, agf, logflags); | |
2219 | xfs_alloc_log_agf(tp, agbp, logflags); | |
2220 | ||
16259e7d CH |
2221 | ASSERT(be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp)); |
2222 | blockp = &agfl->agfl_bno[be32_to_cpu(agf->agf_fllast)]; | |
e2101005 | 2223 | *blockp = cpu_to_be32(bno); |
92821e2b DC |
2224 | TRACE_MODAGF(NULL, agf, logflags); |
2225 | xfs_alloc_log_agf(tp, agbp, logflags); | |
1da177e4 LT |
2226 | xfs_trans_log_buf(tp, agflbp, |
2227 | (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl), | |
2228 | (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl + | |
2229 | sizeof(xfs_agblock_t) - 1)); | |
2230 | return 0; | |
2231 | } | |
2232 | ||
2233 | /* | |
2234 | * Read in the allocation group header (free/alloc section). | |
2235 | */ | |
2236 | int /* error */ | |
2237 | xfs_alloc_read_agf( | |
2238 | xfs_mount_t *mp, /* mount point structure */ | |
2239 | xfs_trans_t *tp, /* transaction pointer */ | |
2240 | xfs_agnumber_t agno, /* allocation group number */ | |
2241 | int flags, /* XFS_ALLOC_FLAG_... */ | |
2242 | xfs_buf_t **bpp) /* buffer for the ag freelist header */ | |
2243 | { | |
2244 | xfs_agf_t *agf; /* ag freelist header */ | |
2245 | int agf_ok; /* set if agf is consistent */ | |
2246 | xfs_buf_t *bp; /* return value */ | |
2247 | xfs_perag_t *pag; /* per allocation group data */ | |
2248 | int error; | |
2249 | ||
2250 | ASSERT(agno != NULLAGNUMBER); | |
2251 | error = xfs_trans_read_buf( | |
2252 | mp, tp, mp->m_ddev_targp, | |
2253 | XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)), | |
2254 | XFS_FSS_TO_BB(mp, 1), | |
2255 | (flags & XFS_ALLOC_FLAG_TRYLOCK) ? XFS_BUF_TRYLOCK : 0U, | |
2256 | &bp); | |
2257 | if (error) | |
2258 | return error; | |
2259 | ASSERT(!bp || !XFS_BUF_GETERROR(bp)); | |
2260 | if (!bp) { | |
2261 | *bpp = NULL; | |
2262 | return 0; | |
2263 | } | |
2264 | /* | |
2265 | * Validate the magic number of the agf block. | |
2266 | */ | |
2267 | agf = XFS_BUF_TO_AGF(bp); | |
2268 | agf_ok = | |
16259e7d CH |
2269 | be32_to_cpu(agf->agf_magicnum) == XFS_AGF_MAGIC && |
2270 | XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) && | |
2271 | be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) && | |
2272 | be32_to_cpu(agf->agf_flfirst) < XFS_AGFL_SIZE(mp) && | |
2273 | be32_to_cpu(agf->agf_fllast) < XFS_AGFL_SIZE(mp) && | |
2274 | be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp); | |
89b28393 BN |
2275 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
2276 | agf_ok = agf_ok && be32_to_cpu(agf->agf_btreeblks) <= | |
2277 | be32_to_cpu(agf->agf_length); | |
1da177e4 LT |
2278 | if (unlikely(XFS_TEST_ERROR(!agf_ok, mp, XFS_ERRTAG_ALLOC_READ_AGF, |
2279 | XFS_RANDOM_ALLOC_READ_AGF))) { | |
2280 | XFS_CORRUPTION_ERROR("xfs_alloc_read_agf", | |
2281 | XFS_ERRLEVEL_LOW, mp, agf); | |
2282 | xfs_trans_brelse(tp, bp); | |
2283 | return XFS_ERROR(EFSCORRUPTED); | |
2284 | } | |
2285 | pag = &mp->m_perag[agno]; | |
2286 | if (!pag->pagf_init) { | |
16259e7d | 2287 | pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks); |
92821e2b | 2288 | pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks); |
16259e7d CH |
2289 | pag->pagf_flcount = be32_to_cpu(agf->agf_flcount); |
2290 | pag->pagf_longest = be32_to_cpu(agf->agf_longest); | |
1da177e4 | 2291 | pag->pagf_levels[XFS_BTNUM_BNOi] = |
16259e7d | 2292 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]); |
1da177e4 | 2293 | pag->pagf_levels[XFS_BTNUM_CNTi] = |
16259e7d | 2294 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]); |
007c61c6 | 2295 | spin_lock_init(&pag->pagb_lock); |
1da177e4 LT |
2296 | pag->pagb_list = kmem_zalloc(XFS_PAGB_NUM_SLOTS * |
2297 | sizeof(xfs_perag_busy_t), KM_SLEEP); | |
2298 | pag->pagf_init = 1; | |
2299 | } | |
2300 | #ifdef DEBUG | |
2301 | else if (!XFS_FORCED_SHUTDOWN(mp)) { | |
16259e7d | 2302 | ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks)); |
89b28393 | 2303 | ASSERT(pag->pagf_btreeblks == be32_to_cpu(agf->agf_btreeblks)); |
16259e7d CH |
2304 | ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount)); |
2305 | ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest)); | |
1da177e4 | 2306 | ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] == |
16259e7d | 2307 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi])); |
1da177e4 | 2308 | ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] == |
16259e7d | 2309 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi])); |
1da177e4 LT |
2310 | } |
2311 | #endif | |
2312 | XFS_BUF_SET_VTYPE_REF(bp, B_FS_AGF, XFS_AGF_REF); | |
2313 | *bpp = bp; | |
2314 | return 0; | |
2315 | } | |
2316 | ||
2317 | /* | |
2318 | * Allocate an extent (variable-size). | |
2319 | * Depending on the allocation type, we either look in a single allocation | |
2320 | * group or loop over the allocation groups to find the result. | |
2321 | */ | |
2322 | int /* error */ | |
2323 | xfs_alloc_vextent( | |
2324 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
2325 | { | |
2326 | xfs_agblock_t agsize; /* allocation group size */ | |
2327 | int error; | |
2328 | int flags; /* XFS_ALLOC_FLAG_... locking flags */ | |
1da177e4 LT |
2329 | xfs_extlen_t minleft;/* minimum left value, temp copy */ |
2330 | xfs_mount_t *mp; /* mount structure pointer */ | |
2331 | xfs_agnumber_t sagno; /* starting allocation group number */ | |
2332 | xfs_alloctype_t type; /* input allocation type */ | |
2333 | int bump_rotor = 0; | |
2334 | int no_min = 0; | |
2335 | xfs_agnumber_t rotorstep = xfs_rotorstep; /* inode32 agf stepper */ | |
2336 | ||
2337 | mp = args->mp; | |
2338 | type = args->otype = args->type; | |
2339 | args->agbno = NULLAGBLOCK; | |
2340 | /* | |
2341 | * Just fix this up, for the case where the last a.g. is shorter | |
2342 | * (or there's only one a.g.) and the caller couldn't easily figure | |
2343 | * that out (xfs_bmap_alloc). | |
2344 | */ | |
2345 | agsize = mp->m_sb.sb_agblocks; | |
2346 | if (args->maxlen > agsize) | |
2347 | args->maxlen = agsize; | |
2348 | if (args->alignment == 0) | |
2349 | args->alignment = 1; | |
2350 | ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount); | |
2351 | ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize); | |
2352 | ASSERT(args->minlen <= args->maxlen); | |
2353 | ASSERT(args->minlen <= agsize); | |
2354 | ASSERT(args->mod < args->prod); | |
2355 | if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount || | |
2356 | XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize || | |
2357 | args->minlen > args->maxlen || args->minlen > agsize || | |
2358 | args->mod >= args->prod) { | |
2359 | args->fsbno = NULLFSBLOCK; | |
2360 | TRACE_ALLOC("badargs", args); | |
2361 | return 0; | |
2362 | } | |
2363 | minleft = args->minleft; | |
2364 | ||
2365 | switch (type) { | |
2366 | case XFS_ALLOCTYPE_THIS_AG: | |
2367 | case XFS_ALLOCTYPE_NEAR_BNO: | |
2368 | case XFS_ALLOCTYPE_THIS_BNO: | |
2369 | /* | |
2370 | * These three force us into a single a.g. | |
2371 | */ | |
2372 | args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
2373 | down_read(&mp->m_peraglock); | |
2374 | args->pag = &mp->m_perag[args->agno]; | |
2375 | args->minleft = 0; | |
2376 | error = xfs_alloc_fix_freelist(args, 0); | |
2377 | args->minleft = minleft; | |
2378 | if (error) { | |
2379 | TRACE_ALLOC("nofix", args); | |
2380 | goto error0; | |
2381 | } | |
2382 | if (!args->agbp) { | |
2383 | up_read(&mp->m_peraglock); | |
2384 | TRACE_ALLOC("noagbp", args); | |
2385 | break; | |
2386 | } | |
2387 | args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno); | |
2388 | if ((error = xfs_alloc_ag_vextent(args))) | |
2389 | goto error0; | |
2390 | up_read(&mp->m_peraglock); | |
2391 | break; | |
2392 | case XFS_ALLOCTYPE_START_BNO: | |
2393 | /* | |
2394 | * Try near allocation first, then anywhere-in-ag after | |
2395 | * the first a.g. fails. | |
2396 | */ | |
2397 | if ((args->userdata == XFS_ALLOC_INITIAL_USER_DATA) && | |
2398 | (mp->m_flags & XFS_MOUNT_32BITINODES)) { | |
2399 | args->fsbno = XFS_AGB_TO_FSB(mp, | |
2400 | ((mp->m_agfrotor / rotorstep) % | |
2401 | mp->m_sb.sb_agcount), 0); | |
2402 | bump_rotor = 1; | |
2403 | } | |
2404 | args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno); | |
2405 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
2406 | /* FALLTHROUGH */ | |
2407 | case XFS_ALLOCTYPE_ANY_AG: | |
2408 | case XFS_ALLOCTYPE_START_AG: | |
2409 | case XFS_ALLOCTYPE_FIRST_AG: | |
2410 | /* | |
2411 | * Rotate through the allocation groups looking for a winner. | |
2412 | */ | |
2413 | if (type == XFS_ALLOCTYPE_ANY_AG) { | |
2414 | /* | |
2415 | * Start with the last place we left off. | |
2416 | */ | |
2417 | args->agno = sagno = (mp->m_agfrotor / rotorstep) % | |
2418 | mp->m_sb.sb_agcount; | |
2419 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2420 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
2421 | } else if (type == XFS_ALLOCTYPE_FIRST_AG) { | |
2422 | /* | |
2423 | * Start with allocation group given by bno. | |
2424 | */ | |
2425 | args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
2426 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2427 | sagno = 0; | |
2428 | flags = 0; | |
2429 | } else { | |
2430 | if (type == XFS_ALLOCTYPE_START_AG) | |
2431 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2432 | /* | |
2433 | * Start with the given allocation group. | |
2434 | */ | |
2435 | args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
2436 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
2437 | } | |
2438 | /* | |
2439 | * Loop over allocation groups twice; first time with | |
2440 | * trylock set, second time without. | |
2441 | */ | |
2442 | down_read(&mp->m_peraglock); | |
2443 | for (;;) { | |
2444 | args->pag = &mp->m_perag[args->agno]; | |
2445 | if (no_min) args->minleft = 0; | |
2446 | error = xfs_alloc_fix_freelist(args, flags); | |
2447 | args->minleft = minleft; | |
2448 | if (error) { | |
2449 | TRACE_ALLOC("nofix", args); | |
2450 | goto error0; | |
2451 | } | |
2452 | /* | |
2453 | * If we get a buffer back then the allocation will fly. | |
2454 | */ | |
2455 | if (args->agbp) { | |
2456 | if ((error = xfs_alloc_ag_vextent(args))) | |
2457 | goto error0; | |
2458 | break; | |
2459 | } | |
2460 | TRACE_ALLOC("loopfailed", args); | |
2461 | /* | |
2462 | * Didn't work, figure out the next iteration. | |
2463 | */ | |
2464 | if (args->agno == sagno && | |
2465 | type == XFS_ALLOCTYPE_START_BNO) | |
2466 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
d210a28c YL |
2467 | /* |
2468 | * For the first allocation, we can try any AG to get | |
2469 | * space. However, if we already have allocated a | |
2470 | * block, we don't want to try AGs whose number is below | |
2471 | * sagno. Otherwise, we may end up with out-of-order | |
2472 | * locking of AGF, which might cause deadlock. | |
2473 | */ | |
2474 | if (++(args->agno) == mp->m_sb.sb_agcount) { | |
2475 | if (args->firstblock != NULLFSBLOCK) | |
2476 | args->agno = sagno; | |
2477 | else | |
2478 | args->agno = 0; | |
2479 | } | |
1da177e4 LT |
2480 | /* |
2481 | * Reached the starting a.g., must either be done | |
2482 | * or switch to non-trylock mode. | |
2483 | */ | |
2484 | if (args->agno == sagno) { | |
2485 | if (no_min == 1) { | |
2486 | args->agbno = NULLAGBLOCK; | |
2487 | TRACE_ALLOC("allfailed", args); | |
2488 | break; | |
2489 | } | |
2490 | if (flags == 0) { | |
2491 | no_min = 1; | |
2492 | } else { | |
2493 | flags = 0; | |
2494 | if (type == XFS_ALLOCTYPE_START_BNO) { | |
2495 | args->agbno = XFS_FSB_TO_AGBNO(mp, | |
2496 | args->fsbno); | |
2497 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
2498 | } | |
2499 | } | |
2500 | } | |
2501 | } | |
2502 | up_read(&mp->m_peraglock); | |
2503 | if (bump_rotor || (type == XFS_ALLOCTYPE_ANY_AG)) { | |
2504 | if (args->agno == sagno) | |
2505 | mp->m_agfrotor = (mp->m_agfrotor + 1) % | |
2506 | (mp->m_sb.sb_agcount * rotorstep); | |
2507 | else | |
2508 | mp->m_agfrotor = (args->agno * rotorstep + 1) % | |
2509 | (mp->m_sb.sb_agcount * rotorstep); | |
2510 | } | |
2511 | break; | |
2512 | default: | |
2513 | ASSERT(0); | |
2514 | /* NOTREACHED */ | |
2515 | } | |
2516 | if (args->agbno == NULLAGBLOCK) | |
2517 | args->fsbno = NULLFSBLOCK; | |
2518 | else { | |
2519 | args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno); | |
2520 | #ifdef DEBUG | |
2521 | ASSERT(args->len >= args->minlen); | |
2522 | ASSERT(args->len <= args->maxlen); | |
2523 | ASSERT(args->agbno % args->alignment == 0); | |
2524 | XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno), | |
2525 | args->len); | |
2526 | #endif | |
2527 | } | |
2528 | return 0; | |
2529 | error0: | |
2530 | up_read(&mp->m_peraglock); | |
2531 | return error; | |
2532 | } | |
2533 | ||
2534 | /* | |
2535 | * Free an extent. | |
2536 | * Just break up the extent address and hand off to xfs_free_ag_extent | |
2537 | * after fixing up the freelist. | |
2538 | */ | |
2539 | int /* error */ | |
2540 | xfs_free_extent( | |
2541 | xfs_trans_t *tp, /* transaction pointer */ | |
2542 | xfs_fsblock_t bno, /* starting block number of extent */ | |
2543 | xfs_extlen_t len) /* length of extent */ | |
2544 | { | |
0e1edbd9 | 2545 | xfs_alloc_arg_t args; |
1da177e4 LT |
2546 | int error; |
2547 | ||
2548 | ASSERT(len != 0); | |
0e1edbd9 | 2549 | memset(&args, 0, sizeof(xfs_alloc_arg_t)); |
1da177e4 LT |
2550 | args.tp = tp; |
2551 | args.mp = tp->t_mountp; | |
2552 | args.agno = XFS_FSB_TO_AGNO(args.mp, bno); | |
2553 | ASSERT(args.agno < args.mp->m_sb.sb_agcount); | |
2554 | args.agbno = XFS_FSB_TO_AGBNO(args.mp, bno); | |
1da177e4 LT |
2555 | down_read(&args.mp->m_peraglock); |
2556 | args.pag = &args.mp->m_perag[args.agno]; | |
d210a28c | 2557 | if ((error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING))) |
1da177e4 LT |
2558 | goto error0; |
2559 | #ifdef DEBUG | |
2560 | ASSERT(args.agbp != NULL); | |
0e1edbd9 NS |
2561 | ASSERT((args.agbno + len) <= |
2562 | be32_to_cpu(XFS_BUF_TO_AGF(args.agbp)->agf_length)); | |
1da177e4 | 2563 | #endif |
0e1edbd9 | 2564 | error = xfs_free_ag_extent(tp, args.agbp, args.agno, args.agbno, len, 0); |
1da177e4 LT |
2565 | error0: |
2566 | up_read(&args.mp->m_peraglock); | |
2567 | return error; | |
2568 | } | |
2569 | ||
2570 | ||
2571 | /* | |
2572 | * AG Busy list management | |
2573 | * The busy list contains block ranges that have been freed but whose | |
c41564b5 | 2574 | * transactions have not yet hit disk. If any block listed in a busy |
1da177e4 LT |
2575 | * list is reused, the transaction that freed it must be forced to disk |
2576 | * before continuing to use the block. | |
2577 | * | |
2578 | * xfs_alloc_mark_busy - add to the per-ag busy list | |
2579 | * xfs_alloc_clear_busy - remove an item from the per-ag busy list | |
2580 | */ | |
2581 | void | |
2582 | xfs_alloc_mark_busy(xfs_trans_t *tp, | |
2583 | xfs_agnumber_t agno, | |
2584 | xfs_agblock_t bno, | |
2585 | xfs_extlen_t len) | |
2586 | { | |
2587 | xfs_mount_t *mp; | |
2588 | xfs_perag_busy_t *bsy; | |
2589 | int n; | |
1da177e4 LT |
2590 | |
2591 | mp = tp->t_mountp; | |
64137e56 | 2592 | spin_lock(&mp->m_perag[agno].pagb_lock); |
1da177e4 LT |
2593 | |
2594 | /* search pagb_list for an open slot */ | |
2595 | for (bsy = mp->m_perag[agno].pagb_list, n = 0; | |
2596 | n < XFS_PAGB_NUM_SLOTS; | |
2597 | bsy++, n++) { | |
2598 | if (bsy->busy_tp == NULL) { | |
2599 | break; | |
2600 | } | |
2601 | } | |
2602 | ||
2603 | if (n < XFS_PAGB_NUM_SLOTS) { | |
2604 | bsy = &mp->m_perag[agno].pagb_list[n]; | |
2605 | mp->m_perag[agno].pagb_count++; | |
2606 | TRACE_BUSY("xfs_alloc_mark_busy", "got", agno, bno, len, n, tp); | |
2607 | bsy->busy_start = bno; | |
2608 | bsy->busy_length = len; | |
2609 | bsy->busy_tp = tp; | |
2610 | xfs_trans_add_busy(tp, agno, n); | |
2611 | } else { | |
2612 | TRACE_BUSY("xfs_alloc_mark_busy", "FULL", agno, bno, len, -1, tp); | |
2613 | /* | |
2614 | * The busy list is full! Since it is now not possible to | |
2615 | * track the free block, make this a synchronous transaction | |
2616 | * to insure that the block is not reused before this | |
2617 | * transaction commits. | |
2618 | */ | |
2619 | xfs_trans_set_sync(tp); | |
2620 | } | |
2621 | ||
64137e56 | 2622 | spin_unlock(&mp->m_perag[agno].pagb_lock); |
1da177e4 LT |
2623 | } |
2624 | ||
2625 | void | |
2626 | xfs_alloc_clear_busy(xfs_trans_t *tp, | |
2627 | xfs_agnumber_t agno, | |
2628 | int idx) | |
2629 | { | |
2630 | xfs_mount_t *mp; | |
2631 | xfs_perag_busy_t *list; | |
1da177e4 LT |
2632 | |
2633 | mp = tp->t_mountp; | |
2634 | ||
64137e56 | 2635 | spin_lock(&mp->m_perag[agno].pagb_lock); |
1da177e4 LT |
2636 | list = mp->m_perag[agno].pagb_list; |
2637 | ||
2638 | ASSERT(idx < XFS_PAGB_NUM_SLOTS); | |
2639 | if (list[idx].busy_tp == tp) { | |
2640 | TRACE_UNBUSY("xfs_alloc_clear_busy", "found", agno, idx, tp); | |
2641 | list[idx].busy_tp = NULL; | |
2642 | mp->m_perag[agno].pagb_count--; | |
2643 | } else { | |
2644 | TRACE_UNBUSY("xfs_alloc_clear_busy", "missing", agno, idx, tp); | |
2645 | } | |
2646 | ||
64137e56 | 2647 | spin_unlock(&mp->m_perag[agno].pagb_lock); |
1da177e4 LT |
2648 | } |
2649 | ||
2650 | ||
2651 | /* | |
f4586e40 DC |
2652 | * If we find the extent in the busy list, force the log out to get the |
2653 | * extent out of the busy list so the caller can use it straight away. | |
1da177e4 | 2654 | */ |
f4586e40 | 2655 | STATIC void |
1da177e4 LT |
2656 | xfs_alloc_search_busy(xfs_trans_t *tp, |
2657 | xfs_agnumber_t agno, | |
2658 | xfs_agblock_t bno, | |
2659 | xfs_extlen_t len) | |
2660 | { | |
2661 | xfs_mount_t *mp; | |
2662 | xfs_perag_busy_t *bsy; | |
1da177e4 LT |
2663 | xfs_agblock_t uend, bend; |
2664 | xfs_lsn_t lsn; | |
2665 | int cnt; | |
1da177e4 LT |
2666 | |
2667 | mp = tp->t_mountp; | |
2668 | ||
64137e56 | 2669 | spin_lock(&mp->m_perag[agno].pagb_lock); |
1da177e4 LT |
2670 | cnt = mp->m_perag[agno].pagb_count; |
2671 | ||
2672 | uend = bno + len - 1; | |
2673 | ||
2674 | /* search pagb_list for this slot, skipping open slots */ | |
f4586e40 | 2675 | for (bsy = mp->m_perag[agno].pagb_list; cnt; bsy++) { |
1da177e4 LT |
2676 | |
2677 | /* | |
2678 | * (start1,length1) within (start2, length2) | |
2679 | */ | |
2680 | if (bsy->busy_tp != NULL) { | |
2681 | bend = bsy->busy_start + bsy->busy_length - 1; | |
f4586e40 | 2682 | if ((bno > bend) || (uend < bsy->busy_start)) { |
1da177e4 LT |
2683 | cnt--; |
2684 | } else { | |
2685 | TRACE_BUSYSEARCH("xfs_alloc_search_busy", | |
f4586e40 | 2686 | "found1", agno, bno, len, tp); |
1da177e4 LT |
2687 | break; |
2688 | } | |
2689 | } | |
2690 | } | |
2691 | ||
2692 | /* | |
2693 | * If a block was found, force the log through the LSN of the | |
2694 | * transaction that freed the block | |
2695 | */ | |
2696 | if (cnt) { | |
f4586e40 | 2697 | TRACE_BUSYSEARCH("xfs_alloc_search_busy", "found", agno, bno, len, tp); |
1da177e4 | 2698 | lsn = bsy->busy_tp->t_commit_lsn; |
64137e56 | 2699 | spin_unlock(&mp->m_perag[agno].pagb_lock); |
1da177e4 LT |
2700 | xfs_log_force(mp, lsn, XFS_LOG_FORCE|XFS_LOG_SYNC); |
2701 | } else { | |
f4586e40 | 2702 | TRACE_BUSYSEARCH("xfs_alloc_search_busy", "not-found", agno, bno, len, tp); |
64137e56 | 2703 | spin_unlock(&mp->m_perag[agno].pagb_lock); |
1da177e4 | 2704 | } |
1da177e4 | 2705 | } |