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 | } | |
7cc95a82 | 383 | |
1da177e4 | 384 | #ifdef DEBUG |
7cc95a82 CH |
385 | if (bno_cur->bc_nlevels == 1 && cnt_cur->bc_nlevels == 1) { |
386 | struct xfs_btree_block *bnoblock; | |
387 | struct xfs_btree_block *cntblock; | |
388 | ||
389 | bnoblock = XFS_BUF_TO_BLOCK(bno_cur->bc_bufs[0]); | |
390 | cntblock = XFS_BUF_TO_BLOCK(cnt_cur->bc_bufs[0]); | |
1da177e4 | 391 | |
7cc95a82 CH |
392 | XFS_WANT_CORRUPTED_RETURN( |
393 | bnoblock->bb_numrecs == cntblock->bb_numrecs); | |
1da177e4 LT |
394 | } |
395 | #endif | |
7cc95a82 | 396 | |
1da177e4 LT |
397 | /* |
398 | * Deal with all four cases: the allocated record is contained | |
399 | * within the freespace record, so we can have new freespace | |
400 | * at either (or both) end, or no freespace remaining. | |
401 | */ | |
402 | if (rbno == fbno && rlen == flen) | |
403 | nfbno1 = nfbno2 = NULLAGBLOCK; | |
404 | else if (rbno == fbno) { | |
405 | nfbno1 = rbno + rlen; | |
406 | nflen1 = flen - rlen; | |
407 | nfbno2 = NULLAGBLOCK; | |
408 | } else if (rbno + rlen == fbno + flen) { | |
409 | nfbno1 = fbno; | |
410 | nflen1 = flen - rlen; | |
411 | nfbno2 = NULLAGBLOCK; | |
412 | } else { | |
413 | nfbno1 = fbno; | |
414 | nflen1 = rbno - fbno; | |
415 | nfbno2 = rbno + rlen; | |
416 | nflen2 = (fbno + flen) - nfbno2; | |
417 | } | |
418 | /* | |
419 | * Delete the entry from the by-size btree. | |
420 | */ | |
91cca5df | 421 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 LT |
422 | return error; |
423 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
424 | /* | |
425 | * Add new by-size btree entry(s). | |
426 | */ | |
427 | if (nfbno1 != NULLAGBLOCK) { | |
428 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno1, nflen1, &i))) | |
429 | return error; | |
430 | XFS_WANT_CORRUPTED_RETURN(i == 0); | |
4b22a571 | 431 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
1da177e4 LT |
432 | return error; |
433 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
434 | } | |
435 | if (nfbno2 != NULLAGBLOCK) { | |
436 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno2, nflen2, &i))) | |
437 | return error; | |
438 | XFS_WANT_CORRUPTED_RETURN(i == 0); | |
4b22a571 | 439 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
1da177e4 LT |
440 | return error; |
441 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
442 | } | |
443 | /* | |
444 | * Fix up the by-block btree entry(s). | |
445 | */ | |
446 | if (nfbno1 == NULLAGBLOCK) { | |
447 | /* | |
448 | * No remaining freespace, just delete the by-block tree entry. | |
449 | */ | |
91cca5df | 450 | if ((error = xfs_btree_delete(bno_cur, &i))) |
1da177e4 LT |
451 | return error; |
452 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
453 | } else { | |
454 | /* | |
455 | * Update the by-block entry to start later|be shorter. | |
456 | */ | |
457 | if ((error = xfs_alloc_update(bno_cur, nfbno1, nflen1))) | |
458 | return error; | |
459 | } | |
460 | if (nfbno2 != NULLAGBLOCK) { | |
461 | /* | |
462 | * 2 resulting free entries, need to add one. | |
463 | */ | |
464 | if ((error = xfs_alloc_lookup_eq(bno_cur, nfbno2, nflen2, &i))) | |
465 | return error; | |
466 | XFS_WANT_CORRUPTED_RETURN(i == 0); | |
4b22a571 | 467 | if ((error = xfs_btree_insert(bno_cur, &i))) |
1da177e4 LT |
468 | return error; |
469 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
470 | } | |
471 | return 0; | |
472 | } | |
473 | ||
474 | /* | |
475 | * Read in the allocation group free block array. | |
476 | */ | |
477 | STATIC int /* error */ | |
478 | xfs_alloc_read_agfl( | |
479 | xfs_mount_t *mp, /* mount point structure */ | |
480 | xfs_trans_t *tp, /* transaction pointer */ | |
481 | xfs_agnumber_t agno, /* allocation group number */ | |
482 | xfs_buf_t **bpp) /* buffer for the ag free block array */ | |
483 | { | |
484 | xfs_buf_t *bp; /* return value */ | |
485 | int error; | |
486 | ||
487 | ASSERT(agno != NULLAGNUMBER); | |
488 | error = xfs_trans_read_buf( | |
489 | mp, tp, mp->m_ddev_targp, | |
490 | XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)), | |
491 | XFS_FSS_TO_BB(mp, 1), 0, &bp); | |
492 | if (error) | |
493 | return error; | |
494 | ASSERT(bp); | |
495 | ASSERT(!XFS_BUF_GETERROR(bp)); | |
496 | XFS_BUF_SET_VTYPE_REF(bp, B_FS_AGFL, XFS_AGFL_REF); | |
497 | *bpp = bp; | |
498 | return 0; | |
499 | } | |
500 | ||
501 | #if defined(XFS_ALLOC_TRACE) | |
502 | /* | |
503 | * Add an allocation trace entry for an alloc call. | |
504 | */ | |
505 | STATIC void | |
506 | xfs_alloc_trace_alloc( | |
3a59c94c | 507 | const char *name, /* function tag string */ |
1da177e4 LT |
508 | char *str, /* additional string */ |
509 | xfs_alloc_arg_t *args, /* allocation argument structure */ | |
510 | int line) /* source line number */ | |
511 | { | |
512 | ktrace_enter(xfs_alloc_trace_buf, | |
513 | (void *)(__psint_t)(XFS_ALLOC_KTRACE_ALLOC | (line << 16)), | |
514 | (void *)name, | |
515 | (void *)str, | |
516 | (void *)args->mp, | |
517 | (void *)(__psunsigned_t)args->agno, | |
518 | (void *)(__psunsigned_t)args->agbno, | |
519 | (void *)(__psunsigned_t)args->minlen, | |
520 | (void *)(__psunsigned_t)args->maxlen, | |
521 | (void *)(__psunsigned_t)args->mod, | |
522 | (void *)(__psunsigned_t)args->prod, | |
523 | (void *)(__psunsigned_t)args->minleft, | |
524 | (void *)(__psunsigned_t)args->total, | |
525 | (void *)(__psunsigned_t)args->alignment, | |
526 | (void *)(__psunsigned_t)args->len, | |
527 | (void *)((((__psint_t)args->type) << 16) | | |
528 | (__psint_t)args->otype), | |
529 | (void *)(__psint_t)((args->wasdel << 3) | | |
530 | (args->wasfromfl << 2) | | |
531 | (args->isfl << 1) | | |
532 | (args->userdata << 0))); | |
533 | } | |
534 | ||
535 | /* | |
536 | * Add an allocation trace entry for a free call. | |
537 | */ | |
538 | STATIC void | |
539 | xfs_alloc_trace_free( | |
3a59c94c | 540 | const char *name, /* function tag string */ |
1da177e4 LT |
541 | char *str, /* additional string */ |
542 | xfs_mount_t *mp, /* file system mount point */ | |
543 | xfs_agnumber_t agno, /* allocation group number */ | |
544 | xfs_agblock_t agbno, /* a.g. relative block number */ | |
545 | xfs_extlen_t len, /* length of extent */ | |
546 | int isfl, /* set if is freelist allocation/free */ | |
547 | int line) /* source line number */ | |
548 | { | |
549 | ktrace_enter(xfs_alloc_trace_buf, | |
550 | (void *)(__psint_t)(XFS_ALLOC_KTRACE_FREE | (line << 16)), | |
551 | (void *)name, | |
552 | (void *)str, | |
553 | (void *)mp, | |
554 | (void *)(__psunsigned_t)agno, | |
555 | (void *)(__psunsigned_t)agbno, | |
556 | (void *)(__psunsigned_t)len, | |
557 | (void *)(__psint_t)isfl, | |
558 | NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL); | |
559 | } | |
560 | ||
561 | /* | |
562 | * Add an allocation trace entry for modifying an agf. | |
563 | */ | |
564 | STATIC void | |
565 | xfs_alloc_trace_modagf( | |
3a59c94c | 566 | const char *name, /* function tag string */ |
1da177e4 LT |
567 | char *str, /* additional string */ |
568 | xfs_mount_t *mp, /* file system mount point */ | |
569 | xfs_agf_t *agf, /* new agf value */ | |
570 | int flags, /* logging flags for agf */ | |
571 | int line) /* source line number */ | |
572 | { | |
573 | ktrace_enter(xfs_alloc_trace_buf, | |
574 | (void *)(__psint_t)(XFS_ALLOC_KTRACE_MODAGF | (line << 16)), | |
575 | (void *)name, | |
576 | (void *)str, | |
577 | (void *)mp, | |
578 | (void *)(__psint_t)flags, | |
16259e7d CH |
579 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_seqno), |
580 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_length), | |
581 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_roots[XFS_BTNUM_BNO]), | |
582 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_roots[XFS_BTNUM_CNT]), | |
583 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]), | |
584 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]), | |
585 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_flfirst), | |
586 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_fllast), | |
587 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_flcount), | |
588 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_freeblks), | |
589 | (void *)(__psunsigned_t)be32_to_cpu(agf->agf_longest)); | |
1da177e4 LT |
590 | } |
591 | ||
592 | STATIC void | |
593 | xfs_alloc_trace_busy( | |
3a59c94c | 594 | const char *name, /* function tag string */ |
1da177e4 | 595 | char *str, /* additional string */ |
c41564b5 | 596 | xfs_mount_t *mp, /* file system mount point */ |
1da177e4 LT |
597 | xfs_agnumber_t agno, /* allocation group number */ |
598 | xfs_agblock_t agbno, /* a.g. relative block number */ | |
599 | xfs_extlen_t len, /* length of extent */ | |
600 | int slot, /* perag Busy slot */ | |
601 | xfs_trans_t *tp, | |
602 | int trtype, /* type: add, delete, search */ | |
603 | int line) /* source line number */ | |
604 | { | |
605 | ktrace_enter(xfs_alloc_trace_buf, | |
606 | (void *)(__psint_t)(trtype | (line << 16)), | |
607 | (void *)name, | |
608 | (void *)str, | |
609 | (void *)mp, | |
610 | (void *)(__psunsigned_t)agno, | |
611 | (void *)(__psunsigned_t)agbno, | |
612 | (void *)(__psunsigned_t)len, | |
613 | (void *)(__psint_t)slot, | |
614 | (void *)tp, | |
615 | NULL, NULL, NULL, NULL, NULL, NULL, NULL); | |
616 | } | |
617 | #endif /* XFS_ALLOC_TRACE */ | |
618 | ||
619 | /* | |
620 | * Allocation group level functions. | |
621 | */ | |
622 | ||
623 | /* | |
624 | * Allocate a variable extent in the allocation group agno. | |
625 | * Type and bno are used to determine where in the allocation group the | |
626 | * extent will start. | |
627 | * Extent's length (returned in *len) will be between minlen and maxlen, | |
628 | * and of the form k * prod + mod unless there's nothing that large. | |
629 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
630 | */ | |
631 | STATIC int /* error */ | |
632 | xfs_alloc_ag_vextent( | |
633 | xfs_alloc_arg_t *args) /* argument structure for allocation */ | |
634 | { | |
635 | int error=0; | |
1da177e4 LT |
636 | |
637 | ASSERT(args->minlen > 0); | |
638 | ASSERT(args->maxlen > 0); | |
639 | ASSERT(args->minlen <= args->maxlen); | |
640 | ASSERT(args->mod < args->prod); | |
641 | ASSERT(args->alignment > 0); | |
642 | /* | |
643 | * Branch to correct routine based on the type. | |
644 | */ | |
645 | args->wasfromfl = 0; | |
646 | switch (args->type) { | |
647 | case XFS_ALLOCTYPE_THIS_AG: | |
648 | error = xfs_alloc_ag_vextent_size(args); | |
649 | break; | |
650 | case XFS_ALLOCTYPE_NEAR_BNO: | |
651 | error = xfs_alloc_ag_vextent_near(args); | |
652 | break; | |
653 | case XFS_ALLOCTYPE_THIS_BNO: | |
654 | error = xfs_alloc_ag_vextent_exact(args); | |
655 | break; | |
656 | default: | |
657 | ASSERT(0); | |
658 | /* NOTREACHED */ | |
659 | } | |
660 | if (error) | |
661 | return error; | |
662 | /* | |
663 | * If the allocation worked, need to change the agf structure | |
664 | * (and log it), and the superblock. | |
665 | */ | |
666 | if (args->agbno != NULLAGBLOCK) { | |
667 | xfs_agf_t *agf; /* allocation group freelist header */ | |
668 | #ifdef XFS_ALLOC_TRACE | |
669 | xfs_mount_t *mp = args->mp; | |
670 | #endif | |
671 | long slen = (long)args->len; | |
672 | ||
673 | ASSERT(args->len >= args->minlen && args->len <= args->maxlen); | |
674 | ASSERT(!(args->wasfromfl) || !args->isfl); | |
675 | ASSERT(args->agbno % args->alignment == 0); | |
676 | if (!(args->wasfromfl)) { | |
677 | ||
678 | agf = XFS_BUF_TO_AGF(args->agbp); | |
413d57c9 | 679 | be32_add_cpu(&agf->agf_freeblks, -(args->len)); |
1da177e4 LT |
680 | xfs_trans_agblocks_delta(args->tp, |
681 | -((long)(args->len))); | |
682 | args->pag->pagf_freeblks -= args->len; | |
16259e7d CH |
683 | ASSERT(be32_to_cpu(agf->agf_freeblks) <= |
684 | be32_to_cpu(agf->agf_length)); | |
1da177e4 LT |
685 | TRACE_MODAGF(NULL, agf, XFS_AGF_FREEBLKS); |
686 | xfs_alloc_log_agf(args->tp, args->agbp, | |
687 | XFS_AGF_FREEBLKS); | |
688 | /* search the busylist for these blocks */ | |
689 | xfs_alloc_search_busy(args->tp, args->agno, | |
690 | args->agbno, args->len); | |
691 | } | |
692 | if (!args->isfl) | |
693 | xfs_trans_mod_sb(args->tp, | |
694 | args->wasdel ? XFS_TRANS_SB_RES_FDBLOCKS : | |
695 | XFS_TRANS_SB_FDBLOCKS, -slen); | |
696 | XFS_STATS_INC(xs_allocx); | |
697 | XFS_STATS_ADD(xs_allocb, args->len); | |
698 | } | |
699 | return 0; | |
700 | } | |
701 | ||
702 | /* | |
703 | * Allocate a variable extent at exactly agno/bno. | |
704 | * Extent's length (returned in *len) will be between minlen and maxlen, | |
705 | * and of the form k * prod + mod unless there's nothing that large. | |
706 | * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it. | |
707 | */ | |
708 | STATIC int /* error */ | |
709 | xfs_alloc_ag_vextent_exact( | |
710 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
711 | { | |
712 | xfs_btree_cur_t *bno_cur;/* by block-number btree cursor */ | |
713 | xfs_btree_cur_t *cnt_cur;/* by count btree cursor */ | |
714 | xfs_agblock_t end; /* end of allocated extent */ | |
715 | int error; | |
716 | xfs_agblock_t fbno; /* start block of found extent */ | |
717 | xfs_agblock_t fend; /* end block of found extent */ | |
718 | xfs_extlen_t flen; /* length of found extent */ | |
1da177e4 LT |
719 | int i; /* success/failure of operation */ |
720 | xfs_agblock_t maxend; /* end of maximal extent */ | |
721 | xfs_agblock_t minend; /* end of minimal extent */ | |
722 | xfs_extlen_t rlen; /* length of returned extent */ | |
723 | ||
724 | ASSERT(args->alignment == 1); | |
725 | /* | |
726 | * Allocate/initialize a cursor for the by-number freespace btree. | |
727 | */ | |
561f7d17 CH |
728 | bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
729 | args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
730 | /* |
731 | * Lookup bno and minlen in the btree (minlen is irrelevant, really). | |
732 | * Look for the closest free block <= bno, it must contain bno | |
733 | * if any free block does. | |
734 | */ | |
735 | if ((error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i))) | |
736 | goto error0; | |
737 | if (!i) { | |
738 | /* | |
739 | * Didn't find it, return null. | |
740 | */ | |
741 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
742 | args->agbno = NULLAGBLOCK; | |
743 | return 0; | |
744 | } | |
745 | /* | |
746 | * Grab the freespace record. | |
747 | */ | |
748 | if ((error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i))) | |
749 | goto error0; | |
750 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
751 | ASSERT(fbno <= args->agbno); | |
752 | minend = args->agbno + args->minlen; | |
753 | maxend = args->agbno + args->maxlen; | |
754 | fend = fbno + flen; | |
755 | /* | |
756 | * Give up if the freespace isn't long enough for the minimum request. | |
757 | */ | |
758 | if (fend < minend) { | |
759 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
760 | args->agbno = NULLAGBLOCK; | |
761 | return 0; | |
762 | } | |
763 | /* | |
764 | * End of extent will be smaller of the freespace end and the | |
765 | * maximal requested end. | |
766 | */ | |
767 | end = XFS_AGBLOCK_MIN(fend, maxend); | |
768 | /* | |
769 | * Fix the length according to mod and prod if given. | |
770 | */ | |
771 | args->len = end - args->agbno; | |
772 | xfs_alloc_fix_len(args); | |
773 | if (!xfs_alloc_fix_minleft(args)) { | |
774 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
775 | return 0; | |
776 | } | |
777 | rlen = args->len; | |
778 | ASSERT(args->agbno + rlen <= fend); | |
779 | end = args->agbno + rlen; | |
780 | /* | |
781 | * We are allocating agbno for rlen [agbno .. end] | |
782 | * Allocate/initialize a cursor for the by-size btree. | |
783 | */ | |
561f7d17 CH |
784 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
785 | args->agno, XFS_BTNUM_CNT); | |
1da177e4 | 786 | ASSERT(args->agbno + args->len <= |
16259e7d | 787 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
1da177e4 LT |
788 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, |
789 | args->agbno, args->len, XFSA_FIXUP_BNO_OK))) { | |
790 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
791 | goto error0; | |
792 | } | |
793 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
794 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
795 | TRACE_ALLOC("normal", args); | |
796 | args->wasfromfl = 0; | |
797 | return 0; | |
798 | ||
799 | error0: | |
800 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
801 | TRACE_ALLOC("error", args); | |
802 | return error; | |
803 | } | |
804 | ||
805 | /* | |
806 | * Allocate a variable extent near bno in the allocation group agno. | |
807 | * Extent's length (returned in len) will be between minlen and maxlen, | |
808 | * and of the form k * prod + mod unless there's nothing that large. | |
809 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
810 | */ | |
811 | STATIC int /* error */ | |
812 | xfs_alloc_ag_vextent_near( | |
813 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
814 | { | |
815 | xfs_btree_cur_t *bno_cur_gt; /* cursor for bno btree, right side */ | |
816 | xfs_btree_cur_t *bno_cur_lt; /* cursor for bno btree, left side */ | |
817 | xfs_btree_cur_t *cnt_cur; /* cursor for count btree */ | |
1da177e4 LT |
818 | xfs_agblock_t gtbno; /* start bno of right side entry */ |
819 | xfs_agblock_t gtbnoa; /* aligned ... */ | |
820 | xfs_extlen_t gtdiff; /* difference to right side entry */ | |
821 | xfs_extlen_t gtlen; /* length of right side entry */ | |
822 | xfs_extlen_t gtlena; /* aligned ... */ | |
823 | xfs_agblock_t gtnew; /* useful start bno of right side */ | |
824 | int error; /* error code */ | |
825 | int i; /* result code, temporary */ | |
826 | int j; /* result code, temporary */ | |
827 | xfs_agblock_t ltbno; /* start bno of left side entry */ | |
828 | xfs_agblock_t ltbnoa; /* aligned ... */ | |
829 | xfs_extlen_t ltdiff; /* difference to left side entry */ | |
830 | /*REFERENCED*/ | |
831 | xfs_agblock_t ltend; /* end bno of left side entry */ | |
832 | xfs_extlen_t ltlen; /* length of left side entry */ | |
833 | xfs_extlen_t ltlena; /* aligned ... */ | |
834 | xfs_agblock_t ltnew; /* useful start bno of left side */ | |
835 | xfs_extlen_t rlen; /* length of returned extent */ | |
836 | #if defined(DEBUG) && defined(__KERNEL__) | |
837 | /* | |
838 | * Randomly don't execute the first algorithm. | |
839 | */ | |
840 | int dofirst; /* set to do first algorithm */ | |
841 | ||
e7a23a9b | 842 | dofirst = random32() & 1; |
1da177e4 LT |
843 | #endif |
844 | /* | |
845 | * Get a cursor for the by-size btree. | |
846 | */ | |
561f7d17 CH |
847 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
848 | args->agno, XFS_BTNUM_CNT); | |
1da177e4 LT |
849 | ltlen = 0; |
850 | bno_cur_lt = bno_cur_gt = NULL; | |
851 | /* | |
852 | * See if there are any free extents as big as maxlen. | |
853 | */ | |
854 | if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, args->maxlen, &i))) | |
855 | goto error0; | |
856 | /* | |
857 | * If none, then pick up the last entry in the tree unless the | |
858 | * tree is empty. | |
859 | */ | |
860 | if (!i) { | |
861 | if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, <bno, | |
862 | <len, &i))) | |
863 | goto error0; | |
864 | if (i == 0 || ltlen == 0) { | |
865 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
866 | return 0; | |
867 | } | |
868 | ASSERT(i == 1); | |
869 | } | |
870 | args->wasfromfl = 0; | |
871 | /* | |
872 | * First algorithm. | |
873 | * If the requested extent is large wrt the freespaces available | |
874 | * in this a.g., then the cursor will be pointing to a btree entry | |
875 | * near the right edge of the tree. If it's in the last btree leaf | |
876 | * block, then we just examine all the entries in that block | |
877 | * that are big enough, and pick the best one. | |
878 | * This is written as a while loop so we can break out of it, | |
879 | * but we never loop back to the top. | |
880 | */ | |
881 | while (xfs_btree_islastblock(cnt_cur, 0)) { | |
882 | xfs_extlen_t bdiff; | |
883 | int besti=0; | |
884 | xfs_extlen_t blen=0; | |
885 | xfs_agblock_t bnew=0; | |
886 | ||
887 | #if defined(DEBUG) && defined(__KERNEL__) | |
888 | if (!dofirst) | |
889 | break; | |
890 | #endif | |
891 | /* | |
892 | * Start from the entry that lookup found, sequence through | |
893 | * all larger free blocks. If we're actually pointing at a | |
894 | * record smaller than maxlen, go to the start of this block, | |
895 | * and skip all those smaller than minlen. | |
896 | */ | |
897 | if (ltlen || args->alignment > 1) { | |
898 | cnt_cur->bc_ptrs[0] = 1; | |
899 | do { | |
900 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, | |
901 | <len, &i))) | |
902 | goto error0; | |
903 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
904 | if (ltlen >= args->minlen) | |
905 | break; | |
637aa50f | 906 | if ((error = xfs_btree_increment(cnt_cur, 0, &i))) |
1da177e4 LT |
907 | goto error0; |
908 | } while (i); | |
909 | ASSERT(ltlen >= args->minlen); | |
910 | if (!i) | |
911 | break; | |
912 | } | |
913 | i = cnt_cur->bc_ptrs[0]; | |
914 | for (j = 1, blen = 0, bdiff = 0; | |
915 | !error && j && (blen < args->maxlen || bdiff > 0); | |
637aa50f | 916 | error = xfs_btree_increment(cnt_cur, 0, &j)) { |
1da177e4 LT |
917 | /* |
918 | * For each entry, decide if it's better than | |
919 | * the previous best entry. | |
920 | */ | |
921 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i))) | |
922 | goto error0; | |
923 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
12375c82 DC |
924 | xfs_alloc_compute_aligned(ltbno, ltlen, args->alignment, |
925 | args->minlen, <bnoa, <lena); | |
e6430037 | 926 | if (ltlena < args->minlen) |
1da177e4 LT |
927 | continue; |
928 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); | |
929 | xfs_alloc_fix_len(args); | |
930 | ASSERT(args->len >= args->minlen); | |
931 | if (args->len < blen) | |
932 | continue; | |
933 | ltdiff = xfs_alloc_compute_diff(args->agbno, args->len, | |
934 | args->alignment, ltbno, ltlen, <new); | |
935 | if (ltnew != NULLAGBLOCK && | |
936 | (args->len > blen || ltdiff < bdiff)) { | |
937 | bdiff = ltdiff; | |
938 | bnew = ltnew; | |
939 | blen = args->len; | |
940 | besti = cnt_cur->bc_ptrs[0]; | |
941 | } | |
942 | } | |
943 | /* | |
944 | * It didn't work. We COULD be in a case where | |
945 | * there's a good record somewhere, so try again. | |
946 | */ | |
947 | if (blen == 0) | |
948 | break; | |
949 | /* | |
950 | * Point at the best entry, and retrieve it again. | |
951 | */ | |
952 | cnt_cur->bc_ptrs[0] = besti; | |
953 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i))) | |
954 | goto error0; | |
955 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
956 | ltend = ltbno + ltlen; | |
16259e7d | 957 | ASSERT(ltend <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
1da177e4 LT |
958 | args->len = blen; |
959 | if (!xfs_alloc_fix_minleft(args)) { | |
960 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
961 | TRACE_ALLOC("nominleft", args); | |
962 | return 0; | |
963 | } | |
964 | blen = args->len; | |
965 | /* | |
966 | * We are allocating starting at bnew for blen blocks. | |
967 | */ | |
968 | args->agbno = bnew; | |
969 | ASSERT(bnew >= ltbno); | |
970 | ASSERT(bnew + blen <= ltend); | |
971 | /* | |
972 | * Set up a cursor for the by-bno tree. | |
973 | */ | |
561f7d17 CH |
974 | bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, |
975 | args->agbp, args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
976 | /* |
977 | * Fix up the btree entries. | |
978 | */ | |
979 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, | |
980 | ltlen, bnew, blen, XFSA_FIXUP_CNT_OK))) | |
981 | goto error0; | |
982 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
983 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); | |
984 | TRACE_ALLOC("first", args); | |
985 | return 0; | |
986 | } | |
987 | /* | |
988 | * Second algorithm. | |
989 | * Search in the by-bno tree to the left and to the right | |
990 | * simultaneously, until in each case we find a space big enough, | |
991 | * or run into the edge of the tree. When we run into the edge, | |
992 | * we deallocate that cursor. | |
993 | * If both searches succeed, we compare the two spaces and pick | |
994 | * the better one. | |
995 | * With alignment, it's possible for both to fail; the upper | |
996 | * level algorithm that picks allocation groups for allocations | |
997 | * is not supposed to do this. | |
998 | */ | |
999 | /* | |
1000 | * Allocate and initialize the cursor for the leftward search. | |
1001 | */ | |
561f7d17 CH |
1002 | bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1003 | args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
1004 | /* |
1005 | * Lookup <= bno to find the leftward search's starting point. | |
1006 | */ | |
1007 | if ((error = xfs_alloc_lookup_le(bno_cur_lt, args->agbno, args->maxlen, &i))) | |
1008 | goto error0; | |
1009 | if (!i) { | |
1010 | /* | |
1011 | * Didn't find anything; use this cursor for the rightward | |
1012 | * search. | |
1013 | */ | |
1014 | bno_cur_gt = bno_cur_lt; | |
1015 | bno_cur_lt = NULL; | |
1016 | } | |
1017 | /* | |
1018 | * Found something. Duplicate the cursor for the rightward search. | |
1019 | */ | |
1020 | else if ((error = xfs_btree_dup_cursor(bno_cur_lt, &bno_cur_gt))) | |
1021 | goto error0; | |
1022 | /* | |
1023 | * Increment the cursor, so we will point at the entry just right | |
1024 | * of the leftward entry if any, or to the leftmost entry. | |
1025 | */ | |
637aa50f | 1026 | if ((error = xfs_btree_increment(bno_cur_gt, 0, &i))) |
1da177e4 LT |
1027 | goto error0; |
1028 | if (!i) { | |
1029 | /* | |
1030 | * It failed, there are no rightward entries. | |
1031 | */ | |
1032 | xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_NOERROR); | |
1033 | bno_cur_gt = NULL; | |
1034 | } | |
1035 | /* | |
1036 | * Loop going left with the leftward cursor, right with the | |
1037 | * rightward cursor, until either both directions give up or | |
1038 | * we find an entry at least as big as minlen. | |
1039 | */ | |
1040 | do { | |
1041 | if (bno_cur_lt) { | |
1042 | if ((error = xfs_alloc_get_rec(bno_cur_lt, <bno, <len, &i))) | |
1043 | goto error0; | |
1044 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
12375c82 DC |
1045 | xfs_alloc_compute_aligned(ltbno, ltlen, args->alignment, |
1046 | args->minlen, <bnoa, <lena); | |
1047 | if (ltlena >= args->minlen) | |
1da177e4 | 1048 | break; |
8df4da4a | 1049 | if ((error = xfs_btree_decrement(bno_cur_lt, 0, &i))) |
1da177e4 LT |
1050 | goto error0; |
1051 | if (!i) { | |
1052 | xfs_btree_del_cursor(bno_cur_lt, | |
1053 | XFS_BTREE_NOERROR); | |
1054 | bno_cur_lt = NULL; | |
1055 | } | |
1056 | } | |
1057 | if (bno_cur_gt) { | |
1058 | if ((error = xfs_alloc_get_rec(bno_cur_gt, >bno, >len, &i))) | |
1059 | goto error0; | |
1060 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
12375c82 DC |
1061 | xfs_alloc_compute_aligned(gtbno, gtlen, args->alignment, |
1062 | args->minlen, >bnoa, >lena); | |
1063 | if (gtlena >= args->minlen) | |
1da177e4 | 1064 | break; |
637aa50f | 1065 | if ((error = xfs_btree_increment(bno_cur_gt, 0, &i))) |
1da177e4 LT |
1066 | goto error0; |
1067 | if (!i) { | |
1068 | xfs_btree_del_cursor(bno_cur_gt, | |
1069 | XFS_BTREE_NOERROR); | |
1070 | bno_cur_gt = NULL; | |
1071 | } | |
1072 | } | |
1073 | } while (bno_cur_lt || bno_cur_gt); | |
1074 | /* | |
1075 | * Got both cursors still active, need to find better entry. | |
1076 | */ | |
1077 | if (bno_cur_lt && bno_cur_gt) { | |
1078 | /* | |
1079 | * Left side is long enough, look for a right side entry. | |
1080 | */ | |
1081 | if (ltlena >= args->minlen) { | |
1082 | /* | |
1083 | * Fix up the length. | |
1084 | */ | |
1085 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); | |
1086 | xfs_alloc_fix_len(args); | |
1087 | rlen = args->len; | |
1088 | ltdiff = xfs_alloc_compute_diff(args->agbno, rlen, | |
1089 | args->alignment, ltbno, ltlen, <new); | |
1090 | /* | |
1091 | * Not perfect. | |
1092 | */ | |
1093 | if (ltdiff) { | |
1094 | /* | |
1095 | * Look until we find a better one, run out of | |
1096 | * space, or run off the end. | |
1097 | */ | |
1098 | while (bno_cur_lt && bno_cur_gt) { | |
1099 | if ((error = xfs_alloc_get_rec( | |
1100 | bno_cur_gt, >bno, | |
1101 | >len, &i))) | |
1102 | goto error0; | |
1103 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1104 | xfs_alloc_compute_aligned(gtbno, gtlen, | |
1105 | args->alignment, args->minlen, | |
1106 | >bnoa, >lena); | |
1107 | /* | |
1108 | * The left one is clearly better. | |
1109 | */ | |
1110 | if (gtbnoa >= args->agbno + ltdiff) { | |
1111 | xfs_btree_del_cursor( | |
1112 | bno_cur_gt, | |
1113 | XFS_BTREE_NOERROR); | |
1114 | bno_cur_gt = NULL; | |
1115 | break; | |
1116 | } | |
1117 | /* | |
1118 | * If we reach a big enough entry, | |
1119 | * compare the two and pick the best. | |
1120 | */ | |
1121 | if (gtlena >= args->minlen) { | |
1122 | args->len = | |
1123 | XFS_EXTLEN_MIN(gtlena, | |
1124 | args->maxlen); | |
1125 | xfs_alloc_fix_len(args); | |
1126 | rlen = args->len; | |
1127 | gtdiff = xfs_alloc_compute_diff( | |
1128 | args->agbno, rlen, | |
1129 | args->alignment, | |
1130 | gtbno, gtlen, >new); | |
1131 | /* | |
1132 | * Right side is better. | |
1133 | */ | |
1134 | if (gtdiff < ltdiff) { | |
1135 | xfs_btree_del_cursor( | |
1136 | bno_cur_lt, | |
1137 | XFS_BTREE_NOERROR); | |
1138 | bno_cur_lt = NULL; | |
1139 | } | |
1140 | /* | |
1141 | * Left side is better. | |
1142 | */ | |
1143 | else { | |
1144 | xfs_btree_del_cursor( | |
1145 | bno_cur_gt, | |
1146 | XFS_BTREE_NOERROR); | |
1147 | bno_cur_gt = NULL; | |
1148 | } | |
1149 | break; | |
1150 | } | |
1151 | /* | |
1152 | * Fell off the right end. | |
1153 | */ | |
637aa50f | 1154 | if ((error = xfs_btree_increment( |
1da177e4 LT |
1155 | bno_cur_gt, 0, &i))) |
1156 | goto error0; | |
1157 | if (!i) { | |
1158 | xfs_btree_del_cursor( | |
1159 | bno_cur_gt, | |
1160 | XFS_BTREE_NOERROR); | |
1161 | bno_cur_gt = NULL; | |
1162 | break; | |
1163 | } | |
1164 | } | |
1165 | } | |
1166 | /* | |
1167 | * The left side is perfect, trash the right side. | |
1168 | */ | |
1169 | else { | |
1170 | xfs_btree_del_cursor(bno_cur_gt, | |
1171 | XFS_BTREE_NOERROR); | |
1172 | bno_cur_gt = NULL; | |
1173 | } | |
1174 | } | |
1175 | /* | |
1176 | * It's the right side that was found first, look left. | |
1177 | */ | |
1178 | else { | |
1179 | /* | |
1180 | * Fix up the length. | |
1181 | */ | |
1182 | args->len = XFS_EXTLEN_MIN(gtlena, args->maxlen); | |
1183 | xfs_alloc_fix_len(args); | |
1184 | rlen = args->len; | |
1185 | gtdiff = xfs_alloc_compute_diff(args->agbno, rlen, | |
1186 | args->alignment, gtbno, gtlen, >new); | |
1187 | /* | |
1188 | * Right side entry isn't perfect. | |
1189 | */ | |
1190 | if (gtdiff) { | |
1191 | /* | |
1192 | * Look until we find a better one, run out of | |
1193 | * space, or run off the end. | |
1194 | */ | |
1195 | while (bno_cur_lt && bno_cur_gt) { | |
1196 | if ((error = xfs_alloc_get_rec( | |
1197 | bno_cur_lt, <bno, | |
1198 | <len, &i))) | |
1199 | goto error0; | |
1200 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1201 | xfs_alloc_compute_aligned(ltbno, ltlen, | |
1202 | args->alignment, args->minlen, | |
1203 | <bnoa, <lena); | |
1204 | /* | |
1205 | * The right one is clearly better. | |
1206 | */ | |
1207 | if (ltbnoa <= args->agbno - gtdiff) { | |
1208 | xfs_btree_del_cursor( | |
1209 | bno_cur_lt, | |
1210 | XFS_BTREE_NOERROR); | |
1211 | bno_cur_lt = NULL; | |
1212 | break; | |
1213 | } | |
1214 | /* | |
1215 | * If we reach a big enough entry, | |
1216 | * compare the two and pick the best. | |
1217 | */ | |
1218 | if (ltlena >= args->minlen) { | |
1219 | args->len = XFS_EXTLEN_MIN( | |
1220 | ltlena, args->maxlen); | |
1221 | xfs_alloc_fix_len(args); | |
1222 | rlen = args->len; | |
1223 | ltdiff = xfs_alloc_compute_diff( | |
1224 | args->agbno, rlen, | |
1225 | args->alignment, | |
1226 | ltbno, ltlen, <new); | |
1227 | /* | |
1228 | * Left side is better. | |
1229 | */ | |
1230 | if (ltdiff < gtdiff) { | |
1231 | xfs_btree_del_cursor( | |
1232 | bno_cur_gt, | |
1233 | XFS_BTREE_NOERROR); | |
1234 | bno_cur_gt = NULL; | |
1235 | } | |
1236 | /* | |
1237 | * Right side is better. | |
1238 | */ | |
1239 | else { | |
1240 | xfs_btree_del_cursor( | |
1241 | bno_cur_lt, | |
1242 | XFS_BTREE_NOERROR); | |
1243 | bno_cur_lt = NULL; | |
1244 | } | |
1245 | break; | |
1246 | } | |
1247 | /* | |
1248 | * Fell off the left end. | |
1249 | */ | |
8df4da4a | 1250 | if ((error = xfs_btree_decrement( |
1da177e4 LT |
1251 | bno_cur_lt, 0, &i))) |
1252 | goto error0; | |
1253 | if (!i) { | |
1254 | xfs_btree_del_cursor(bno_cur_lt, | |
1255 | XFS_BTREE_NOERROR); | |
1256 | bno_cur_lt = NULL; | |
1257 | break; | |
1258 | } | |
1259 | } | |
1260 | } | |
1261 | /* | |
1262 | * The right side is perfect, trash the left side. | |
1263 | */ | |
1264 | else { | |
1265 | xfs_btree_del_cursor(bno_cur_lt, | |
1266 | XFS_BTREE_NOERROR); | |
1267 | bno_cur_lt = NULL; | |
1268 | } | |
1269 | } | |
1270 | } | |
1271 | /* | |
1272 | * If we couldn't get anything, give up. | |
1273 | */ | |
1274 | if (bno_cur_lt == NULL && bno_cur_gt == NULL) { | |
1275 | TRACE_ALLOC("neither", args); | |
1276 | args->agbno = NULLAGBLOCK; | |
1277 | return 0; | |
1278 | } | |
1279 | /* | |
1280 | * At this point we have selected a freespace entry, either to the | |
1281 | * left or to the right. If it's on the right, copy all the | |
1282 | * useful variables to the "left" set so we only have one | |
1283 | * copy of this code. | |
1284 | */ | |
1285 | if (bno_cur_gt) { | |
1286 | bno_cur_lt = bno_cur_gt; | |
1287 | bno_cur_gt = NULL; | |
1288 | ltbno = gtbno; | |
1289 | ltbnoa = gtbnoa; | |
1290 | ltlen = gtlen; | |
1291 | ltlena = gtlena; | |
1292 | j = 1; | |
1293 | } else | |
1294 | j = 0; | |
1295 | /* | |
1296 | * Fix up the length and compute the useful address. | |
1297 | */ | |
1298 | ltend = ltbno + ltlen; | |
1299 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); | |
1300 | xfs_alloc_fix_len(args); | |
1301 | if (!xfs_alloc_fix_minleft(args)) { | |
1302 | TRACE_ALLOC("nominleft", args); | |
1303 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); | |
1304 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1305 | return 0; | |
1306 | } | |
1307 | rlen = args->len; | |
1308 | (void)xfs_alloc_compute_diff(args->agbno, rlen, args->alignment, ltbno, | |
1309 | ltlen, <new); | |
1310 | ASSERT(ltnew >= ltbno); | |
1311 | ASSERT(ltnew + rlen <= ltend); | |
16259e7d | 1312 | ASSERT(ltnew + rlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
1da177e4 LT |
1313 | args->agbno = ltnew; |
1314 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, ltlen, | |
1315 | ltnew, rlen, XFSA_FIXUP_BNO_OK))) | |
1316 | goto error0; | |
1317 | TRACE_ALLOC(j ? "gt" : "lt", args); | |
1318 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1319 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); | |
1320 | return 0; | |
1321 | ||
1322 | error0: | |
1323 | TRACE_ALLOC("error", args); | |
1324 | if (cnt_cur != NULL) | |
1325 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1326 | if (bno_cur_lt != NULL) | |
1327 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_ERROR); | |
1328 | if (bno_cur_gt != NULL) | |
1329 | xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_ERROR); | |
1330 | return error; | |
1331 | } | |
1332 | ||
1333 | /* | |
1334 | * Allocate a variable extent anywhere in the allocation group agno. | |
1335 | * Extent's length (returned in len) will be between minlen and maxlen, | |
1336 | * and of the form k * prod + mod unless there's nothing that large. | |
1337 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
1338 | */ | |
1339 | STATIC int /* error */ | |
1340 | xfs_alloc_ag_vextent_size( | |
1341 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
1342 | { | |
1343 | xfs_btree_cur_t *bno_cur; /* cursor for bno btree */ | |
1344 | xfs_btree_cur_t *cnt_cur; /* cursor for cnt btree */ | |
1345 | int error; /* error result */ | |
1346 | xfs_agblock_t fbno; /* start of found freespace */ | |
1347 | xfs_extlen_t flen; /* length of found freespace */ | |
1da177e4 LT |
1348 | int i; /* temp status variable */ |
1349 | xfs_agblock_t rbno; /* returned block number */ | |
1350 | xfs_extlen_t rlen; /* length of returned extent */ | |
1351 | ||
1352 | /* | |
1353 | * Allocate and initialize a cursor for the by-size btree. | |
1354 | */ | |
561f7d17 CH |
1355 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1356 | args->agno, XFS_BTNUM_CNT); | |
1da177e4 LT |
1357 | bno_cur = NULL; |
1358 | /* | |
1359 | * Look for an entry >= maxlen+alignment-1 blocks. | |
1360 | */ | |
1361 | if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, | |
1362 | args->maxlen + args->alignment - 1, &i))) | |
1363 | goto error0; | |
1364 | /* | |
1365 | * If none, then pick up the last entry in the tree unless the | |
1366 | * tree is empty. | |
1367 | */ | |
1368 | if (!i) { | |
1369 | if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, &fbno, | |
1370 | &flen, &i))) | |
1371 | goto error0; | |
1372 | if (i == 0 || flen == 0) { | |
1373 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1374 | TRACE_ALLOC("noentry", args); | |
1375 | return 0; | |
1376 | } | |
1377 | ASSERT(i == 1); | |
1378 | } | |
1379 | /* | |
1380 | * There's a freespace as big as maxlen+alignment-1, get it. | |
1381 | */ | |
1382 | else { | |
1383 | if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i))) | |
1384 | goto error0; | |
1385 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1386 | } | |
1387 | /* | |
1388 | * In the first case above, we got the last entry in the | |
1389 | * by-size btree. Now we check to see if the space hits maxlen | |
1390 | * once aligned; if not, we search left for something better. | |
1391 | * This can't happen in the second case above. | |
1392 | */ | |
1393 | xfs_alloc_compute_aligned(fbno, flen, args->alignment, args->minlen, | |
1394 | &rbno, &rlen); | |
1395 | rlen = XFS_EXTLEN_MIN(args->maxlen, rlen); | |
1396 | XFS_WANT_CORRUPTED_GOTO(rlen == 0 || | |
1397 | (rlen <= flen && rbno + rlen <= fbno + flen), error0); | |
1398 | if (rlen < args->maxlen) { | |
1399 | xfs_agblock_t bestfbno; | |
1400 | xfs_extlen_t bestflen; | |
1401 | xfs_agblock_t bestrbno; | |
1402 | xfs_extlen_t bestrlen; | |
1403 | ||
1404 | bestrlen = rlen; | |
1405 | bestrbno = rbno; | |
1406 | bestflen = flen; | |
1407 | bestfbno = fbno; | |
1408 | for (;;) { | |
8df4da4a | 1409 | if ((error = xfs_btree_decrement(cnt_cur, 0, &i))) |
1da177e4 LT |
1410 | goto error0; |
1411 | if (i == 0) | |
1412 | break; | |
1413 | if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, | |
1414 | &i))) | |
1415 | goto error0; | |
1416 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1417 | if (flen < bestrlen) | |
1418 | break; | |
1419 | xfs_alloc_compute_aligned(fbno, flen, args->alignment, | |
1420 | args->minlen, &rbno, &rlen); | |
1421 | rlen = XFS_EXTLEN_MIN(args->maxlen, rlen); | |
1422 | XFS_WANT_CORRUPTED_GOTO(rlen == 0 || | |
1423 | (rlen <= flen && rbno + rlen <= fbno + flen), | |
1424 | error0); | |
1425 | if (rlen > bestrlen) { | |
1426 | bestrlen = rlen; | |
1427 | bestrbno = rbno; | |
1428 | bestflen = flen; | |
1429 | bestfbno = fbno; | |
1430 | if (rlen == args->maxlen) | |
1431 | break; | |
1432 | } | |
1433 | } | |
1434 | if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen, | |
1435 | &i))) | |
1436 | goto error0; | |
1437 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1438 | rlen = bestrlen; | |
1439 | rbno = bestrbno; | |
1440 | flen = bestflen; | |
1441 | fbno = bestfbno; | |
1442 | } | |
1443 | args->wasfromfl = 0; | |
1444 | /* | |
1445 | * Fix up the length. | |
1446 | */ | |
1447 | args->len = rlen; | |
1448 | xfs_alloc_fix_len(args); | |
1449 | if (rlen < args->minlen || !xfs_alloc_fix_minleft(args)) { | |
1450 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1451 | TRACE_ALLOC("nominleft", args); | |
1452 | args->agbno = NULLAGBLOCK; | |
1453 | return 0; | |
1454 | } | |
1455 | rlen = args->len; | |
1456 | XFS_WANT_CORRUPTED_GOTO(rlen <= flen, error0); | |
1457 | /* | |
1458 | * Allocate and initialize a cursor for the by-block tree. | |
1459 | */ | |
561f7d17 CH |
1460 | bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1461 | args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
1462 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, |
1463 | rbno, rlen, XFSA_FIXUP_CNT_OK))) | |
1464 | goto error0; | |
1465 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1466 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
1467 | cnt_cur = bno_cur = NULL; | |
1468 | args->len = rlen; | |
1469 | args->agbno = rbno; | |
1470 | XFS_WANT_CORRUPTED_GOTO( | |
1471 | args->agbno + args->len <= | |
16259e7d | 1472 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length), |
1da177e4 LT |
1473 | error0); |
1474 | TRACE_ALLOC("normal", args); | |
1475 | return 0; | |
1476 | ||
1477 | error0: | |
1478 | TRACE_ALLOC("error", args); | |
1479 | if (cnt_cur) | |
1480 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1481 | if (bno_cur) | |
1482 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
1483 | return error; | |
1484 | } | |
1485 | ||
1486 | /* | |
1487 | * Deal with the case where only small freespaces remain. | |
1488 | * Either return the contents of the last freespace record, | |
1489 | * or allocate space from the freelist if there is nothing in the tree. | |
1490 | */ | |
1491 | STATIC int /* error */ | |
1492 | xfs_alloc_ag_vextent_small( | |
1493 | xfs_alloc_arg_t *args, /* allocation argument structure */ | |
1494 | xfs_btree_cur_t *ccur, /* by-size cursor */ | |
1495 | xfs_agblock_t *fbnop, /* result block number */ | |
1496 | xfs_extlen_t *flenp, /* result length */ | |
1497 | int *stat) /* status: 0-freelist, 1-normal/none */ | |
1498 | { | |
1499 | int error; | |
1500 | xfs_agblock_t fbno; | |
1501 | xfs_extlen_t flen; | |
1da177e4 LT |
1502 | int i; |
1503 | ||
8df4da4a | 1504 | if ((error = xfs_btree_decrement(ccur, 0, &i))) |
1da177e4 LT |
1505 | goto error0; |
1506 | if (i) { | |
1507 | if ((error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i))) | |
1508 | goto error0; | |
1509 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1510 | } | |
1511 | /* | |
1512 | * Nothing in the btree, try the freelist. Make sure | |
1513 | * to respect minleft even when pulling from the | |
1514 | * freelist. | |
1515 | */ | |
1516 | else if (args->minlen == 1 && args->alignment == 1 && !args->isfl && | |
16259e7d CH |
1517 | (be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_flcount) |
1518 | > args->minleft)) { | |
92821e2b DC |
1519 | error = xfs_alloc_get_freelist(args->tp, args->agbp, &fbno, 0); |
1520 | if (error) | |
1da177e4 LT |
1521 | goto error0; |
1522 | if (fbno != NULLAGBLOCK) { | |
1523 | if (args->userdata) { | |
1524 | xfs_buf_t *bp; | |
1525 | ||
1526 | bp = xfs_btree_get_bufs(args->mp, args->tp, | |
1527 | args->agno, fbno, 0); | |
1528 | xfs_trans_binval(args->tp, bp); | |
1529 | } | |
1530 | args->len = 1; | |
1531 | args->agbno = fbno; | |
1532 | XFS_WANT_CORRUPTED_GOTO( | |
1533 | args->agbno + args->len <= | |
16259e7d | 1534 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length), |
1da177e4 LT |
1535 | error0); |
1536 | args->wasfromfl = 1; | |
1537 | TRACE_ALLOC("freelist", args); | |
1538 | *stat = 0; | |
1539 | return 0; | |
1540 | } | |
1541 | /* | |
1542 | * Nothing in the freelist. | |
1543 | */ | |
1544 | else | |
1545 | flen = 0; | |
1546 | } | |
1547 | /* | |
1548 | * Can't allocate from the freelist for some reason. | |
1549 | */ | |
d432c80e NS |
1550 | else { |
1551 | fbno = NULLAGBLOCK; | |
1da177e4 | 1552 | flen = 0; |
d432c80e | 1553 | } |
1da177e4 LT |
1554 | /* |
1555 | * Can't do the allocation, give up. | |
1556 | */ | |
1557 | if (flen < args->minlen) { | |
1558 | args->agbno = NULLAGBLOCK; | |
1559 | TRACE_ALLOC("notenough", args); | |
1560 | flen = 0; | |
1561 | } | |
1562 | *fbnop = fbno; | |
1563 | *flenp = flen; | |
1564 | *stat = 1; | |
1565 | TRACE_ALLOC("normal", args); | |
1566 | return 0; | |
1567 | ||
1568 | error0: | |
1569 | TRACE_ALLOC("error", args); | |
1570 | return error; | |
1571 | } | |
1572 | ||
1573 | /* | |
1574 | * Free the extent starting at agno/bno for length. | |
1575 | */ | |
1576 | STATIC int /* error */ | |
1577 | xfs_free_ag_extent( | |
1578 | xfs_trans_t *tp, /* transaction pointer */ | |
1579 | xfs_buf_t *agbp, /* buffer for a.g. freelist header */ | |
1580 | xfs_agnumber_t agno, /* allocation group number */ | |
1581 | xfs_agblock_t bno, /* starting block number */ | |
1582 | xfs_extlen_t len, /* length of extent */ | |
1583 | int isfl) /* set if is freelist blocks - no sb acctg */ | |
1584 | { | |
1585 | xfs_btree_cur_t *bno_cur; /* cursor for by-block btree */ | |
1586 | xfs_btree_cur_t *cnt_cur; /* cursor for by-size btree */ | |
1587 | int error; /* error return value */ | |
1da177e4 LT |
1588 | xfs_agblock_t gtbno; /* start of right neighbor block */ |
1589 | xfs_extlen_t gtlen; /* length of right neighbor block */ | |
1590 | int haveleft; /* have a left neighbor block */ | |
1591 | int haveright; /* have a right neighbor block */ | |
1592 | int i; /* temp, result code */ | |
1593 | xfs_agblock_t ltbno; /* start of left neighbor block */ | |
1594 | xfs_extlen_t ltlen; /* length of left neighbor block */ | |
1595 | xfs_mount_t *mp; /* mount point struct for filesystem */ | |
1596 | xfs_agblock_t nbno; /* new starting block of freespace */ | |
1597 | xfs_extlen_t nlen; /* new length of freespace */ | |
1598 | ||
1599 | mp = tp->t_mountp; | |
1600 | /* | |
1601 | * Allocate and initialize a cursor for the by-block btree. | |
1602 | */ | |
561f7d17 | 1603 | bno_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_BNO); |
1da177e4 LT |
1604 | cnt_cur = NULL; |
1605 | /* | |
1606 | * Look for a neighboring block on the left (lower block numbers) | |
1607 | * that is contiguous with this space. | |
1608 | */ | |
1609 | if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft))) | |
1610 | goto error0; | |
1611 | if (haveleft) { | |
1612 | /* | |
1613 | * There is a block to our left. | |
1614 | */ | |
1615 | if ((error = xfs_alloc_get_rec(bno_cur, <bno, <len, &i))) | |
1616 | goto error0; | |
1617 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1618 | /* | |
1619 | * It's not contiguous, though. | |
1620 | */ | |
1621 | if (ltbno + ltlen < bno) | |
1622 | haveleft = 0; | |
1623 | else { | |
1624 | /* | |
1625 | * If this failure happens the request to free this | |
1626 | * space was invalid, it's (partly) already free. | |
1627 | * Very bad. | |
1628 | */ | |
1629 | XFS_WANT_CORRUPTED_GOTO(ltbno + ltlen <= bno, error0); | |
1630 | } | |
1631 | } | |
1632 | /* | |
1633 | * Look for a neighboring block on the right (higher block numbers) | |
1634 | * that is contiguous with this space. | |
1635 | */ | |
637aa50f | 1636 | if ((error = xfs_btree_increment(bno_cur, 0, &haveright))) |
1da177e4 LT |
1637 | goto error0; |
1638 | if (haveright) { | |
1639 | /* | |
1640 | * There is a block to our right. | |
1641 | */ | |
1642 | if ((error = xfs_alloc_get_rec(bno_cur, >bno, >len, &i))) | |
1643 | goto error0; | |
1644 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1645 | /* | |
1646 | * It's not contiguous, though. | |
1647 | */ | |
1648 | if (bno + len < gtbno) | |
1649 | haveright = 0; | |
1650 | else { | |
1651 | /* | |
1652 | * If this failure happens the request to free this | |
1653 | * space was invalid, it's (partly) already free. | |
1654 | * Very bad. | |
1655 | */ | |
1656 | XFS_WANT_CORRUPTED_GOTO(gtbno >= bno + len, error0); | |
1657 | } | |
1658 | } | |
1659 | /* | |
1660 | * Now allocate and initialize a cursor for the by-size tree. | |
1661 | */ | |
561f7d17 | 1662 | cnt_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_CNT); |
1da177e4 LT |
1663 | /* |
1664 | * Have both left and right contiguous neighbors. | |
1665 | * Merge all three into a single free block. | |
1666 | */ | |
1667 | if (haveleft && haveright) { | |
1668 | /* | |
1669 | * Delete the old by-size entry on the left. | |
1670 | */ | |
1671 | if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i))) | |
1672 | goto error0; | |
1673 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
91cca5df | 1674 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 LT |
1675 | goto error0; |
1676 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1677 | /* | |
1678 | * Delete the old by-size entry on the right. | |
1679 | */ | |
1680 | if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i))) | |
1681 | goto error0; | |
1682 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
91cca5df | 1683 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 LT |
1684 | goto error0; |
1685 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1686 | /* | |
1687 | * Delete the old by-block entry for the right block. | |
1688 | */ | |
91cca5df | 1689 | if ((error = xfs_btree_delete(bno_cur, &i))) |
1da177e4 LT |
1690 | goto error0; |
1691 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1692 | /* | |
1693 | * Move the by-block cursor back to the left neighbor. | |
1694 | */ | |
8df4da4a | 1695 | if ((error = xfs_btree_decrement(bno_cur, 0, &i))) |
1da177e4 LT |
1696 | goto error0; |
1697 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1698 | #ifdef DEBUG | |
1699 | /* | |
1700 | * Check that this is the right record: delete didn't | |
1701 | * mangle the cursor. | |
1702 | */ | |
1703 | { | |
1704 | xfs_agblock_t xxbno; | |
1705 | xfs_extlen_t xxlen; | |
1706 | ||
1707 | if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen, | |
1708 | &i))) | |
1709 | goto error0; | |
1710 | XFS_WANT_CORRUPTED_GOTO( | |
1711 | i == 1 && xxbno == ltbno && xxlen == ltlen, | |
1712 | error0); | |
1713 | } | |
1714 | #endif | |
1715 | /* | |
1716 | * Update remaining by-block entry to the new, joined block. | |
1717 | */ | |
1718 | nbno = ltbno; | |
1719 | nlen = len + ltlen + gtlen; | |
1720 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
1721 | goto error0; | |
1722 | } | |
1723 | /* | |
1724 | * Have only a left contiguous neighbor. | |
1725 | * Merge it together with the new freespace. | |
1726 | */ | |
1727 | else if (haveleft) { | |
1728 | /* | |
1729 | * Delete the old by-size entry on the left. | |
1730 | */ | |
1731 | if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i))) | |
1732 | goto error0; | |
1733 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
91cca5df | 1734 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 LT |
1735 | goto error0; |
1736 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1737 | /* | |
1738 | * Back up the by-block cursor to the left neighbor, and | |
1739 | * update its length. | |
1740 | */ | |
8df4da4a | 1741 | if ((error = xfs_btree_decrement(bno_cur, 0, &i))) |
1da177e4 LT |
1742 | goto error0; |
1743 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1744 | nbno = ltbno; | |
1745 | nlen = len + ltlen; | |
1746 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
1747 | goto error0; | |
1748 | } | |
1749 | /* | |
1750 | * Have only a right contiguous neighbor. | |
1751 | * Merge it together with the new freespace. | |
1752 | */ | |
1753 | else if (haveright) { | |
1754 | /* | |
1755 | * Delete the old by-size entry on the right. | |
1756 | */ | |
1757 | if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i))) | |
1758 | goto error0; | |
1759 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
91cca5df | 1760 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 LT |
1761 | goto error0; |
1762 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1763 | /* | |
1764 | * Update the starting block and length of the right | |
1765 | * neighbor in the by-block tree. | |
1766 | */ | |
1767 | nbno = bno; | |
1768 | nlen = len + gtlen; | |
1769 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
1770 | goto error0; | |
1771 | } | |
1772 | /* | |
1773 | * No contiguous neighbors. | |
1774 | * Insert the new freespace into the by-block tree. | |
1775 | */ | |
1776 | else { | |
1777 | nbno = bno; | |
1778 | nlen = len; | |
4b22a571 | 1779 | if ((error = xfs_btree_insert(bno_cur, &i))) |
1da177e4 LT |
1780 | goto error0; |
1781 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1782 | } | |
1783 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
1784 | bno_cur = NULL; | |
1785 | /* | |
1786 | * In all cases we need to insert the new freespace in the by-size tree. | |
1787 | */ | |
1788 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i))) | |
1789 | goto error0; | |
1790 | XFS_WANT_CORRUPTED_GOTO(i == 0, error0); | |
4b22a571 | 1791 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
1da177e4 LT |
1792 | goto error0; |
1793 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1794 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1795 | cnt_cur = NULL; | |
1796 | /* | |
1797 | * Update the freespace totals in the ag and superblock. | |
1798 | */ | |
1799 | { | |
1800 | xfs_agf_t *agf; | |
1801 | xfs_perag_t *pag; /* per allocation group data */ | |
1802 | ||
1803 | agf = XFS_BUF_TO_AGF(agbp); | |
1804 | pag = &mp->m_perag[agno]; | |
413d57c9 | 1805 | be32_add_cpu(&agf->agf_freeblks, len); |
1da177e4 LT |
1806 | xfs_trans_agblocks_delta(tp, len); |
1807 | pag->pagf_freeblks += len; | |
1808 | XFS_WANT_CORRUPTED_GOTO( | |
16259e7d CH |
1809 | be32_to_cpu(agf->agf_freeblks) <= |
1810 | be32_to_cpu(agf->agf_length), | |
1da177e4 LT |
1811 | error0); |
1812 | TRACE_MODAGF(NULL, agf, XFS_AGF_FREEBLKS); | |
1813 | xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS); | |
1814 | if (!isfl) | |
1815 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, (long)len); | |
1816 | XFS_STATS_INC(xs_freex); | |
1817 | XFS_STATS_ADD(xs_freeb, len); | |
1818 | } | |
1819 | TRACE_FREE(haveleft ? | |
1820 | (haveright ? "both" : "left") : | |
1821 | (haveright ? "right" : "none"), | |
1822 | agno, bno, len, isfl); | |
1823 | ||
1824 | /* | |
1825 | * Since blocks move to the free list without the coordination | |
1826 | * used in xfs_bmap_finish, we can't allow block to be available | |
1827 | * for reallocation and non-transaction writing (user data) | |
1828 | * until we know that the transaction that moved it to the free | |
1829 | * list is permanently on disk. We track the blocks by declaring | |
1830 | * these blocks as "busy"; the busy list is maintained on a per-ag | |
1831 | * basis and each transaction records which entries should be removed | |
1832 | * when the iclog commits to disk. If a busy block is allocated, | |
1833 | * the iclog is pushed up to the LSN that freed the block. | |
1834 | */ | |
1835 | xfs_alloc_mark_busy(tp, agno, bno, len); | |
1836 | return 0; | |
1837 | ||
1838 | error0: | |
1839 | TRACE_FREE("error", agno, bno, len, isfl); | |
1840 | if (bno_cur) | |
1841 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
1842 | if (cnt_cur) | |
1843 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1844 | return error; | |
1845 | } | |
1846 | ||
1847 | /* | |
1848 | * Visible (exported) allocation/free functions. | |
1849 | * Some of these are used just by xfs_alloc_btree.c and this file. | |
1850 | */ | |
1851 | ||
1852 | /* | |
1853 | * Compute and fill in value of m_ag_maxlevels. | |
1854 | */ | |
1855 | void | |
1856 | xfs_alloc_compute_maxlevels( | |
1857 | xfs_mount_t *mp) /* file system mount structure */ | |
1858 | { | |
1859 | int level; | |
1860 | uint maxblocks; | |
1861 | uint maxleafents; | |
1862 | int minleafrecs; | |
1863 | int minnoderecs; | |
1864 | ||
1865 | maxleafents = (mp->m_sb.sb_agblocks + 1) / 2; | |
1866 | minleafrecs = mp->m_alloc_mnr[0]; | |
1867 | minnoderecs = mp->m_alloc_mnr[1]; | |
1868 | maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs; | |
1869 | for (level = 1; maxblocks > 1; level++) | |
1870 | maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs; | |
1871 | mp->m_ag_maxlevels = level; | |
1872 | } | |
1873 | ||
6cc87645 DC |
1874 | /* |
1875 | * Find the length of the longest extent in an AG. | |
1876 | */ | |
1877 | xfs_extlen_t | |
1878 | xfs_alloc_longest_free_extent( | |
1879 | struct xfs_mount *mp, | |
1880 | struct xfs_perag *pag) | |
1881 | { | |
1882 | xfs_extlen_t need, delta = 0; | |
1883 | ||
1884 | need = XFS_MIN_FREELIST_PAG(pag, mp); | |
1885 | if (need > pag->pagf_flcount) | |
1886 | delta = need - pag->pagf_flcount; | |
1887 | ||
1888 | if (pag->pagf_longest > delta) | |
1889 | return pag->pagf_longest - delta; | |
1890 | return pag->pagf_flcount > 0 || pag->pagf_longest > 0; | |
1891 | } | |
1892 | ||
1da177e4 LT |
1893 | /* |
1894 | * Decide whether to use this allocation group for this allocation. | |
1895 | * If so, fix up the btree freelist's size. | |
1896 | */ | |
1897 | STATIC int /* error */ | |
1898 | xfs_alloc_fix_freelist( | |
1899 | xfs_alloc_arg_t *args, /* allocation argument structure */ | |
1900 | int flags) /* XFS_ALLOC_FLAG_... */ | |
1901 | { | |
1902 | xfs_buf_t *agbp; /* agf buffer pointer */ | |
1903 | xfs_agf_t *agf; /* a.g. freespace structure pointer */ | |
1904 | xfs_buf_t *agflbp;/* agfl buffer pointer */ | |
1905 | xfs_agblock_t bno; /* freelist block */ | |
1906 | xfs_extlen_t delta; /* new blocks needed in freelist */ | |
1907 | int error; /* error result code */ | |
1908 | xfs_extlen_t longest;/* longest extent in allocation group */ | |
1909 | xfs_mount_t *mp; /* file system mount point structure */ | |
1910 | xfs_extlen_t need; /* total blocks needed in freelist */ | |
1911 | xfs_perag_t *pag; /* per-ag information structure */ | |
1912 | xfs_alloc_arg_t targs; /* local allocation arguments */ | |
1913 | xfs_trans_t *tp; /* transaction pointer */ | |
1914 | ||
1915 | mp = args->mp; | |
1916 | ||
1917 | pag = args->pag; | |
1918 | tp = args->tp; | |
1919 | if (!pag->pagf_init) { | |
1920 | if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags, | |
1921 | &agbp))) | |
1922 | return error; | |
1923 | if (!pag->pagf_init) { | |
0e1edbd9 NS |
1924 | ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK); |
1925 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
1da177e4 LT |
1926 | args->agbp = NULL; |
1927 | return 0; | |
1928 | } | |
1929 | } else | |
1930 | agbp = NULL; | |
1931 | ||
0e1edbd9 NS |
1932 | /* |
1933 | * If this is a metadata preferred pag and we are user data | |
1da177e4 LT |
1934 | * then try somewhere else if we are not being asked to |
1935 | * try harder at this point | |
1936 | */ | |
0e1edbd9 NS |
1937 | if (pag->pagf_metadata && args->userdata && |
1938 | (flags & XFS_ALLOC_FLAG_TRYLOCK)) { | |
1939 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
1da177e4 LT |
1940 | args->agbp = NULL; |
1941 | return 0; | |
1942 | } | |
1943 | ||
0e1edbd9 | 1944 | if (!(flags & XFS_ALLOC_FLAG_FREEING)) { |
0e1edbd9 NS |
1945 | /* |
1946 | * If it looks like there isn't a long enough extent, or enough | |
1947 | * total blocks, reject it. | |
1948 | */ | |
6cc87645 DC |
1949 | need = XFS_MIN_FREELIST_PAG(pag, mp); |
1950 | longest = xfs_alloc_longest_free_extent(mp, pag); | |
0e1edbd9 NS |
1951 | if ((args->minlen + args->alignment + args->minalignslop - 1) > |
1952 | longest || | |
1953 | ((int)(pag->pagf_freeblks + pag->pagf_flcount - | |
1954 | need - args->total) < (int)args->minleft)) { | |
1955 | if (agbp) | |
1956 | xfs_trans_brelse(tp, agbp); | |
1957 | args->agbp = NULL; | |
1958 | return 0; | |
1959 | } | |
1da177e4 | 1960 | } |
0e1edbd9 | 1961 | |
1da177e4 LT |
1962 | /* |
1963 | * Get the a.g. freespace buffer. | |
1964 | * Can fail if we're not blocking on locks, and it's held. | |
1965 | */ | |
1966 | if (agbp == NULL) { | |
1967 | if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags, | |
1968 | &agbp))) | |
1969 | return error; | |
1970 | if (agbp == NULL) { | |
0e1edbd9 NS |
1971 | ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK); |
1972 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
1da177e4 LT |
1973 | args->agbp = NULL; |
1974 | return 0; | |
1975 | } | |
1976 | } | |
1977 | /* | |
1978 | * Figure out how many blocks we should have in the freelist. | |
1979 | */ | |
1980 | agf = XFS_BUF_TO_AGF(agbp); | |
1981 | need = XFS_MIN_FREELIST(agf, mp); | |
1da177e4 LT |
1982 | /* |
1983 | * If there isn't enough total or single-extent, reject it. | |
1984 | */ | |
0e1edbd9 NS |
1985 | if (!(flags & XFS_ALLOC_FLAG_FREEING)) { |
1986 | delta = need > be32_to_cpu(agf->agf_flcount) ? | |
1987 | (need - be32_to_cpu(agf->agf_flcount)) : 0; | |
1988 | longest = be32_to_cpu(agf->agf_longest); | |
1989 | longest = (longest > delta) ? (longest - delta) : | |
1990 | (be32_to_cpu(agf->agf_flcount) > 0 || longest > 0); | |
1991 | if ((args->minlen + args->alignment + args->minalignslop - 1) > | |
1992 | longest || | |
1993 | ((int)(be32_to_cpu(agf->agf_freeblks) + | |
1994 | be32_to_cpu(agf->agf_flcount) - need - args->total) < | |
1995 | (int)args->minleft)) { | |
1996 | xfs_trans_brelse(tp, agbp); | |
1997 | args->agbp = NULL; | |
1998 | return 0; | |
1999 | } | |
1da177e4 LT |
2000 | } |
2001 | /* | |
2002 | * Make the freelist shorter if it's too long. | |
2003 | */ | |
16259e7d | 2004 | while (be32_to_cpu(agf->agf_flcount) > need) { |
1da177e4 LT |
2005 | xfs_buf_t *bp; |
2006 | ||
92821e2b DC |
2007 | error = xfs_alloc_get_freelist(tp, agbp, &bno, 0); |
2008 | if (error) | |
1da177e4 LT |
2009 | return error; |
2010 | if ((error = xfs_free_ag_extent(tp, agbp, args->agno, bno, 1, 1))) | |
2011 | return error; | |
2012 | bp = xfs_btree_get_bufs(mp, tp, args->agno, bno, 0); | |
2013 | xfs_trans_binval(tp, bp); | |
2014 | } | |
2015 | /* | |
2016 | * Initialize the args structure. | |
2017 | */ | |
2018 | targs.tp = tp; | |
2019 | targs.mp = mp; | |
2020 | targs.agbp = agbp; | |
2021 | targs.agno = args->agno; | |
2022 | targs.mod = targs.minleft = targs.wasdel = targs.userdata = | |
2023 | targs.minalignslop = 0; | |
2024 | targs.alignment = targs.minlen = targs.prod = targs.isfl = 1; | |
2025 | targs.type = XFS_ALLOCTYPE_THIS_AG; | |
2026 | targs.pag = pag; | |
2027 | if ((error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp))) | |
2028 | return error; | |
2029 | /* | |
2030 | * Make the freelist longer if it's too short. | |
2031 | */ | |
16259e7d | 2032 | while (be32_to_cpu(agf->agf_flcount) < need) { |
1da177e4 | 2033 | targs.agbno = 0; |
16259e7d | 2034 | targs.maxlen = need - be32_to_cpu(agf->agf_flcount); |
1da177e4 LT |
2035 | /* |
2036 | * Allocate as many blocks as possible at once. | |
2037 | */ | |
e63a3690 NS |
2038 | if ((error = xfs_alloc_ag_vextent(&targs))) { |
2039 | xfs_trans_brelse(tp, agflbp); | |
1da177e4 | 2040 | return error; |
e63a3690 | 2041 | } |
1da177e4 LT |
2042 | /* |
2043 | * Stop if we run out. Won't happen if callers are obeying | |
2044 | * the restrictions correctly. Can happen for free calls | |
2045 | * on a completely full ag. | |
2046 | */ | |
d210a28c | 2047 | if (targs.agbno == NULLAGBLOCK) { |
0e1edbd9 NS |
2048 | if (flags & XFS_ALLOC_FLAG_FREEING) |
2049 | break; | |
2050 | xfs_trans_brelse(tp, agflbp); | |
2051 | args->agbp = NULL; | |
2052 | return 0; | |
d210a28c | 2053 | } |
1da177e4 LT |
2054 | /* |
2055 | * Put each allocated block on the list. | |
2056 | */ | |
2057 | for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) { | |
92821e2b DC |
2058 | error = xfs_alloc_put_freelist(tp, agbp, |
2059 | agflbp, bno, 0); | |
2060 | if (error) | |
1da177e4 LT |
2061 | return error; |
2062 | } | |
2063 | } | |
e63a3690 | 2064 | xfs_trans_brelse(tp, agflbp); |
1da177e4 LT |
2065 | args->agbp = agbp; |
2066 | return 0; | |
2067 | } | |
2068 | ||
2069 | /* | |
2070 | * Get a block from the freelist. | |
2071 | * Returns with the buffer for the block gotten. | |
2072 | */ | |
2073 | int /* error */ | |
2074 | xfs_alloc_get_freelist( | |
2075 | xfs_trans_t *tp, /* transaction pointer */ | |
2076 | xfs_buf_t *agbp, /* buffer containing the agf structure */ | |
92821e2b DC |
2077 | xfs_agblock_t *bnop, /* block address retrieved from freelist */ |
2078 | int btreeblk) /* destination is a AGF btree */ | |
1da177e4 LT |
2079 | { |
2080 | xfs_agf_t *agf; /* a.g. freespace structure */ | |
2081 | xfs_agfl_t *agfl; /* a.g. freelist structure */ | |
2082 | xfs_buf_t *agflbp;/* buffer for a.g. freelist structure */ | |
2083 | xfs_agblock_t bno; /* block number returned */ | |
2084 | int error; | |
92821e2b | 2085 | int logflags; |
1da177e4 LT |
2086 | xfs_mount_t *mp; /* mount structure */ |
2087 | xfs_perag_t *pag; /* per allocation group data */ | |
2088 | ||
2089 | agf = XFS_BUF_TO_AGF(agbp); | |
2090 | /* | |
2091 | * Freelist is empty, give up. | |
2092 | */ | |
2093 | if (!agf->agf_flcount) { | |
2094 | *bnop = NULLAGBLOCK; | |
2095 | return 0; | |
2096 | } | |
2097 | /* | |
2098 | * Read the array of free blocks. | |
2099 | */ | |
2100 | mp = tp->t_mountp; | |
2101 | if ((error = xfs_alloc_read_agfl(mp, tp, | |
16259e7d | 2102 | be32_to_cpu(agf->agf_seqno), &agflbp))) |
1da177e4 LT |
2103 | return error; |
2104 | agfl = XFS_BUF_TO_AGFL(agflbp); | |
2105 | /* | |
2106 | * Get the block number and update the data structures. | |
2107 | */ | |
e2101005 | 2108 | bno = be32_to_cpu(agfl->agfl_bno[be32_to_cpu(agf->agf_flfirst)]); |
413d57c9 | 2109 | be32_add_cpu(&agf->agf_flfirst, 1); |
1da177e4 | 2110 | xfs_trans_brelse(tp, agflbp); |
16259e7d | 2111 | if (be32_to_cpu(agf->agf_flfirst) == XFS_AGFL_SIZE(mp)) |
1da177e4 | 2112 | agf->agf_flfirst = 0; |
16259e7d | 2113 | pag = &mp->m_perag[be32_to_cpu(agf->agf_seqno)]; |
413d57c9 | 2114 | be32_add_cpu(&agf->agf_flcount, -1); |
1da177e4 LT |
2115 | xfs_trans_agflist_delta(tp, -1); |
2116 | pag->pagf_flcount--; | |
92821e2b DC |
2117 | |
2118 | logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT; | |
2119 | if (btreeblk) { | |
413d57c9 | 2120 | be32_add_cpu(&agf->agf_btreeblks, 1); |
92821e2b DC |
2121 | pag->pagf_btreeblks++; |
2122 | logflags |= XFS_AGF_BTREEBLKS; | |
2123 | } | |
2124 | ||
2125 | TRACE_MODAGF(NULL, agf, logflags); | |
2126 | xfs_alloc_log_agf(tp, agbp, logflags); | |
1da177e4 LT |
2127 | *bnop = bno; |
2128 | ||
2129 | /* | |
2130 | * As blocks are freed, they are added to the per-ag busy list | |
2131 | * and remain there until the freeing transaction is committed to | |
2132 | * disk. Now that we have allocated blocks, this list must be | |
2133 | * searched to see if a block is being reused. If one is, then | |
2134 | * the freeing transaction must be pushed to disk NOW by forcing | |
2135 | * to disk all iclogs up that transaction's LSN. | |
2136 | */ | |
16259e7d | 2137 | xfs_alloc_search_busy(tp, be32_to_cpu(agf->agf_seqno), bno, 1); |
1da177e4 LT |
2138 | return 0; |
2139 | } | |
2140 | ||
2141 | /* | |
2142 | * Log the given fields from the agf structure. | |
2143 | */ | |
2144 | void | |
2145 | xfs_alloc_log_agf( | |
2146 | xfs_trans_t *tp, /* transaction pointer */ | |
2147 | xfs_buf_t *bp, /* buffer for a.g. freelist header */ | |
2148 | int fields) /* mask of fields to be logged (XFS_AGF_...) */ | |
2149 | { | |
2150 | int first; /* first byte offset */ | |
2151 | int last; /* last byte offset */ | |
2152 | static const short offsets[] = { | |
2153 | offsetof(xfs_agf_t, agf_magicnum), | |
2154 | offsetof(xfs_agf_t, agf_versionnum), | |
2155 | offsetof(xfs_agf_t, agf_seqno), | |
2156 | offsetof(xfs_agf_t, agf_length), | |
2157 | offsetof(xfs_agf_t, agf_roots[0]), | |
2158 | offsetof(xfs_agf_t, agf_levels[0]), | |
2159 | offsetof(xfs_agf_t, agf_flfirst), | |
2160 | offsetof(xfs_agf_t, agf_fllast), | |
2161 | offsetof(xfs_agf_t, agf_flcount), | |
2162 | offsetof(xfs_agf_t, agf_freeblks), | |
2163 | offsetof(xfs_agf_t, agf_longest), | |
92821e2b | 2164 | offsetof(xfs_agf_t, agf_btreeblks), |
1da177e4 LT |
2165 | sizeof(xfs_agf_t) |
2166 | }; | |
2167 | ||
2168 | xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last); | |
2169 | xfs_trans_log_buf(tp, bp, (uint)first, (uint)last); | |
2170 | } | |
2171 | ||
2172 | /* | |
2173 | * Interface for inode allocation to force the pag data to be initialized. | |
2174 | */ | |
2175 | int /* error */ | |
2176 | xfs_alloc_pagf_init( | |
2177 | xfs_mount_t *mp, /* file system mount structure */ | |
2178 | xfs_trans_t *tp, /* transaction pointer */ | |
2179 | xfs_agnumber_t agno, /* allocation group number */ | |
2180 | int flags) /* XFS_ALLOC_FLAGS_... */ | |
2181 | { | |
2182 | xfs_buf_t *bp; | |
2183 | int error; | |
2184 | ||
2185 | if ((error = xfs_alloc_read_agf(mp, tp, agno, flags, &bp))) | |
2186 | return error; | |
2187 | if (bp) | |
2188 | xfs_trans_brelse(tp, bp); | |
2189 | return 0; | |
2190 | } | |
2191 | ||
2192 | /* | |
2193 | * Put the block on the freelist for the allocation group. | |
2194 | */ | |
2195 | int /* error */ | |
2196 | xfs_alloc_put_freelist( | |
2197 | xfs_trans_t *tp, /* transaction pointer */ | |
2198 | xfs_buf_t *agbp, /* buffer for a.g. freelist header */ | |
2199 | xfs_buf_t *agflbp,/* buffer for a.g. free block array */ | |
92821e2b DC |
2200 | xfs_agblock_t bno, /* block being freed */ |
2201 | int btreeblk) /* block came from a AGF btree */ | |
1da177e4 LT |
2202 | { |
2203 | xfs_agf_t *agf; /* a.g. freespace structure */ | |
2204 | xfs_agfl_t *agfl; /* a.g. free block array */ | |
e2101005 | 2205 | __be32 *blockp;/* pointer to array entry */ |
1da177e4 | 2206 | int error; |
92821e2b | 2207 | int logflags; |
1da177e4 LT |
2208 | xfs_mount_t *mp; /* mount structure */ |
2209 | xfs_perag_t *pag; /* per allocation group data */ | |
2210 | ||
2211 | agf = XFS_BUF_TO_AGF(agbp); | |
2212 | mp = tp->t_mountp; | |
2213 | ||
2214 | if (!agflbp && (error = xfs_alloc_read_agfl(mp, tp, | |
16259e7d | 2215 | be32_to_cpu(agf->agf_seqno), &agflbp))) |
1da177e4 LT |
2216 | return error; |
2217 | agfl = XFS_BUF_TO_AGFL(agflbp); | |
413d57c9 | 2218 | be32_add_cpu(&agf->agf_fllast, 1); |
16259e7d | 2219 | if (be32_to_cpu(agf->agf_fllast) == XFS_AGFL_SIZE(mp)) |
1da177e4 | 2220 | agf->agf_fllast = 0; |
16259e7d | 2221 | pag = &mp->m_perag[be32_to_cpu(agf->agf_seqno)]; |
413d57c9 | 2222 | be32_add_cpu(&agf->agf_flcount, 1); |
1da177e4 LT |
2223 | xfs_trans_agflist_delta(tp, 1); |
2224 | pag->pagf_flcount++; | |
92821e2b DC |
2225 | |
2226 | logflags = XFS_AGF_FLLAST | XFS_AGF_FLCOUNT; | |
2227 | if (btreeblk) { | |
413d57c9 | 2228 | be32_add_cpu(&agf->agf_btreeblks, -1); |
92821e2b DC |
2229 | pag->pagf_btreeblks--; |
2230 | logflags |= XFS_AGF_BTREEBLKS; | |
2231 | } | |
2232 | ||
2233 | TRACE_MODAGF(NULL, agf, logflags); | |
2234 | xfs_alloc_log_agf(tp, agbp, logflags); | |
2235 | ||
16259e7d CH |
2236 | ASSERT(be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp)); |
2237 | blockp = &agfl->agfl_bno[be32_to_cpu(agf->agf_fllast)]; | |
e2101005 | 2238 | *blockp = cpu_to_be32(bno); |
92821e2b DC |
2239 | TRACE_MODAGF(NULL, agf, logflags); |
2240 | xfs_alloc_log_agf(tp, agbp, logflags); | |
1da177e4 LT |
2241 | xfs_trans_log_buf(tp, agflbp, |
2242 | (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl), | |
2243 | (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl + | |
2244 | sizeof(xfs_agblock_t) - 1)); | |
2245 | return 0; | |
2246 | } | |
2247 | ||
2248 | /* | |
2249 | * Read in the allocation group header (free/alloc section). | |
2250 | */ | |
2251 | int /* error */ | |
4805621a FCH |
2252 | xfs_read_agf( |
2253 | struct xfs_mount *mp, /* mount point structure */ | |
2254 | struct xfs_trans *tp, /* transaction pointer */ | |
2255 | xfs_agnumber_t agno, /* allocation group number */ | |
2256 | int flags, /* XFS_BUF_ */ | |
2257 | struct xfs_buf **bpp) /* buffer for the ag freelist header */ | |
1da177e4 | 2258 | { |
4805621a | 2259 | struct xfs_agf *agf; /* ag freelist header */ |
1da177e4 | 2260 | int agf_ok; /* set if agf is consistent */ |
1da177e4 LT |
2261 | int error; |
2262 | ||
2263 | ASSERT(agno != NULLAGNUMBER); | |
2264 | error = xfs_trans_read_buf( | |
2265 | mp, tp, mp->m_ddev_targp, | |
2266 | XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)), | |
4805621a | 2267 | XFS_FSS_TO_BB(mp, 1), flags, bpp); |
1da177e4 LT |
2268 | if (error) |
2269 | return error; | |
4805621a | 2270 | if (!*bpp) |
1da177e4 | 2271 | return 0; |
4805621a FCH |
2272 | |
2273 | ASSERT(!XFS_BUF_GETERROR(*bpp)); | |
2274 | agf = XFS_BUF_TO_AGF(*bpp); | |
2275 | ||
1da177e4 LT |
2276 | /* |
2277 | * Validate the magic number of the agf block. | |
2278 | */ | |
1da177e4 | 2279 | agf_ok = |
16259e7d CH |
2280 | be32_to_cpu(agf->agf_magicnum) == XFS_AGF_MAGIC && |
2281 | XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) && | |
2282 | be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) && | |
2283 | be32_to_cpu(agf->agf_flfirst) < XFS_AGFL_SIZE(mp) && | |
2284 | be32_to_cpu(agf->agf_fllast) < XFS_AGFL_SIZE(mp) && | |
4805621a FCH |
2285 | be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp) && |
2286 | be32_to_cpu(agf->agf_seqno) == agno; | |
89b28393 BN |
2287 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
2288 | agf_ok = agf_ok && be32_to_cpu(agf->agf_btreeblks) <= | |
2289 | be32_to_cpu(agf->agf_length); | |
1da177e4 LT |
2290 | if (unlikely(XFS_TEST_ERROR(!agf_ok, mp, XFS_ERRTAG_ALLOC_READ_AGF, |
2291 | XFS_RANDOM_ALLOC_READ_AGF))) { | |
2292 | XFS_CORRUPTION_ERROR("xfs_alloc_read_agf", | |
2293 | XFS_ERRLEVEL_LOW, mp, agf); | |
4805621a | 2294 | xfs_trans_brelse(tp, *bpp); |
1da177e4 LT |
2295 | return XFS_ERROR(EFSCORRUPTED); |
2296 | } | |
4805621a FCH |
2297 | |
2298 | XFS_BUF_SET_VTYPE_REF(*bpp, B_FS_AGF, XFS_AGF_REF); | |
2299 | return 0; | |
2300 | } | |
2301 | ||
2302 | /* | |
2303 | * Read in the allocation group header (free/alloc section). | |
2304 | */ | |
2305 | int /* error */ | |
2306 | xfs_alloc_read_agf( | |
2307 | struct xfs_mount *mp, /* mount point structure */ | |
2308 | struct xfs_trans *tp, /* transaction pointer */ | |
2309 | xfs_agnumber_t agno, /* allocation group number */ | |
2310 | int flags, /* XFS_ALLOC_FLAG_... */ | |
2311 | struct xfs_buf **bpp) /* buffer for the ag freelist header */ | |
2312 | { | |
2313 | struct xfs_agf *agf; /* ag freelist header */ | |
2314 | struct xfs_perag *pag; /* per allocation group data */ | |
2315 | int error; | |
2316 | ||
2317 | ASSERT(agno != NULLAGNUMBER); | |
2318 | ||
2319 | error = xfs_read_agf(mp, tp, agno, | |
2320 | (flags & XFS_ALLOC_FLAG_TRYLOCK) ? XFS_BUF_TRYLOCK : 0, | |
2321 | bpp); | |
2322 | if (error) | |
2323 | return error; | |
2324 | if (!*bpp) | |
2325 | return 0; | |
2326 | ASSERT(!XFS_BUF_GETERROR(*bpp)); | |
2327 | ||
2328 | agf = XFS_BUF_TO_AGF(*bpp); | |
1da177e4 LT |
2329 | pag = &mp->m_perag[agno]; |
2330 | if (!pag->pagf_init) { | |
16259e7d | 2331 | pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks); |
92821e2b | 2332 | pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks); |
16259e7d CH |
2333 | pag->pagf_flcount = be32_to_cpu(agf->agf_flcount); |
2334 | pag->pagf_longest = be32_to_cpu(agf->agf_longest); | |
1da177e4 | 2335 | pag->pagf_levels[XFS_BTNUM_BNOi] = |
16259e7d | 2336 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]); |
1da177e4 | 2337 | pag->pagf_levels[XFS_BTNUM_CNTi] = |
16259e7d | 2338 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]); |
007c61c6 | 2339 | spin_lock_init(&pag->pagb_lock); |
1da177e4 LT |
2340 | pag->pagb_list = kmem_zalloc(XFS_PAGB_NUM_SLOTS * |
2341 | sizeof(xfs_perag_busy_t), KM_SLEEP); | |
2342 | pag->pagf_init = 1; | |
2343 | } | |
2344 | #ifdef DEBUG | |
2345 | else if (!XFS_FORCED_SHUTDOWN(mp)) { | |
16259e7d | 2346 | ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks)); |
89b28393 | 2347 | ASSERT(pag->pagf_btreeblks == be32_to_cpu(agf->agf_btreeblks)); |
16259e7d CH |
2348 | ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount)); |
2349 | ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest)); | |
1da177e4 | 2350 | ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] == |
16259e7d | 2351 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi])); |
1da177e4 | 2352 | ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] == |
16259e7d | 2353 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi])); |
1da177e4 LT |
2354 | } |
2355 | #endif | |
1da177e4 LT |
2356 | return 0; |
2357 | } | |
2358 | ||
2359 | /* | |
2360 | * Allocate an extent (variable-size). | |
2361 | * Depending on the allocation type, we either look in a single allocation | |
2362 | * group or loop over the allocation groups to find the result. | |
2363 | */ | |
2364 | int /* error */ | |
2365 | xfs_alloc_vextent( | |
2366 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
2367 | { | |
2368 | xfs_agblock_t agsize; /* allocation group size */ | |
2369 | int error; | |
2370 | int flags; /* XFS_ALLOC_FLAG_... locking flags */ | |
1da177e4 LT |
2371 | xfs_extlen_t minleft;/* minimum left value, temp copy */ |
2372 | xfs_mount_t *mp; /* mount structure pointer */ | |
2373 | xfs_agnumber_t sagno; /* starting allocation group number */ | |
2374 | xfs_alloctype_t type; /* input allocation type */ | |
2375 | int bump_rotor = 0; | |
2376 | int no_min = 0; | |
2377 | xfs_agnumber_t rotorstep = xfs_rotorstep; /* inode32 agf stepper */ | |
2378 | ||
2379 | mp = args->mp; | |
2380 | type = args->otype = args->type; | |
2381 | args->agbno = NULLAGBLOCK; | |
2382 | /* | |
2383 | * Just fix this up, for the case where the last a.g. is shorter | |
2384 | * (or there's only one a.g.) and the caller couldn't easily figure | |
2385 | * that out (xfs_bmap_alloc). | |
2386 | */ | |
2387 | agsize = mp->m_sb.sb_agblocks; | |
2388 | if (args->maxlen > agsize) | |
2389 | args->maxlen = agsize; | |
2390 | if (args->alignment == 0) | |
2391 | args->alignment = 1; | |
2392 | ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount); | |
2393 | ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize); | |
2394 | ASSERT(args->minlen <= args->maxlen); | |
2395 | ASSERT(args->minlen <= agsize); | |
2396 | ASSERT(args->mod < args->prod); | |
2397 | if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount || | |
2398 | XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize || | |
2399 | args->minlen > args->maxlen || args->minlen > agsize || | |
2400 | args->mod >= args->prod) { | |
2401 | args->fsbno = NULLFSBLOCK; | |
2402 | TRACE_ALLOC("badargs", args); | |
2403 | return 0; | |
2404 | } | |
2405 | minleft = args->minleft; | |
2406 | ||
2407 | switch (type) { | |
2408 | case XFS_ALLOCTYPE_THIS_AG: | |
2409 | case XFS_ALLOCTYPE_NEAR_BNO: | |
2410 | case XFS_ALLOCTYPE_THIS_BNO: | |
2411 | /* | |
2412 | * These three force us into a single a.g. | |
2413 | */ | |
2414 | args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
2415 | down_read(&mp->m_peraglock); | |
2416 | args->pag = &mp->m_perag[args->agno]; | |
2417 | args->minleft = 0; | |
2418 | error = xfs_alloc_fix_freelist(args, 0); | |
2419 | args->minleft = minleft; | |
2420 | if (error) { | |
2421 | TRACE_ALLOC("nofix", args); | |
2422 | goto error0; | |
2423 | } | |
2424 | if (!args->agbp) { | |
2425 | up_read(&mp->m_peraglock); | |
2426 | TRACE_ALLOC("noagbp", args); | |
2427 | break; | |
2428 | } | |
2429 | args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno); | |
2430 | if ((error = xfs_alloc_ag_vextent(args))) | |
2431 | goto error0; | |
2432 | up_read(&mp->m_peraglock); | |
2433 | break; | |
2434 | case XFS_ALLOCTYPE_START_BNO: | |
2435 | /* | |
2436 | * Try near allocation first, then anywhere-in-ag after | |
2437 | * the first a.g. fails. | |
2438 | */ | |
2439 | if ((args->userdata == XFS_ALLOC_INITIAL_USER_DATA) && | |
2440 | (mp->m_flags & XFS_MOUNT_32BITINODES)) { | |
2441 | args->fsbno = XFS_AGB_TO_FSB(mp, | |
2442 | ((mp->m_agfrotor / rotorstep) % | |
2443 | mp->m_sb.sb_agcount), 0); | |
2444 | bump_rotor = 1; | |
2445 | } | |
2446 | args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno); | |
2447 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
2448 | /* FALLTHROUGH */ | |
2449 | case XFS_ALLOCTYPE_ANY_AG: | |
2450 | case XFS_ALLOCTYPE_START_AG: | |
2451 | case XFS_ALLOCTYPE_FIRST_AG: | |
2452 | /* | |
2453 | * Rotate through the allocation groups looking for a winner. | |
2454 | */ | |
2455 | if (type == XFS_ALLOCTYPE_ANY_AG) { | |
2456 | /* | |
2457 | * Start with the last place we left off. | |
2458 | */ | |
2459 | args->agno = sagno = (mp->m_agfrotor / rotorstep) % | |
2460 | mp->m_sb.sb_agcount; | |
2461 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2462 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
2463 | } else if (type == XFS_ALLOCTYPE_FIRST_AG) { | |
2464 | /* | |
2465 | * Start with allocation group given by bno. | |
2466 | */ | |
2467 | args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
2468 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2469 | sagno = 0; | |
2470 | flags = 0; | |
2471 | } else { | |
2472 | if (type == XFS_ALLOCTYPE_START_AG) | |
2473 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2474 | /* | |
2475 | * Start with the given allocation group. | |
2476 | */ | |
2477 | args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
2478 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
2479 | } | |
2480 | /* | |
2481 | * Loop over allocation groups twice; first time with | |
2482 | * trylock set, second time without. | |
2483 | */ | |
2484 | down_read(&mp->m_peraglock); | |
2485 | for (;;) { | |
2486 | args->pag = &mp->m_perag[args->agno]; | |
2487 | if (no_min) args->minleft = 0; | |
2488 | error = xfs_alloc_fix_freelist(args, flags); | |
2489 | args->minleft = minleft; | |
2490 | if (error) { | |
2491 | TRACE_ALLOC("nofix", args); | |
2492 | goto error0; | |
2493 | } | |
2494 | /* | |
2495 | * If we get a buffer back then the allocation will fly. | |
2496 | */ | |
2497 | if (args->agbp) { | |
2498 | if ((error = xfs_alloc_ag_vextent(args))) | |
2499 | goto error0; | |
2500 | break; | |
2501 | } | |
2502 | TRACE_ALLOC("loopfailed", args); | |
2503 | /* | |
2504 | * Didn't work, figure out the next iteration. | |
2505 | */ | |
2506 | if (args->agno == sagno && | |
2507 | type == XFS_ALLOCTYPE_START_BNO) | |
2508 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
d210a28c YL |
2509 | /* |
2510 | * For the first allocation, we can try any AG to get | |
2511 | * space. However, if we already have allocated a | |
2512 | * block, we don't want to try AGs whose number is below | |
2513 | * sagno. Otherwise, we may end up with out-of-order | |
2514 | * locking of AGF, which might cause deadlock. | |
2515 | */ | |
2516 | if (++(args->agno) == mp->m_sb.sb_agcount) { | |
2517 | if (args->firstblock != NULLFSBLOCK) | |
2518 | args->agno = sagno; | |
2519 | else | |
2520 | args->agno = 0; | |
2521 | } | |
1da177e4 LT |
2522 | /* |
2523 | * Reached the starting a.g., must either be done | |
2524 | * or switch to non-trylock mode. | |
2525 | */ | |
2526 | if (args->agno == sagno) { | |
2527 | if (no_min == 1) { | |
2528 | args->agbno = NULLAGBLOCK; | |
2529 | TRACE_ALLOC("allfailed", args); | |
2530 | break; | |
2531 | } | |
2532 | if (flags == 0) { | |
2533 | no_min = 1; | |
2534 | } else { | |
2535 | flags = 0; | |
2536 | if (type == XFS_ALLOCTYPE_START_BNO) { | |
2537 | args->agbno = XFS_FSB_TO_AGBNO(mp, | |
2538 | args->fsbno); | |
2539 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
2540 | } | |
2541 | } | |
2542 | } | |
2543 | } | |
2544 | up_read(&mp->m_peraglock); | |
2545 | if (bump_rotor || (type == XFS_ALLOCTYPE_ANY_AG)) { | |
2546 | if (args->agno == sagno) | |
2547 | mp->m_agfrotor = (mp->m_agfrotor + 1) % | |
2548 | (mp->m_sb.sb_agcount * rotorstep); | |
2549 | else | |
2550 | mp->m_agfrotor = (args->agno * rotorstep + 1) % | |
2551 | (mp->m_sb.sb_agcount * rotorstep); | |
2552 | } | |
2553 | break; | |
2554 | default: | |
2555 | ASSERT(0); | |
2556 | /* NOTREACHED */ | |
2557 | } | |
2558 | if (args->agbno == NULLAGBLOCK) | |
2559 | args->fsbno = NULLFSBLOCK; | |
2560 | else { | |
2561 | args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno); | |
2562 | #ifdef DEBUG | |
2563 | ASSERT(args->len >= args->minlen); | |
2564 | ASSERT(args->len <= args->maxlen); | |
2565 | ASSERT(args->agbno % args->alignment == 0); | |
2566 | XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno), | |
2567 | args->len); | |
2568 | #endif | |
2569 | } | |
2570 | return 0; | |
2571 | error0: | |
2572 | up_read(&mp->m_peraglock); | |
2573 | return error; | |
2574 | } | |
2575 | ||
2576 | /* | |
2577 | * Free an extent. | |
2578 | * Just break up the extent address and hand off to xfs_free_ag_extent | |
2579 | * after fixing up the freelist. | |
2580 | */ | |
2581 | int /* error */ | |
2582 | xfs_free_extent( | |
2583 | xfs_trans_t *tp, /* transaction pointer */ | |
2584 | xfs_fsblock_t bno, /* starting block number of extent */ | |
2585 | xfs_extlen_t len) /* length of extent */ | |
2586 | { | |
0e1edbd9 | 2587 | xfs_alloc_arg_t args; |
1da177e4 LT |
2588 | int error; |
2589 | ||
2590 | ASSERT(len != 0); | |
0e1edbd9 | 2591 | memset(&args, 0, sizeof(xfs_alloc_arg_t)); |
1da177e4 LT |
2592 | args.tp = tp; |
2593 | args.mp = tp->t_mountp; | |
2594 | args.agno = XFS_FSB_TO_AGNO(args.mp, bno); | |
2595 | ASSERT(args.agno < args.mp->m_sb.sb_agcount); | |
2596 | args.agbno = XFS_FSB_TO_AGBNO(args.mp, bno); | |
1da177e4 LT |
2597 | down_read(&args.mp->m_peraglock); |
2598 | args.pag = &args.mp->m_perag[args.agno]; | |
d210a28c | 2599 | if ((error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING))) |
1da177e4 LT |
2600 | goto error0; |
2601 | #ifdef DEBUG | |
2602 | ASSERT(args.agbp != NULL); | |
0e1edbd9 NS |
2603 | ASSERT((args.agbno + len) <= |
2604 | be32_to_cpu(XFS_BUF_TO_AGF(args.agbp)->agf_length)); | |
1da177e4 | 2605 | #endif |
0e1edbd9 | 2606 | error = xfs_free_ag_extent(tp, args.agbp, args.agno, args.agbno, len, 0); |
1da177e4 LT |
2607 | error0: |
2608 | up_read(&args.mp->m_peraglock); | |
2609 | return error; | |
2610 | } | |
2611 | ||
2612 | ||
2613 | /* | |
2614 | * AG Busy list management | |
2615 | * The busy list contains block ranges that have been freed but whose | |
c41564b5 | 2616 | * transactions have not yet hit disk. If any block listed in a busy |
1da177e4 LT |
2617 | * list is reused, the transaction that freed it must be forced to disk |
2618 | * before continuing to use the block. | |
2619 | * | |
2620 | * xfs_alloc_mark_busy - add to the per-ag busy list | |
2621 | * xfs_alloc_clear_busy - remove an item from the per-ag busy list | |
2622 | */ | |
2623 | void | |
2624 | xfs_alloc_mark_busy(xfs_trans_t *tp, | |
2625 | xfs_agnumber_t agno, | |
2626 | xfs_agblock_t bno, | |
2627 | xfs_extlen_t len) | |
2628 | { | |
2629 | xfs_mount_t *mp; | |
2630 | xfs_perag_busy_t *bsy; | |
2631 | int n; | |
1da177e4 LT |
2632 | |
2633 | mp = tp->t_mountp; | |
64137e56 | 2634 | spin_lock(&mp->m_perag[agno].pagb_lock); |
1da177e4 LT |
2635 | |
2636 | /* search pagb_list for an open slot */ | |
2637 | for (bsy = mp->m_perag[agno].pagb_list, n = 0; | |
2638 | n < XFS_PAGB_NUM_SLOTS; | |
2639 | bsy++, n++) { | |
2640 | if (bsy->busy_tp == NULL) { | |
2641 | break; | |
2642 | } | |
2643 | } | |
2644 | ||
2645 | if (n < XFS_PAGB_NUM_SLOTS) { | |
2646 | bsy = &mp->m_perag[agno].pagb_list[n]; | |
2647 | mp->m_perag[agno].pagb_count++; | |
2648 | TRACE_BUSY("xfs_alloc_mark_busy", "got", agno, bno, len, n, tp); | |
2649 | bsy->busy_start = bno; | |
2650 | bsy->busy_length = len; | |
2651 | bsy->busy_tp = tp; | |
2652 | xfs_trans_add_busy(tp, agno, n); | |
2653 | } else { | |
2654 | TRACE_BUSY("xfs_alloc_mark_busy", "FULL", agno, bno, len, -1, tp); | |
2655 | /* | |
2656 | * The busy list is full! Since it is now not possible to | |
2657 | * track the free block, make this a synchronous transaction | |
2658 | * to insure that the block is not reused before this | |
2659 | * transaction commits. | |
2660 | */ | |
2661 | xfs_trans_set_sync(tp); | |
2662 | } | |
2663 | ||
64137e56 | 2664 | spin_unlock(&mp->m_perag[agno].pagb_lock); |
1da177e4 LT |
2665 | } |
2666 | ||
2667 | void | |
2668 | xfs_alloc_clear_busy(xfs_trans_t *tp, | |
2669 | xfs_agnumber_t agno, | |
2670 | int idx) | |
2671 | { | |
2672 | xfs_mount_t *mp; | |
2673 | xfs_perag_busy_t *list; | |
1da177e4 LT |
2674 | |
2675 | mp = tp->t_mountp; | |
2676 | ||
64137e56 | 2677 | spin_lock(&mp->m_perag[agno].pagb_lock); |
1da177e4 LT |
2678 | list = mp->m_perag[agno].pagb_list; |
2679 | ||
2680 | ASSERT(idx < XFS_PAGB_NUM_SLOTS); | |
2681 | if (list[idx].busy_tp == tp) { | |
2682 | TRACE_UNBUSY("xfs_alloc_clear_busy", "found", agno, idx, tp); | |
2683 | list[idx].busy_tp = NULL; | |
2684 | mp->m_perag[agno].pagb_count--; | |
2685 | } else { | |
2686 | TRACE_UNBUSY("xfs_alloc_clear_busy", "missing", agno, idx, tp); | |
2687 | } | |
2688 | ||
64137e56 | 2689 | spin_unlock(&mp->m_perag[agno].pagb_lock); |
1da177e4 LT |
2690 | } |
2691 | ||
2692 | ||
2693 | /* | |
f4586e40 DC |
2694 | * If we find the extent in the busy list, force the log out to get the |
2695 | * extent out of the busy list so the caller can use it straight away. | |
1da177e4 | 2696 | */ |
f4586e40 | 2697 | STATIC void |
1da177e4 LT |
2698 | xfs_alloc_search_busy(xfs_trans_t *tp, |
2699 | xfs_agnumber_t agno, | |
2700 | xfs_agblock_t bno, | |
2701 | xfs_extlen_t len) | |
2702 | { | |
2703 | xfs_mount_t *mp; | |
2704 | xfs_perag_busy_t *bsy; | |
1da177e4 LT |
2705 | xfs_agblock_t uend, bend; |
2706 | xfs_lsn_t lsn; | |
2707 | int cnt; | |
1da177e4 LT |
2708 | |
2709 | mp = tp->t_mountp; | |
2710 | ||
64137e56 | 2711 | spin_lock(&mp->m_perag[agno].pagb_lock); |
1da177e4 LT |
2712 | cnt = mp->m_perag[agno].pagb_count; |
2713 | ||
2714 | uend = bno + len - 1; | |
2715 | ||
2716 | /* search pagb_list for this slot, skipping open slots */ | |
f4586e40 | 2717 | for (bsy = mp->m_perag[agno].pagb_list; cnt; bsy++) { |
1da177e4 LT |
2718 | |
2719 | /* | |
2720 | * (start1,length1) within (start2, length2) | |
2721 | */ | |
2722 | if (bsy->busy_tp != NULL) { | |
2723 | bend = bsy->busy_start + bsy->busy_length - 1; | |
f4586e40 | 2724 | if ((bno > bend) || (uend < bsy->busy_start)) { |
1da177e4 LT |
2725 | cnt--; |
2726 | } else { | |
2727 | TRACE_BUSYSEARCH("xfs_alloc_search_busy", | |
f4586e40 | 2728 | "found1", agno, bno, len, tp); |
1da177e4 LT |
2729 | break; |
2730 | } | |
2731 | } | |
2732 | } | |
2733 | ||
2734 | /* | |
2735 | * If a block was found, force the log through the LSN of the | |
2736 | * transaction that freed the block | |
2737 | */ | |
2738 | if (cnt) { | |
f4586e40 | 2739 | TRACE_BUSYSEARCH("xfs_alloc_search_busy", "found", agno, bno, len, tp); |
1da177e4 | 2740 | lsn = bsy->busy_tp->t_commit_lsn; |
64137e56 | 2741 | spin_unlock(&mp->m_perag[agno].pagb_lock); |
1da177e4 LT |
2742 | xfs_log_force(mp, lsn, XFS_LOG_FORCE|XFS_LOG_SYNC); |
2743 | } else { | |
f4586e40 | 2744 | TRACE_BUSYSEARCH("xfs_alloc_search_busy", "not-found", agno, bno, len, tp); |
64137e56 | 2745 | spin_unlock(&mp->m_perag[agno].pagb_lock); |
1da177e4 | 2746 | } |
1da177e4 | 2747 | } |