diff options
-rw-r--r-- | fs/xfs/xfs_ialloc.c | 175 |
1 files changed, 95 insertions, 80 deletions
diff --git a/fs/xfs/xfs_ialloc.c b/fs/xfs/xfs_ialloc.c index 3120a3a5e20f..ce9edf7f4cb4 100644 --- a/fs/xfs/xfs_ialloc.c +++ b/fs/xfs/xfs_ialloc.c | |||
@@ -153,6 +153,87 @@ xfs_inobt_get_rec( | |||
153 | } | 153 | } |
154 | 154 | ||
155 | /* | 155 | /* |
156 | * Initialise a new set of inodes. | ||
157 | */ | ||
158 | STATIC void | ||
159 | xfs_ialloc_inode_init( | ||
160 | struct xfs_mount *mp, | ||
161 | struct xfs_trans *tp, | ||
162 | xfs_agnumber_t agno, | ||
163 | xfs_agblock_t agbno, | ||
164 | xfs_agblock_t length, | ||
165 | unsigned int gen) | ||
166 | { | ||
167 | struct xfs_buf *fbuf; | ||
168 | struct xfs_dinode *free; | ||
169 | int blks_per_cluster, nbufs, ninodes; | ||
170 | int version; | ||
171 | int i, j; | ||
172 | xfs_daddr_t d; | ||
173 | |||
174 | /* | ||
175 | * Loop over the new block(s), filling in the inodes. | ||
176 | * For small block sizes, manipulate the inodes in buffers | ||
177 | * which are multiples of the blocks size. | ||
178 | */ | ||
179 | if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) { | ||
180 | blks_per_cluster = 1; | ||
181 | nbufs = length; | ||
182 | ninodes = mp->m_sb.sb_inopblock; | ||
183 | } else { | ||
184 | blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) / | ||
185 | mp->m_sb.sb_blocksize; | ||
186 | nbufs = length / blks_per_cluster; | ||
187 | ninodes = blks_per_cluster * mp->m_sb.sb_inopblock; | ||
188 | } | ||
189 | |||
190 | /* | ||
191 | * Figure out what version number to use in the inodes we create. | ||
192 | * If the superblock version has caught up to the one that supports | ||
193 | * the new inode format, then use the new inode version. Otherwise | ||
194 | * use the old version so that old kernels will continue to be | ||
195 | * able to use the file system. | ||
196 | */ | ||
197 | if (xfs_sb_version_hasnlink(&mp->m_sb)) | ||
198 | version = 2; | ||
199 | else | ||
200 | version = 1; | ||
201 | |||
202 | for (j = 0; j < nbufs; j++) { | ||
203 | /* | ||
204 | * Get the block. | ||
205 | */ | ||
206 | d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster)); | ||
207 | fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | ||
208 | mp->m_bsize * blks_per_cluster, | ||
209 | XFS_BUF_LOCK); | ||
210 | ASSERT(fbuf); | ||
211 | ASSERT(!XFS_BUF_GETERROR(fbuf)); | ||
212 | |||
213 | /* | ||
214 | * Initialize all inodes in this buffer and then log them. | ||
215 | * | ||
216 | * XXX: It would be much better if we had just one transaction | ||
217 | * to log a whole cluster of inodes instead of all the | ||
218 | * individual transactions causing a lot of log traffic. | ||
219 | */ | ||
220 | xfs_biozero(fbuf, 0, ninodes << mp->m_sb.sb_inodelog); | ||
221 | for (i = 0; i < ninodes; i++) { | ||
222 | int ioffset = i << mp->m_sb.sb_inodelog; | ||
223 | uint isize = sizeof(struct xfs_dinode); | ||
224 | |||
225 | free = xfs_make_iptr(mp, fbuf, i); | ||
226 | free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC); | ||
227 | free->di_version = version; | ||
228 | free->di_gen = cpu_to_be32(gen); | ||
229 | free->di_next_unlinked = cpu_to_be32(NULLAGINO); | ||
230 | xfs_trans_log_buf(tp, fbuf, ioffset, ioffset + isize - 1); | ||
231 | } | ||
232 | xfs_trans_inode_alloc_buf(tp, fbuf); | ||
233 | } | ||
234 | } | ||
235 | |||
236 | /* | ||
156 | * Allocate new inodes in the allocation group specified by agbp. | 237 | * Allocate new inodes in the allocation group specified by agbp. |
157 | * Return 0 for success, else error code. | 238 | * Return 0 for success, else error code. |
158 | */ | 239 | */ |
@@ -164,24 +245,15 @@ xfs_ialloc_ag_alloc( | |||
164 | { | 245 | { |
165 | xfs_agi_t *agi; /* allocation group header */ | 246 | xfs_agi_t *agi; /* allocation group header */ |
166 | xfs_alloc_arg_t args; /* allocation argument structure */ | 247 | xfs_alloc_arg_t args; /* allocation argument structure */ |
167 | int blks_per_cluster; /* fs blocks per inode cluster */ | ||
168 | xfs_btree_cur_t *cur; /* inode btree cursor */ | 248 | xfs_btree_cur_t *cur; /* inode btree cursor */ |
169 | xfs_daddr_t d; /* disk addr of buffer */ | ||
170 | xfs_agnumber_t agno; | 249 | xfs_agnumber_t agno; |
171 | int error; | 250 | int error; |
172 | xfs_buf_t *fbuf; /* new free inodes' buffer */ | 251 | int i; |
173 | xfs_dinode_t *free; /* new free inode structure */ | ||
174 | int i; /* inode counter */ | ||
175 | int j; /* block counter */ | ||
176 | int nbufs; /* num bufs of new inodes */ | ||
177 | xfs_agino_t newino; /* new first inode's number */ | 252 | xfs_agino_t newino; /* new first inode's number */ |
178 | xfs_agino_t newlen; /* new number of inodes */ | 253 | xfs_agino_t newlen; /* new number of inodes */ |
179 | int ninodes; /* num inodes per buf */ | ||
180 | xfs_agino_t thisino; /* current inode number, for loop */ | 254 | xfs_agino_t thisino; /* current inode number, for loop */ |
181 | int version; /* inode version number to use */ | ||
182 | int isaligned = 0; /* inode allocation at stripe unit */ | 255 | int isaligned = 0; /* inode allocation at stripe unit */ |
183 | /* boundary */ | 256 | /* boundary */ |
184 | unsigned int gen; | ||
185 | 257 | ||
186 | args.tp = tp; | 258 | args.tp = tp; |
187 | args.mp = tp->t_mountp; | 259 | args.mp = tp->t_mountp; |
@@ -202,12 +274,12 @@ xfs_ialloc_ag_alloc( | |||
202 | */ | 274 | */ |
203 | agi = XFS_BUF_TO_AGI(agbp); | 275 | agi = XFS_BUF_TO_AGI(agbp); |
204 | newino = be32_to_cpu(agi->agi_newino); | 276 | newino = be32_to_cpu(agi->agi_newino); |
277 | agno = be32_to_cpu(agi->agi_seqno); | ||
205 | args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) + | 278 | args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) + |
206 | XFS_IALLOC_BLOCKS(args.mp); | 279 | XFS_IALLOC_BLOCKS(args.mp); |
207 | if (likely(newino != NULLAGINO && | 280 | if (likely(newino != NULLAGINO && |
208 | (args.agbno < be32_to_cpu(agi->agi_length)))) { | 281 | (args.agbno < be32_to_cpu(agi->agi_length)))) { |
209 | args.fsbno = XFS_AGB_TO_FSB(args.mp, | 282 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
210 | be32_to_cpu(agi->agi_seqno), args.agbno); | ||
211 | args.type = XFS_ALLOCTYPE_THIS_BNO; | 283 | args.type = XFS_ALLOCTYPE_THIS_BNO; |
212 | args.mod = args.total = args.wasdel = args.isfl = | 284 | args.mod = args.total = args.wasdel = args.isfl = |
213 | args.userdata = args.minalignslop = 0; | 285 | args.userdata = args.minalignslop = 0; |
@@ -258,8 +330,7 @@ xfs_ialloc_ag_alloc( | |||
258 | * For now, just allocate blocks up front. | 330 | * For now, just allocate blocks up front. |
259 | */ | 331 | */ |
260 | args.agbno = be32_to_cpu(agi->agi_root); | 332 | args.agbno = be32_to_cpu(agi->agi_root); |
261 | args.fsbno = XFS_AGB_TO_FSB(args.mp, | 333 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
262 | be32_to_cpu(agi->agi_seqno), args.agbno); | ||
263 | /* | 334 | /* |
264 | * Allocate a fixed-size extent of inodes. | 335 | * Allocate a fixed-size extent of inodes. |
265 | */ | 336 | */ |
@@ -282,8 +353,7 @@ xfs_ialloc_ag_alloc( | |||
282 | if (isaligned && args.fsbno == NULLFSBLOCK) { | 353 | if (isaligned && args.fsbno == NULLFSBLOCK) { |
283 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | 354 | args.type = XFS_ALLOCTYPE_NEAR_BNO; |
284 | args.agbno = be32_to_cpu(agi->agi_root); | 355 | args.agbno = be32_to_cpu(agi->agi_root); |
285 | args.fsbno = XFS_AGB_TO_FSB(args.mp, | 356 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
286 | be32_to_cpu(agi->agi_seqno), args.agbno); | ||
287 | args.alignment = xfs_ialloc_cluster_alignment(&args); | 357 | args.alignment = xfs_ialloc_cluster_alignment(&args); |
288 | if ((error = xfs_alloc_vextent(&args))) | 358 | if ((error = xfs_alloc_vextent(&args))) |
289 | return error; | 359 | return error; |
@@ -294,85 +364,30 @@ xfs_ialloc_ag_alloc( | |||
294 | return 0; | 364 | return 0; |
295 | } | 365 | } |
296 | ASSERT(args.len == args.minlen); | 366 | ASSERT(args.len == args.minlen); |
297 | /* | ||
298 | * Convert the results. | ||
299 | */ | ||
300 | newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0); | ||
301 | /* | ||
302 | * Loop over the new block(s), filling in the inodes. | ||
303 | * For small block sizes, manipulate the inodes in buffers | ||
304 | * which are multiples of the blocks size. | ||
305 | */ | ||
306 | if (args.mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(args.mp)) { | ||
307 | blks_per_cluster = 1; | ||
308 | nbufs = (int)args.len; | ||
309 | ninodes = args.mp->m_sb.sb_inopblock; | ||
310 | } else { | ||
311 | blks_per_cluster = XFS_INODE_CLUSTER_SIZE(args.mp) / | ||
312 | args.mp->m_sb.sb_blocksize; | ||
313 | nbufs = (int)args.len / blks_per_cluster; | ||
314 | ninodes = blks_per_cluster * args.mp->m_sb.sb_inopblock; | ||
315 | } | ||
316 | /* | ||
317 | * Figure out what version number to use in the inodes we create. | ||
318 | * If the superblock version has caught up to the one that supports | ||
319 | * the new inode format, then use the new inode version. Otherwise | ||
320 | * use the old version so that old kernels will continue to be | ||
321 | * able to use the file system. | ||
322 | */ | ||
323 | if (xfs_sb_version_hasnlink(&args.mp->m_sb)) | ||
324 | version = 2; | ||
325 | else | ||
326 | version = 1; | ||
327 | 367 | ||
328 | /* | 368 | /* |
369 | * Stamp and write the inode buffers. | ||
370 | * | ||
329 | * Seed the new inode cluster with a random generation number. This | 371 | * Seed the new inode cluster with a random generation number. This |
330 | * prevents short-term reuse of generation numbers if a chunk is | 372 | * prevents short-term reuse of generation numbers if a chunk is |
331 | * freed and then immediately reallocated. We use random numbers | 373 | * freed and then immediately reallocated. We use random numbers |
332 | * rather than a linear progression to prevent the next generation | 374 | * rather than a linear progression to prevent the next generation |
333 | * number from being easily guessable. | 375 | * number from being easily guessable. |
334 | */ | 376 | */ |
335 | gen = random32(); | 377 | xfs_ialloc_inode_init(args.mp, tp, agno, args.agbno, args.len, |
336 | for (j = 0; j < nbufs; j++) { | 378 | random32()); |
337 | /* | ||
338 | * Get the block. | ||
339 | */ | ||
340 | d = XFS_AGB_TO_DADDR(args.mp, be32_to_cpu(agi->agi_seqno), | ||
341 | args.agbno + (j * blks_per_cluster)); | ||
342 | fbuf = xfs_trans_get_buf(tp, args.mp->m_ddev_targp, d, | ||
343 | args.mp->m_bsize * blks_per_cluster, | ||
344 | XFS_BUF_LOCK); | ||
345 | ASSERT(fbuf); | ||
346 | ASSERT(!XFS_BUF_GETERROR(fbuf)); | ||
347 | 379 | ||
348 | /* | 380 | /* |
349 | * Initialize all inodes in this buffer and then log them. | 381 | * Convert the results. |
350 | * | 382 | */ |
351 | * XXX: It would be much better if we had just one transaction to | 383 | newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0); |
352 | * log a whole cluster of inodes instead of all the individual | ||
353 | * transactions causing a lot of log traffic. | ||
354 | */ | ||
355 | xfs_biozero(fbuf, 0, ninodes << args.mp->m_sb.sb_inodelog); | ||
356 | for (i = 0; i < ninodes; i++) { | ||
357 | int ioffset = i << args.mp->m_sb.sb_inodelog; | ||
358 | uint isize = sizeof(struct xfs_dinode); | ||
359 | |||
360 | free = xfs_make_iptr(args.mp, fbuf, i); | ||
361 | free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC); | ||
362 | free->di_version = version; | ||
363 | free->di_gen = cpu_to_be32(gen); | ||
364 | free->di_next_unlinked = cpu_to_be32(NULLAGINO); | ||
365 | xfs_trans_log_buf(tp, fbuf, ioffset, ioffset + isize - 1); | ||
366 | } | ||
367 | xfs_trans_inode_alloc_buf(tp, fbuf); | ||
368 | } | ||
369 | be32_add_cpu(&agi->agi_count, newlen); | 384 | be32_add_cpu(&agi->agi_count, newlen); |
370 | be32_add_cpu(&agi->agi_freecount, newlen); | 385 | be32_add_cpu(&agi->agi_freecount, newlen); |
371 | agno = be32_to_cpu(agi->agi_seqno); | ||
372 | down_read(&args.mp->m_peraglock); | 386 | down_read(&args.mp->m_peraglock); |
373 | args.mp->m_perag[agno].pagi_freecount += newlen; | 387 | args.mp->m_perag[agno].pagi_freecount += newlen; |
374 | up_read(&args.mp->m_peraglock); | 388 | up_read(&args.mp->m_peraglock); |
375 | agi->agi_newino = cpu_to_be32(newino); | 389 | agi->agi_newino = cpu_to_be32(newino); |
390 | |||
376 | /* | 391 | /* |
377 | * Insert records describing the new inode chunk into the btree. | 392 | * Insert records describing the new inode chunk into the btree. |
378 | */ | 393 | */ |