Commit | Line | Data |
---|---|---|
d7e09d03 PT |
1 | /* |
2 | * GPL HEADER START | |
3 | * | |
4 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License version 2 only, | |
8 | * as published by the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, but | |
11 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
13 | * General Public License version 2 for more details (a copy is included | |
14 | * in the LICENSE file that accompanied this code). | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * version 2 along with this program; If not, see | |
18 | * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf | |
19 | * | |
20 | * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
21 | * CA 95054 USA or visit www.sun.com if you need additional information or | |
22 | * have any questions. | |
23 | * | |
24 | * GPL HEADER END | |
25 | */ | |
26 | /* | |
27 | * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved. | |
28 | * Use is subject to license terms. | |
29 | * | |
30 | * Copyright (c) 2011, 2012, Intel Corporation. | |
31 | */ | |
32 | /* | |
33 | * This file is part of Lustre, http://www.lustre.org/ | |
34 | * Lustre is a trademark of Sun Microsystems, Inc. | |
35 | * | |
36 | * cl code shared between vvp and liblustre (and other Lustre clients in the | |
37 | * future). | |
38 | * | |
39 | * Author: Nikita Danilov <nikita.danilov@sun.com> | |
40 | */ | |
41 | ||
42 | #define DEBUG_SUBSYSTEM S_LLITE | |
43 | ||
44 | # include <linux/libcfs/libcfs.h> | |
45 | # include <linux/fs.h> | |
46 | # include <linux/sched.h> | |
47 | # include <linux/mm.h> | |
48 | # include <linux/quotaops.h> | |
49 | # include <linux/highmem.h> | |
50 | # include <linux/pagemap.h> | |
51 | # include <linux/rbtree.h> | |
52 | ||
53 | #include <obd.h> | |
54 | #include <obd_support.h> | |
55 | #include <lustre_fid.h> | |
56 | #include <lustre_lite.h> | |
57 | #include <lustre_dlm.h> | |
58 | #include <lustre_ver.h> | |
59 | #include <lustre_mdc.h> | |
60 | #include <cl_object.h> | |
61 | ||
62 | #include <lclient.h> | |
63 | ||
64 | #include "../llite/llite_internal.h" | |
65 | ||
66 | const struct cl_req_operations ccc_req_ops; | |
67 | ||
68 | /* | |
69 | * ccc_ prefix stands for "Common Client Code". | |
70 | */ | |
71 | ||
72 | static struct kmem_cache *ccc_lock_kmem; | |
73 | static struct kmem_cache *ccc_object_kmem; | |
74 | static struct kmem_cache *ccc_thread_kmem; | |
75 | static struct kmem_cache *ccc_session_kmem; | |
76 | static struct kmem_cache *ccc_req_kmem; | |
77 | ||
78 | static struct lu_kmem_descr ccc_caches[] = { | |
79 | { | |
80 | .ckd_cache = &ccc_lock_kmem, | |
81 | .ckd_name = "ccc_lock_kmem", | |
82 | .ckd_size = sizeof (struct ccc_lock) | |
83 | }, | |
84 | { | |
85 | .ckd_cache = &ccc_object_kmem, | |
86 | .ckd_name = "ccc_object_kmem", | |
87 | .ckd_size = sizeof (struct ccc_object) | |
88 | }, | |
89 | { | |
90 | .ckd_cache = &ccc_thread_kmem, | |
91 | .ckd_name = "ccc_thread_kmem", | |
92 | .ckd_size = sizeof (struct ccc_thread_info), | |
93 | }, | |
94 | { | |
95 | .ckd_cache = &ccc_session_kmem, | |
96 | .ckd_name = "ccc_session_kmem", | |
97 | .ckd_size = sizeof (struct ccc_session) | |
98 | }, | |
99 | { | |
100 | .ckd_cache = &ccc_req_kmem, | |
101 | .ckd_name = "ccc_req_kmem", | |
102 | .ckd_size = sizeof (struct ccc_req) | |
103 | }, | |
104 | { | |
105 | .ckd_cache = NULL | |
106 | } | |
107 | }; | |
108 | ||
109 | /***************************************************************************** | |
110 | * | |
111 | * Vvp device and device type functions. | |
112 | * | |
113 | */ | |
114 | ||
115 | void *ccc_key_init(const struct lu_context *ctx, | |
116 | struct lu_context_key *key) | |
117 | { | |
118 | struct ccc_thread_info *info; | |
119 | ||
120 | OBD_SLAB_ALLOC_PTR_GFP(info, ccc_thread_kmem, __GFP_IO); | |
121 | if (info == NULL) | |
122 | info = ERR_PTR(-ENOMEM); | |
123 | return info; | |
124 | } | |
125 | ||
126 | void ccc_key_fini(const struct lu_context *ctx, | |
127 | struct lu_context_key *key, void *data) | |
128 | { | |
129 | struct ccc_thread_info *info = data; | |
130 | OBD_SLAB_FREE_PTR(info, ccc_thread_kmem); | |
131 | } | |
132 | ||
133 | void *ccc_session_key_init(const struct lu_context *ctx, | |
134 | struct lu_context_key *key) | |
135 | { | |
136 | struct ccc_session *session; | |
137 | ||
138 | OBD_SLAB_ALLOC_PTR_GFP(session, ccc_session_kmem, __GFP_IO); | |
139 | if (session == NULL) | |
140 | session = ERR_PTR(-ENOMEM); | |
141 | return session; | |
142 | } | |
143 | ||
144 | void ccc_session_key_fini(const struct lu_context *ctx, | |
145 | struct lu_context_key *key, void *data) | |
146 | { | |
147 | struct ccc_session *session = data; | |
148 | OBD_SLAB_FREE_PTR(session, ccc_session_kmem); | |
149 | } | |
150 | ||
151 | struct lu_context_key ccc_key = { | |
152 | .lct_tags = LCT_CL_THREAD, | |
153 | .lct_init = ccc_key_init, | |
154 | .lct_fini = ccc_key_fini | |
155 | }; | |
156 | ||
157 | struct lu_context_key ccc_session_key = { | |
158 | .lct_tags = LCT_SESSION, | |
159 | .lct_init = ccc_session_key_init, | |
160 | .lct_fini = ccc_session_key_fini | |
161 | }; | |
162 | ||
163 | ||
164 | /* type constructor/destructor: ccc_type_{init,fini,start,stop}(). */ | |
165 | // LU_TYPE_INIT_FINI(ccc, &ccc_key, &ccc_session_key); | |
166 | ||
167 | int ccc_device_init(const struct lu_env *env, struct lu_device *d, | |
168 | const char *name, struct lu_device *next) | |
169 | { | |
170 | struct ccc_device *vdv; | |
171 | int rc; | |
172 | ENTRY; | |
173 | ||
174 | vdv = lu2ccc_dev(d); | |
175 | vdv->cdv_next = lu2cl_dev(next); | |
176 | ||
177 | LASSERT(d->ld_site != NULL && next->ld_type != NULL); | |
178 | next->ld_site = d->ld_site; | |
179 | rc = next->ld_type->ldt_ops->ldto_device_init( | |
180 | env, next, next->ld_type->ldt_name, NULL); | |
181 | if (rc == 0) { | |
182 | lu_device_get(next); | |
183 | lu_ref_add(&next->ld_reference, "lu-stack", &lu_site_init); | |
184 | } | |
185 | RETURN(rc); | |
186 | } | |
187 | ||
188 | struct lu_device *ccc_device_fini(const struct lu_env *env, | |
189 | struct lu_device *d) | |
190 | { | |
191 | return cl2lu_dev(lu2ccc_dev(d)->cdv_next); | |
192 | } | |
193 | ||
194 | struct lu_device *ccc_device_alloc(const struct lu_env *env, | |
195 | struct lu_device_type *t, | |
196 | struct lustre_cfg *cfg, | |
197 | const struct lu_device_operations *luops, | |
198 | const struct cl_device_operations *clops) | |
199 | { | |
200 | struct ccc_device *vdv; | |
201 | struct lu_device *lud; | |
202 | struct cl_site *site; | |
203 | int rc; | |
204 | ENTRY; | |
205 | ||
206 | OBD_ALLOC_PTR(vdv); | |
207 | if (vdv == NULL) | |
208 | RETURN(ERR_PTR(-ENOMEM)); | |
209 | ||
210 | lud = &vdv->cdv_cl.cd_lu_dev; | |
211 | cl_device_init(&vdv->cdv_cl, t); | |
212 | ccc2lu_dev(vdv)->ld_ops = luops; | |
213 | vdv->cdv_cl.cd_ops = clops; | |
214 | ||
215 | OBD_ALLOC_PTR(site); | |
216 | if (site != NULL) { | |
217 | rc = cl_site_init(site, &vdv->cdv_cl); | |
218 | if (rc == 0) | |
219 | rc = lu_site_init_finish(&site->cs_lu); | |
220 | else { | |
221 | LASSERT(lud->ld_site == NULL); | |
222 | CERROR("Cannot init lu_site, rc %d.\n", rc); | |
223 | OBD_FREE_PTR(site); | |
224 | } | |
225 | } else | |
226 | rc = -ENOMEM; | |
227 | if (rc != 0) { | |
228 | ccc_device_free(env, lud); | |
229 | lud = ERR_PTR(rc); | |
230 | } | |
231 | RETURN(lud); | |
232 | } | |
233 | ||
234 | struct lu_device *ccc_device_free(const struct lu_env *env, | |
235 | struct lu_device *d) | |
236 | { | |
237 | struct ccc_device *vdv = lu2ccc_dev(d); | |
238 | struct cl_site *site = lu2cl_site(d->ld_site); | |
239 | struct lu_device *next = cl2lu_dev(vdv->cdv_next); | |
240 | ||
241 | if (d->ld_site != NULL) { | |
242 | cl_site_fini(site); | |
243 | OBD_FREE_PTR(site); | |
244 | } | |
245 | cl_device_fini(lu2cl_dev(d)); | |
246 | OBD_FREE_PTR(vdv); | |
247 | return next; | |
248 | } | |
249 | ||
250 | int ccc_req_init(const struct lu_env *env, struct cl_device *dev, | |
251 | struct cl_req *req) | |
252 | { | |
253 | struct ccc_req *vrq; | |
254 | int result; | |
255 | ||
256 | OBD_SLAB_ALLOC_PTR_GFP(vrq, ccc_req_kmem, __GFP_IO); | |
257 | if (vrq != NULL) { | |
258 | cl_req_slice_add(req, &vrq->crq_cl, dev, &ccc_req_ops); | |
259 | result = 0; | |
260 | } else | |
261 | result = -ENOMEM; | |
262 | return result; | |
263 | } | |
264 | ||
265 | /** | |
266 | * An `emergency' environment used by ccc_inode_fini() when cl_env_get() | |
267 | * fails. Access to this environment is serialized by ccc_inode_fini_guard | |
268 | * mutex. | |
269 | */ | |
270 | static struct lu_env *ccc_inode_fini_env = NULL; | |
271 | ||
272 | /** | |
273 | * A mutex serializing calls to slp_inode_fini() under extreme memory | |
274 | * pressure, when environments cannot be allocated. | |
275 | */ | |
276 | static DEFINE_MUTEX(ccc_inode_fini_guard); | |
277 | static int dummy_refcheck; | |
278 | ||
279 | int ccc_global_init(struct lu_device_type *device_type) | |
280 | { | |
281 | int result; | |
282 | ||
283 | result = lu_kmem_init(ccc_caches); | |
284 | if (result) | |
285 | return result; | |
286 | ||
287 | result = lu_device_type_init(device_type); | |
288 | if (result) | |
289 | goto out_kmem; | |
290 | ||
291 | ccc_inode_fini_env = cl_env_alloc(&dummy_refcheck, | |
292 | LCT_REMEMBER|LCT_NOREF); | |
293 | if (IS_ERR(ccc_inode_fini_env)) { | |
294 | result = PTR_ERR(ccc_inode_fini_env); | |
295 | goto out_device; | |
296 | } | |
297 | ||
298 | ccc_inode_fini_env->le_ctx.lc_cookie = 0x4; | |
299 | return 0; | |
300 | out_device: | |
301 | lu_device_type_fini(device_type); | |
302 | out_kmem: | |
303 | lu_kmem_fini(ccc_caches); | |
304 | return result; | |
305 | } | |
306 | ||
307 | void ccc_global_fini(struct lu_device_type *device_type) | |
308 | { | |
309 | if (ccc_inode_fini_env != NULL) { | |
310 | cl_env_put(ccc_inode_fini_env, &dummy_refcheck); | |
311 | ccc_inode_fini_env = NULL; | |
312 | } | |
313 | lu_device_type_fini(device_type); | |
314 | lu_kmem_fini(ccc_caches); | |
315 | } | |
316 | ||
317 | /***************************************************************************** | |
318 | * | |
319 | * Object operations. | |
320 | * | |
321 | */ | |
322 | ||
323 | struct lu_object *ccc_object_alloc(const struct lu_env *env, | |
324 | const struct lu_object_header *unused, | |
325 | struct lu_device *dev, | |
326 | const struct cl_object_operations *clops, | |
327 | const struct lu_object_operations *luops) | |
328 | { | |
329 | struct ccc_object *vob; | |
330 | struct lu_object *obj; | |
331 | ||
332 | OBD_SLAB_ALLOC_PTR_GFP(vob, ccc_object_kmem, __GFP_IO); | |
333 | if (vob != NULL) { | |
334 | struct cl_object_header *hdr; | |
335 | ||
336 | obj = ccc2lu(vob); | |
337 | hdr = &vob->cob_header; | |
338 | cl_object_header_init(hdr); | |
339 | lu_object_init(obj, &hdr->coh_lu, dev); | |
340 | lu_object_add_top(&hdr->coh_lu, obj); | |
341 | ||
342 | vob->cob_cl.co_ops = clops; | |
343 | obj->lo_ops = luops; | |
344 | } else | |
345 | obj = NULL; | |
346 | return obj; | |
347 | } | |
348 | ||
349 | int ccc_object_init0(const struct lu_env *env, | |
350 | struct ccc_object *vob, | |
351 | const struct cl_object_conf *conf) | |
352 | { | |
353 | vob->cob_inode = conf->coc_inode; | |
354 | vob->cob_transient_pages = 0; | |
355 | cl_object_page_init(&vob->cob_cl, sizeof(struct ccc_page)); | |
356 | return 0; | |
357 | } | |
358 | ||
359 | int ccc_object_init(const struct lu_env *env, struct lu_object *obj, | |
360 | const struct lu_object_conf *conf) | |
361 | { | |
362 | struct ccc_device *dev = lu2ccc_dev(obj->lo_dev); | |
363 | struct ccc_object *vob = lu2ccc(obj); | |
364 | struct lu_object *below; | |
365 | struct lu_device *under; | |
366 | int result; | |
367 | ||
368 | under = &dev->cdv_next->cd_lu_dev; | |
369 | below = under->ld_ops->ldo_object_alloc(env, obj->lo_header, under); | |
370 | if (below != NULL) { | |
371 | const struct cl_object_conf *cconf; | |
372 | ||
373 | cconf = lu2cl_conf(conf); | |
374 | INIT_LIST_HEAD(&vob->cob_pending_list); | |
375 | lu_object_add(obj, below); | |
376 | result = ccc_object_init0(env, vob, cconf); | |
377 | } else | |
378 | result = -ENOMEM; | |
379 | return result; | |
380 | } | |
381 | ||
382 | void ccc_object_free(const struct lu_env *env, struct lu_object *obj) | |
383 | { | |
384 | struct ccc_object *vob = lu2ccc(obj); | |
385 | ||
386 | lu_object_fini(obj); | |
387 | lu_object_header_fini(obj->lo_header); | |
388 | OBD_SLAB_FREE_PTR(vob, ccc_object_kmem); | |
389 | } | |
390 | ||
391 | int ccc_lock_init(const struct lu_env *env, | |
392 | struct cl_object *obj, struct cl_lock *lock, | |
393 | const struct cl_io *unused, | |
394 | const struct cl_lock_operations *lkops) | |
395 | { | |
396 | struct ccc_lock *clk; | |
397 | int result; | |
398 | ||
399 | CLOBINVRNT(env, obj, ccc_object_invariant(obj)); | |
400 | ||
401 | OBD_SLAB_ALLOC_PTR_GFP(clk, ccc_lock_kmem, __GFP_IO); | |
402 | if (clk != NULL) { | |
403 | cl_lock_slice_add(lock, &clk->clk_cl, obj, lkops); | |
404 | result = 0; | |
405 | } else | |
406 | result = -ENOMEM; | |
407 | return result; | |
408 | } | |
409 | ||
410 | int ccc_attr_set(const struct lu_env *env, struct cl_object *obj, | |
411 | const struct cl_attr *attr, unsigned valid) | |
412 | { | |
413 | return 0; | |
414 | } | |
415 | ||
416 | int ccc_object_glimpse(const struct lu_env *env, | |
417 | const struct cl_object *obj, struct ost_lvb *lvb) | |
418 | { | |
419 | struct inode *inode = ccc_object_inode(obj); | |
420 | ||
421 | ENTRY; | |
422 | lvb->lvb_mtime = cl_inode_mtime(inode); | |
423 | lvb->lvb_atime = cl_inode_atime(inode); | |
424 | lvb->lvb_ctime = cl_inode_ctime(inode); | |
425 | /* | |
426 | * LU-417: Add dirty pages block count lest i_blocks reports 0, some | |
427 | * "cp" or "tar" on remote node may think it's a completely sparse file | |
428 | * and skip it. | |
429 | */ | |
430 | if (lvb->lvb_size > 0 && lvb->lvb_blocks == 0) | |
431 | lvb->lvb_blocks = dirty_cnt(inode); | |
432 | RETURN(0); | |
433 | } | |
434 | ||
435 | ||
436 | ||
437 | int ccc_conf_set(const struct lu_env *env, struct cl_object *obj, | |
438 | const struct cl_object_conf *conf) | |
439 | { | |
440 | /* TODO: destroy all pages attached to this object. */ | |
441 | return 0; | |
442 | } | |
443 | ||
444 | static void ccc_object_size_lock(struct cl_object *obj) | |
445 | { | |
446 | struct inode *inode = ccc_object_inode(obj); | |
447 | ||
448 | cl_isize_lock(inode); | |
449 | cl_object_attr_lock(obj); | |
450 | } | |
451 | ||
452 | static void ccc_object_size_unlock(struct cl_object *obj) | |
453 | { | |
454 | struct inode *inode = ccc_object_inode(obj); | |
455 | ||
456 | cl_object_attr_unlock(obj); | |
457 | cl_isize_unlock(inode); | |
458 | } | |
459 | ||
460 | /***************************************************************************** | |
461 | * | |
462 | * Page operations. | |
463 | * | |
464 | */ | |
465 | ||
466 | struct page *ccc_page_vmpage(const struct lu_env *env, | |
467 | const struct cl_page_slice *slice) | |
468 | { | |
469 | return cl2vm_page(slice); | |
470 | } | |
471 | ||
472 | int ccc_page_is_under_lock(const struct lu_env *env, | |
473 | const struct cl_page_slice *slice, | |
474 | struct cl_io *io) | |
475 | { | |
476 | struct ccc_io *cio = ccc_env_io(env); | |
477 | struct cl_lock_descr *desc = &ccc_env_info(env)->cti_descr; | |
478 | struct cl_page *page = slice->cpl_page; | |
479 | ||
480 | int result; | |
481 | ||
482 | ENTRY; | |
483 | ||
484 | if (io->ci_type == CIT_READ || io->ci_type == CIT_WRITE || | |
485 | io->ci_type == CIT_FAULT) { | |
486 | if (cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED) | |
487 | result = -EBUSY; | |
488 | else { | |
489 | desc->cld_start = page->cp_index; | |
490 | desc->cld_end = page->cp_index; | |
491 | desc->cld_obj = page->cp_obj; | |
492 | desc->cld_mode = CLM_READ; | |
493 | result = cl_queue_match(&io->ci_lockset.cls_done, | |
494 | desc) ? -EBUSY : 0; | |
495 | } | |
496 | } else | |
497 | result = 0; | |
498 | RETURN(result); | |
499 | } | |
500 | ||
501 | int ccc_fail(const struct lu_env *env, const struct cl_page_slice *slice) | |
502 | { | |
503 | /* | |
504 | * Cached read? | |
505 | */ | |
506 | LBUG(); | |
507 | return 0; | |
508 | } | |
509 | ||
510 | void ccc_transient_page_verify(const struct cl_page *page) | |
511 | { | |
512 | } | |
513 | ||
514 | int ccc_transient_page_own(const struct lu_env *env, | |
515 | const struct cl_page_slice *slice, | |
516 | struct cl_io *unused, | |
517 | int nonblock) | |
518 | { | |
519 | ccc_transient_page_verify(slice->cpl_page); | |
520 | return 0; | |
521 | } | |
522 | ||
523 | void ccc_transient_page_assume(const struct lu_env *env, | |
524 | const struct cl_page_slice *slice, | |
525 | struct cl_io *unused) | |
526 | { | |
527 | ccc_transient_page_verify(slice->cpl_page); | |
528 | } | |
529 | ||
530 | void ccc_transient_page_unassume(const struct lu_env *env, | |
531 | const struct cl_page_slice *slice, | |
532 | struct cl_io *unused) | |
533 | { | |
534 | ccc_transient_page_verify(slice->cpl_page); | |
535 | } | |
536 | ||
537 | void ccc_transient_page_disown(const struct lu_env *env, | |
538 | const struct cl_page_slice *slice, | |
539 | struct cl_io *unused) | |
540 | { | |
541 | ccc_transient_page_verify(slice->cpl_page); | |
542 | } | |
543 | ||
544 | void ccc_transient_page_discard(const struct lu_env *env, | |
545 | const struct cl_page_slice *slice, | |
546 | struct cl_io *unused) | |
547 | { | |
548 | struct cl_page *page = slice->cpl_page; | |
549 | ||
550 | ccc_transient_page_verify(slice->cpl_page); | |
551 | ||
552 | /* | |
553 | * For transient pages, remove it from the radix tree. | |
554 | */ | |
555 | cl_page_delete(env, page); | |
556 | } | |
557 | ||
558 | int ccc_transient_page_prep(const struct lu_env *env, | |
559 | const struct cl_page_slice *slice, | |
560 | struct cl_io *unused) | |
561 | { | |
562 | ENTRY; | |
563 | /* transient page should always be sent. */ | |
564 | RETURN(0); | |
565 | } | |
566 | ||
567 | /***************************************************************************** | |
568 | * | |
569 | * Lock operations. | |
570 | * | |
571 | */ | |
572 | ||
573 | void ccc_lock_delete(const struct lu_env *env, | |
574 | const struct cl_lock_slice *slice) | |
575 | { | |
576 | CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj)); | |
577 | } | |
578 | ||
579 | void ccc_lock_fini(const struct lu_env *env, struct cl_lock_slice *slice) | |
580 | { | |
581 | struct ccc_lock *clk = cl2ccc_lock(slice); | |
582 | OBD_SLAB_FREE_PTR(clk, ccc_lock_kmem); | |
583 | } | |
584 | ||
585 | int ccc_lock_enqueue(const struct lu_env *env, | |
586 | const struct cl_lock_slice *slice, | |
587 | struct cl_io *unused, __u32 enqflags) | |
588 | { | |
589 | CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj)); | |
590 | return 0; | |
591 | } | |
592 | ||
593 | int ccc_lock_unuse(const struct lu_env *env, const struct cl_lock_slice *slice) | |
594 | { | |
595 | CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj)); | |
596 | return 0; | |
597 | } | |
598 | ||
599 | int ccc_lock_wait(const struct lu_env *env, const struct cl_lock_slice *slice) | |
600 | { | |
601 | CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj)); | |
602 | return 0; | |
603 | } | |
604 | ||
605 | /** | |
606 | * Implementation of cl_lock_operations::clo_fits_into() methods for ccc | |
607 | * layer. This function is executed every time io finds an existing lock in | |
608 | * the lock cache while creating new lock. This function has to decide whether | |
609 | * cached lock "fits" into io. | |
610 | * | |
611 | * \param slice lock to be checked | |
612 | * \param io IO that wants a lock. | |
613 | * | |
614 | * \see lov_lock_fits_into(). | |
615 | */ | |
616 | int ccc_lock_fits_into(const struct lu_env *env, | |
617 | const struct cl_lock_slice *slice, | |
618 | const struct cl_lock_descr *need, | |
619 | const struct cl_io *io) | |
620 | { | |
621 | const struct cl_lock *lock = slice->cls_lock; | |
622 | const struct cl_lock_descr *descr = &lock->cll_descr; | |
623 | const struct ccc_io *cio = ccc_env_io(env); | |
624 | int result; | |
625 | ||
626 | ENTRY; | |
627 | /* | |
628 | * Work around DLM peculiarity: it assumes that glimpse | |
629 | * (LDLM_FL_HAS_INTENT) lock is always LCK_PR, and returns reads lock | |
630 | * when asked for LCK_PW lock with LDLM_FL_HAS_INTENT flag set. Make | |
631 | * sure that glimpse doesn't get CLM_WRITE top-lock, so that it | |
632 | * doesn't enqueue CLM_WRITE sub-locks. | |
633 | */ | |
634 | if (cio->cui_glimpse) | |
635 | result = descr->cld_mode != CLM_WRITE; | |
636 | ||
637 | /* | |
638 | * Also, don't match incomplete write locks for read, otherwise read | |
639 | * would enqueue missing sub-locks in the write mode. | |
640 | */ | |
641 | else if (need->cld_mode != descr->cld_mode) | |
642 | result = lock->cll_state >= CLS_ENQUEUED; | |
643 | else | |
644 | result = 1; | |
645 | RETURN(result); | |
646 | } | |
647 | ||
648 | /** | |
649 | * Implements cl_lock_operations::clo_state() method for ccc layer, invoked | |
650 | * whenever lock state changes. Transfers object attributes, that might be | |
651 | * updated as a result of lock acquiring into inode. | |
652 | */ | |
653 | void ccc_lock_state(const struct lu_env *env, | |
654 | const struct cl_lock_slice *slice, | |
655 | enum cl_lock_state state) | |
656 | { | |
657 | struct cl_lock *lock = slice->cls_lock; | |
658 | ENTRY; | |
659 | ||
660 | /* | |
661 | * Refresh inode attributes when the lock is moving into CLS_HELD | |
662 | * state, and only when this is a result of real enqueue, rather than | |
663 | * of finding lock in the cache. | |
664 | */ | |
665 | if (state == CLS_HELD && lock->cll_state < CLS_HELD) { | |
666 | struct cl_object *obj; | |
667 | struct inode *inode; | |
668 | ||
669 | obj = slice->cls_obj; | |
670 | inode = ccc_object_inode(obj); | |
671 | ||
672 | /* vmtruncate() sets the i_size | |
673 | * under both a DLM lock and the | |
674 | * ll_inode_size_lock(). If we don't get the | |
675 | * ll_inode_size_lock() here we can match the DLM lock and | |
676 | * reset i_size. generic_file_write can then trust the | |
677 | * stale i_size when doing appending writes and effectively | |
678 | * cancel the result of the truncate. Getting the | |
679 | * ll_inode_size_lock() after the enqueue maintains the DLM | |
680 | * -> ll_inode_size_lock() acquiring order. */ | |
681 | if (lock->cll_descr.cld_start == 0 && | |
682 | lock->cll_descr.cld_end == CL_PAGE_EOF) | |
683 | cl_merge_lvb(env, inode); | |
684 | } | |
685 | EXIT; | |
686 | } | |
687 | ||
688 | /***************************************************************************** | |
689 | * | |
690 | * io operations. | |
691 | * | |
692 | */ | |
693 | ||
694 | void ccc_io_fini(const struct lu_env *env, const struct cl_io_slice *ios) | |
695 | { | |
696 | struct cl_io *io = ios->cis_io; | |
697 | ||
698 | CLOBINVRNT(env, io->ci_obj, ccc_object_invariant(io->ci_obj)); | |
699 | } | |
700 | ||
701 | int ccc_io_one_lock_index(const struct lu_env *env, struct cl_io *io, | |
702 | __u32 enqflags, enum cl_lock_mode mode, | |
703 | pgoff_t start, pgoff_t end) | |
704 | { | |
705 | struct ccc_io *cio = ccc_env_io(env); | |
706 | struct cl_lock_descr *descr = &cio->cui_link.cill_descr; | |
707 | struct cl_object *obj = io->ci_obj; | |
708 | ||
709 | CLOBINVRNT(env, obj, ccc_object_invariant(obj)); | |
710 | ENTRY; | |
711 | ||
712 | CDEBUG(D_VFSTRACE, "lock: %d [%lu, %lu]\n", mode, start, end); | |
713 | ||
714 | memset(&cio->cui_link, 0, sizeof cio->cui_link); | |
715 | ||
716 | if (cio->cui_fd && (cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) { | |
717 | descr->cld_mode = CLM_GROUP; | |
718 | descr->cld_gid = cio->cui_fd->fd_grouplock.cg_gid; | |
719 | } else { | |
720 | descr->cld_mode = mode; | |
721 | } | |
722 | descr->cld_obj = obj; | |
723 | descr->cld_start = start; | |
724 | descr->cld_end = end; | |
725 | descr->cld_enq_flags = enqflags; | |
726 | ||
727 | cl_io_lock_add(env, io, &cio->cui_link); | |
728 | RETURN(0); | |
729 | } | |
730 | ||
731 | void ccc_io_update_iov(const struct lu_env *env, | |
732 | struct ccc_io *cio, struct cl_io *io) | |
733 | { | |
734 | int i; | |
735 | size_t size = io->u.ci_rw.crw_count; | |
736 | ||
737 | cio->cui_iov_olen = 0; | |
738 | if (!cl_is_normalio(env, io) || cio->cui_tot_nrsegs == 0) | |
739 | return; | |
740 | ||
741 | for (i = 0; i < cio->cui_tot_nrsegs; i++) { | |
742 | struct iovec *iv = &cio->cui_iov[i]; | |
743 | ||
744 | if (iv->iov_len < size) | |
745 | size -= iv->iov_len; | |
746 | else { | |
747 | if (iv->iov_len > size) { | |
748 | cio->cui_iov_olen = iv->iov_len; | |
749 | iv->iov_len = size; | |
750 | } | |
751 | break; | |
752 | } | |
753 | } | |
754 | ||
755 | cio->cui_nrsegs = i + 1; | |
756 | LASSERTF(cio->cui_tot_nrsegs >= cio->cui_nrsegs, | |
757 | "tot_nrsegs: %lu, nrsegs: %lu\n", | |
758 | cio->cui_tot_nrsegs, cio->cui_nrsegs); | |
759 | } | |
760 | ||
761 | int ccc_io_one_lock(const struct lu_env *env, struct cl_io *io, | |
762 | __u32 enqflags, enum cl_lock_mode mode, | |
763 | loff_t start, loff_t end) | |
764 | { | |
765 | struct cl_object *obj = io->ci_obj; | |
766 | return ccc_io_one_lock_index(env, io, enqflags, mode, | |
767 | cl_index(obj, start), cl_index(obj, end)); | |
768 | } | |
769 | ||
770 | void ccc_io_end(const struct lu_env *env, const struct cl_io_slice *ios) | |
771 | { | |
772 | CLOBINVRNT(env, ios->cis_io->ci_obj, | |
773 | ccc_object_invariant(ios->cis_io->ci_obj)); | |
774 | } | |
775 | ||
776 | void ccc_io_advance(const struct lu_env *env, | |
777 | const struct cl_io_slice *ios, | |
778 | size_t nob) | |
779 | { | |
780 | struct ccc_io *cio = cl2ccc_io(env, ios); | |
781 | struct cl_io *io = ios->cis_io; | |
782 | struct cl_object *obj = ios->cis_io->ci_obj; | |
783 | ||
784 | CLOBINVRNT(env, obj, ccc_object_invariant(obj)); | |
785 | ||
786 | if (!cl_is_normalio(env, io)) | |
787 | return; | |
788 | ||
789 | LASSERT(cio->cui_tot_nrsegs >= cio->cui_nrsegs); | |
790 | LASSERT(cio->cui_tot_count >= nob); | |
791 | ||
792 | cio->cui_iov += cio->cui_nrsegs; | |
793 | cio->cui_tot_nrsegs -= cio->cui_nrsegs; | |
794 | cio->cui_tot_count -= nob; | |
795 | ||
796 | /* update the iov */ | |
797 | if (cio->cui_iov_olen > 0) { | |
798 | struct iovec *iv; | |
799 | ||
800 | cio->cui_iov--; | |
801 | cio->cui_tot_nrsegs++; | |
802 | iv = &cio->cui_iov[0]; | |
803 | if (io->ci_continue) { | |
804 | iv->iov_base += iv->iov_len; | |
805 | LASSERT(cio->cui_iov_olen > iv->iov_len); | |
806 | iv->iov_len = cio->cui_iov_olen - iv->iov_len; | |
807 | } else { | |
808 | /* restore the iov_len, in case of restart io. */ | |
809 | iv->iov_len = cio->cui_iov_olen; | |
810 | } | |
811 | cio->cui_iov_olen = 0; | |
812 | } | |
813 | } | |
814 | ||
815 | /** | |
816 | * Helper function that if necessary adjusts file size (inode->i_size), when | |
817 | * position at the offset \a pos is accessed. File size can be arbitrary stale | |
818 | * on a Lustre client, but client at least knows KMS. If accessed area is | |
819 | * inside [0, KMS], set file size to KMS, otherwise glimpse file size. | |
820 | * | |
821 | * Locking: cl_isize_lock is used to serialize changes to inode size and to | |
822 | * protect consistency between inode size and cl_object | |
823 | * attributes. cl_object_size_lock() protects consistency between cl_attr's of | |
824 | * top-object and sub-objects. | |
825 | */ | |
826 | int ccc_prep_size(const struct lu_env *env, struct cl_object *obj, | |
827 | struct cl_io *io, loff_t start, size_t count, int *exceed) | |
828 | { | |
829 | struct cl_attr *attr = ccc_env_thread_attr(env); | |
830 | struct inode *inode = ccc_object_inode(obj); | |
831 | loff_t pos = start + count - 1; | |
832 | loff_t kms; | |
833 | int result; | |
834 | ||
835 | /* | |
836 | * Consistency guarantees: following possibilities exist for the | |
837 | * relation between region being accessed and real file size at this | |
838 | * moment: | |
839 | * | |
840 | * (A): the region is completely inside of the file; | |
841 | * | |
842 | * (B-x): x bytes of region are inside of the file, the rest is | |
843 | * outside; | |
844 | * | |
845 | * (C): the region is completely outside of the file. | |
846 | * | |
847 | * This classification is stable under DLM lock already acquired by | |
848 | * the caller, because to change the class, other client has to take | |
849 | * DLM lock conflicting with our lock. Also, any updates to ->i_size | |
850 | * by other threads on this client are serialized by | |
851 | * ll_inode_size_lock(). This guarantees that short reads are handled | |
852 | * correctly in the face of concurrent writes and truncates. | |
853 | */ | |
854 | ccc_object_size_lock(obj); | |
855 | result = cl_object_attr_get(env, obj, attr); | |
856 | if (result == 0) { | |
857 | kms = attr->cat_kms; | |
858 | if (pos > kms) { | |
859 | /* | |
860 | * A glimpse is necessary to determine whether we | |
861 | * return a short read (B) or some zeroes at the end | |
862 | * of the buffer (C) | |
863 | */ | |
864 | ccc_object_size_unlock(obj); | |
865 | result = cl_glimpse_lock(env, io, inode, obj, 0); | |
866 | if (result == 0 && exceed != NULL) { | |
867 | /* If objective page index exceed end-of-file | |
868 | * page index, return directly. Do not expect | |
869 | * kernel will check such case correctly. | |
870 | * linux-2.6.18-128.1.1 miss to do that. | |
871 | * --bug 17336 */ | |
872 | loff_t size = cl_isize_read(inode); | |
873 | unsigned long cur_index = start >> PAGE_CACHE_SHIFT; | |
874 | ||
875 | if ((size == 0 && cur_index != 0) || | |
876 | (((size - 1) >> PAGE_CACHE_SHIFT) < cur_index)) | |
877 | *exceed = 1; | |
878 | } | |
879 | return result; | |
880 | } else { | |
881 | /* | |
882 | * region is within kms and, hence, within real file | |
883 | * size (A). We need to increase i_size to cover the | |
884 | * read region so that generic_file_read() will do its | |
885 | * job, but that doesn't mean the kms size is | |
886 | * _correct_, it is only the _minimum_ size. If | |
887 | * someone does a stat they will get the correct size | |
888 | * which will always be >= the kms value here. | |
889 | * b=11081 | |
890 | */ | |
891 | if (cl_isize_read(inode) < kms) { | |
892 | cl_isize_write_nolock(inode, kms); | |
893 | CDEBUG(D_VFSTRACE, | |
894 | DFID" updating i_size "LPU64"\n", | |
895 | PFID(lu_object_fid(&obj->co_lu)), | |
896 | (__u64)cl_isize_read(inode)); | |
897 | ||
898 | } | |
899 | } | |
900 | } | |
901 | ccc_object_size_unlock(obj); | |
902 | return result; | |
903 | } | |
904 | ||
905 | /***************************************************************************** | |
906 | * | |
907 | * Transfer operations. | |
908 | * | |
909 | */ | |
910 | ||
911 | void ccc_req_completion(const struct lu_env *env, | |
912 | const struct cl_req_slice *slice, int ioret) | |
913 | { | |
914 | struct ccc_req *vrq; | |
915 | ||
916 | if (ioret > 0) | |
917 | cl_stats_tally(slice->crs_dev, slice->crs_req->crq_type, ioret); | |
918 | ||
919 | vrq = cl2ccc_req(slice); | |
920 | OBD_SLAB_FREE_PTR(vrq, ccc_req_kmem); | |
921 | } | |
922 | ||
923 | /** | |
924 | * Implementation of struct cl_req_operations::cro_attr_set() for ccc | |
925 | * layer. ccc is responsible for | |
926 | * | |
927 | * - o_[mac]time | |
928 | * | |
929 | * - o_mode | |
930 | * | |
931 | * - o_parent_seq | |
932 | * | |
933 | * - o_[ug]id | |
934 | * | |
935 | * - o_parent_oid | |
936 | * | |
937 | * - o_parent_ver | |
938 | * | |
939 | * - o_ioepoch, | |
940 | * | |
941 | * and capability. | |
942 | */ | |
943 | void ccc_req_attr_set(const struct lu_env *env, | |
944 | const struct cl_req_slice *slice, | |
945 | const struct cl_object *obj, | |
946 | struct cl_req_attr *attr, obd_valid flags) | |
947 | { | |
948 | struct inode *inode; | |
949 | struct obdo *oa; | |
950 | obd_flag valid_flags; | |
951 | ||
952 | oa = attr->cra_oa; | |
953 | inode = ccc_object_inode(obj); | |
954 | valid_flags = OBD_MD_FLTYPE; | |
955 | ||
956 | if ((flags & OBD_MD_FLOSSCAPA) != 0) { | |
957 | LASSERT(attr->cra_capa == NULL); | |
958 | attr->cra_capa = cl_capa_lookup(inode, | |
959 | slice->crs_req->crq_type); | |
960 | } | |
961 | ||
962 | if (slice->crs_req->crq_type == CRT_WRITE) { | |
963 | if (flags & OBD_MD_FLEPOCH) { | |
964 | oa->o_valid |= OBD_MD_FLEPOCH; | |
965 | oa->o_ioepoch = cl_i2info(inode)->lli_ioepoch; | |
966 | valid_flags |= OBD_MD_FLMTIME | OBD_MD_FLCTIME | | |
967 | OBD_MD_FLUID | OBD_MD_FLGID; | |
968 | } | |
969 | } | |
970 | obdo_from_inode(oa, inode, valid_flags & flags); | |
971 | obdo_set_parent_fid(oa, &cl_i2info(inode)->lli_fid); | |
972 | memcpy(attr->cra_jobid, cl_i2info(inode)->lli_jobid, | |
973 | JOBSTATS_JOBID_SIZE); | |
974 | } | |
975 | ||
976 | const struct cl_req_operations ccc_req_ops = { | |
977 | .cro_attr_set = ccc_req_attr_set, | |
978 | .cro_completion = ccc_req_completion | |
979 | }; | |
980 | ||
981 | int cl_setattr_ost(struct inode *inode, const struct iattr *attr, | |
982 | struct obd_capa *capa) | |
983 | { | |
984 | struct lu_env *env; | |
985 | struct cl_io *io; | |
986 | int result; | |
987 | int refcheck; | |
988 | ||
989 | ENTRY; | |
990 | ||
991 | env = cl_env_get(&refcheck); | |
992 | if (IS_ERR(env)) | |
993 | RETURN(PTR_ERR(env)); | |
994 | ||
995 | io = ccc_env_thread_io(env); | |
996 | io->ci_obj = cl_i2info(inode)->lli_clob; | |
997 | ||
998 | io->u.ci_setattr.sa_attr.lvb_atime = LTIME_S(attr->ia_atime); | |
999 | io->u.ci_setattr.sa_attr.lvb_mtime = LTIME_S(attr->ia_mtime); | |
1000 | io->u.ci_setattr.sa_attr.lvb_ctime = LTIME_S(attr->ia_ctime); | |
1001 | io->u.ci_setattr.sa_attr.lvb_size = attr->ia_size; | |
1002 | io->u.ci_setattr.sa_valid = attr->ia_valid; | |
1003 | io->u.ci_setattr.sa_capa = capa; | |
1004 | ||
1005 | again: | |
1006 | if (cl_io_init(env, io, CIT_SETATTR, io->ci_obj) == 0) { | |
1007 | struct ccc_io *cio = ccc_env_io(env); | |
1008 | ||
1009 | if (attr->ia_valid & ATTR_FILE) | |
1010 | /* populate the file descriptor for ftruncate to honor | |
1011 | * group lock - see LU-787 */ | |
1012 | cio->cui_fd = cl_iattr2fd(inode, attr); | |
1013 | ||
1014 | result = cl_io_loop(env, io); | |
1015 | } else { | |
1016 | result = io->ci_result; | |
1017 | } | |
1018 | cl_io_fini(env, io); | |
1019 | if (unlikely(io->ci_need_restart)) | |
1020 | goto again; | |
1021 | cl_env_put(env, &refcheck); | |
1022 | RETURN(result); | |
1023 | } | |
1024 | ||
1025 | /***************************************************************************** | |
1026 | * | |
1027 | * Type conversions. | |
1028 | * | |
1029 | */ | |
1030 | ||
1031 | struct lu_device *ccc2lu_dev(struct ccc_device *vdv) | |
1032 | { | |
1033 | return &vdv->cdv_cl.cd_lu_dev; | |
1034 | } | |
1035 | ||
1036 | struct ccc_device *lu2ccc_dev(const struct lu_device *d) | |
1037 | { | |
1038 | return container_of0(d, struct ccc_device, cdv_cl.cd_lu_dev); | |
1039 | } | |
1040 | ||
1041 | struct ccc_device *cl2ccc_dev(const struct cl_device *d) | |
1042 | { | |
1043 | return container_of0(d, struct ccc_device, cdv_cl); | |
1044 | } | |
1045 | ||
1046 | struct lu_object *ccc2lu(struct ccc_object *vob) | |
1047 | { | |
1048 | return &vob->cob_cl.co_lu; | |
1049 | } | |
1050 | ||
1051 | struct ccc_object *lu2ccc(const struct lu_object *obj) | |
1052 | { | |
1053 | return container_of0(obj, struct ccc_object, cob_cl.co_lu); | |
1054 | } | |
1055 | ||
1056 | struct ccc_object *cl2ccc(const struct cl_object *obj) | |
1057 | { | |
1058 | return container_of0(obj, struct ccc_object, cob_cl); | |
1059 | } | |
1060 | ||
1061 | struct ccc_lock *cl2ccc_lock(const struct cl_lock_slice *slice) | |
1062 | { | |
1063 | return container_of(slice, struct ccc_lock, clk_cl); | |
1064 | } | |
1065 | ||
1066 | struct ccc_io *cl2ccc_io(const struct lu_env *env, | |
1067 | const struct cl_io_slice *slice) | |
1068 | { | |
1069 | struct ccc_io *cio; | |
1070 | ||
1071 | cio = container_of(slice, struct ccc_io, cui_cl); | |
1072 | LASSERT(cio == ccc_env_io(env)); | |
1073 | return cio; | |
1074 | } | |
1075 | ||
1076 | struct ccc_req *cl2ccc_req(const struct cl_req_slice *slice) | |
1077 | { | |
1078 | return container_of0(slice, struct ccc_req, crq_cl); | |
1079 | } | |
1080 | ||
1081 | struct page *cl2vm_page(const struct cl_page_slice *slice) | |
1082 | { | |
1083 | return cl2ccc_page(slice)->cpg_page; | |
1084 | } | |
1085 | ||
1086 | /***************************************************************************** | |
1087 | * | |
1088 | * Accessors. | |
1089 | * | |
1090 | */ | |
1091 | int ccc_object_invariant(const struct cl_object *obj) | |
1092 | { | |
1093 | struct inode *inode = ccc_object_inode(obj); | |
1094 | struct cl_inode_info *lli = cl_i2info(inode); | |
1095 | ||
1096 | return (S_ISREG(cl_inode_mode(inode)) || | |
1097 | /* i_mode of unlinked inode is zeroed. */ | |
1098 | cl_inode_mode(inode) == 0) && lli->lli_clob == obj; | |
1099 | } | |
1100 | ||
1101 | struct inode *ccc_object_inode(const struct cl_object *obj) | |
1102 | { | |
1103 | return cl2ccc(obj)->cob_inode; | |
1104 | } | |
1105 | ||
1106 | /** | |
1107 | * Returns a pointer to cl_page associated with \a vmpage, without acquiring | |
1108 | * additional reference to the resulting page. This is an unsafe version of | |
1109 | * cl_vmpage_page() that can only be used under vmpage lock. | |
1110 | */ | |
1111 | struct cl_page *ccc_vmpage_page_transient(struct page *vmpage) | |
1112 | { | |
1113 | KLASSERT(PageLocked(vmpage)); | |
1114 | return (struct cl_page *)vmpage->private; | |
1115 | } | |
1116 | ||
1117 | /** | |
1118 | * Initialize or update CLIO structures for regular files when new | |
1119 | * meta-data arrives from the server. | |
1120 | * | |
1121 | * \param inode regular file inode | |
1122 | * \param md new file metadata from MDS | |
1123 | * - allocates cl_object if necessary, | |
1124 | * - updated layout, if object was already here. | |
1125 | */ | |
1126 | int cl_file_inode_init(struct inode *inode, struct lustre_md *md) | |
1127 | { | |
1128 | struct lu_env *env; | |
1129 | struct cl_inode_info *lli; | |
1130 | struct cl_object *clob; | |
1131 | struct lu_site *site; | |
1132 | struct lu_fid *fid; | |
1133 | struct cl_object_conf conf = { | |
1134 | .coc_inode = inode, | |
1135 | .u = { | |
1136 | .coc_md = md | |
1137 | } | |
1138 | }; | |
1139 | int result = 0; | |
1140 | int refcheck; | |
1141 | ||
1142 | LASSERT(md->body->valid & OBD_MD_FLID); | |
1143 | LASSERT(S_ISREG(cl_inode_mode(inode))); | |
1144 | ||
1145 | env = cl_env_get(&refcheck); | |
1146 | if (IS_ERR(env)) | |
1147 | return PTR_ERR(env); | |
1148 | ||
1149 | site = cl_i2sbi(inode)->ll_site; | |
1150 | lli = cl_i2info(inode); | |
1151 | fid = &lli->lli_fid; | |
1152 | LASSERT(fid_is_sane(fid)); | |
1153 | ||
1154 | if (lli->lli_clob == NULL) { | |
1155 | /* clob is slave of inode, empty lli_clob means for new inode, | |
1156 | * there is no clob in cache with the given fid, so it is | |
1157 | * unnecessary to perform lookup-alloc-lookup-insert, just | |
1158 | * alloc and insert directly. */ | |
1159 | LASSERT(inode->i_state & I_NEW); | |
1160 | conf.coc_lu.loc_flags = LOC_F_NEW; | |
1161 | clob = cl_object_find(env, lu2cl_dev(site->ls_top_dev), | |
1162 | fid, &conf); | |
1163 | if (!IS_ERR(clob)) { | |
1164 | /* | |
1165 | * No locking is necessary, as new inode is | |
1166 | * locked by I_NEW bit. | |
1167 | */ | |
1168 | lli->lli_clob = clob; | |
1169 | lli->lli_has_smd = md->lsm != NULL; | |
1170 | lu_object_ref_add(&clob->co_lu, "inode", inode); | |
1171 | } else | |
1172 | result = PTR_ERR(clob); | |
1173 | } else { | |
1174 | result = cl_conf_set(env, lli->lli_clob, &conf); | |
1175 | } | |
1176 | ||
1177 | cl_env_put(env, &refcheck); | |
1178 | ||
1179 | if (result != 0) | |
1180 | CERROR("Failure to initialize cl object "DFID": %d\n", | |
1181 | PFID(fid), result); | |
1182 | return result; | |
1183 | } | |
1184 | ||
1185 | /** | |
1186 | * Wait for others drop their references of the object at first, then we drop | |
1187 | * the last one, which will lead to the object be destroyed immediately. | |
1188 | * Must be called after cl_object_kill() against this object. | |
1189 | * | |
1190 | * The reason we want to do this is: destroying top object will wait for sub | |
1191 | * objects being destroyed first, so we can't let bottom layer (e.g. from ASTs) | |
1192 | * to initiate top object destroying which may deadlock. See bz22520. | |
1193 | */ | |
1194 | static void cl_object_put_last(struct lu_env *env, struct cl_object *obj) | |
1195 | { | |
1196 | struct lu_object_header *header = obj->co_lu.lo_header; | |
1197 | wait_queue_t waiter; | |
1198 | ||
1199 | if (unlikely(atomic_read(&header->loh_ref) != 1)) { | |
1200 | struct lu_site *site = obj->co_lu.lo_dev->ld_site; | |
1201 | struct lu_site_bkt_data *bkt; | |
1202 | ||
1203 | bkt = lu_site_bkt_from_fid(site, &header->loh_fid); | |
1204 | ||
1205 | init_waitqueue_entry_current(&waiter); | |
1206 | add_wait_queue(&bkt->lsb_marche_funebre, &waiter); | |
1207 | ||
1208 | while (1) { | |
1209 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1210 | if (atomic_read(&header->loh_ref) == 1) | |
1211 | break; | |
1212 | waitq_wait(&waiter, TASK_UNINTERRUPTIBLE); | |
1213 | } | |
1214 | ||
1215 | set_current_state(TASK_RUNNING); | |
1216 | remove_wait_queue(&bkt->lsb_marche_funebre, &waiter); | |
1217 | } | |
1218 | ||
1219 | cl_object_put(env, obj); | |
1220 | } | |
1221 | ||
1222 | void cl_inode_fini(struct inode *inode) | |
1223 | { | |
1224 | struct lu_env *env; | |
1225 | struct cl_inode_info *lli = cl_i2info(inode); | |
1226 | struct cl_object *clob = lli->lli_clob; | |
1227 | int refcheck; | |
1228 | int emergency; | |
1229 | ||
1230 | if (clob != NULL) { | |
1231 | void *cookie; | |
1232 | ||
1233 | cookie = cl_env_reenter(); | |
1234 | env = cl_env_get(&refcheck); | |
1235 | emergency = IS_ERR(env); | |
1236 | if (emergency) { | |
1237 | mutex_lock(&ccc_inode_fini_guard); | |
1238 | LASSERT(ccc_inode_fini_env != NULL); | |
1239 | cl_env_implant(ccc_inode_fini_env, &refcheck); | |
1240 | env = ccc_inode_fini_env; | |
1241 | } | |
1242 | /* | |
1243 | * cl_object cache is a slave to inode cache (which, in turn | |
1244 | * is a slave to dentry cache), don't keep cl_object in memory | |
1245 | * when its master is evicted. | |
1246 | */ | |
1247 | cl_object_kill(env, clob); | |
1248 | lu_object_ref_del(&clob->co_lu, "inode", inode); | |
1249 | cl_object_put_last(env, clob); | |
1250 | lli->lli_clob = NULL; | |
1251 | if (emergency) { | |
1252 | cl_env_unplant(ccc_inode_fini_env, &refcheck); | |
1253 | mutex_unlock(&ccc_inode_fini_guard); | |
1254 | } else | |
1255 | cl_env_put(env, &refcheck); | |
1256 | cl_env_reexit(cookie); | |
1257 | } | |
1258 | } | |
1259 | ||
1260 | /** | |
1261 | * return IF_* type for given lu_dirent entry. | |
1262 | * IF_* flag shld be converted to particular OS file type in | |
1263 | * platform llite module. | |
1264 | */ | |
1265 | __u16 ll_dirent_type_get(struct lu_dirent *ent) | |
1266 | { | |
1267 | __u16 type = 0; | |
1268 | struct luda_type *lt; | |
1269 | int len = 0; | |
1270 | ||
1271 | if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) { | |
1272 | const unsigned align = sizeof(struct luda_type) - 1; | |
1273 | ||
1274 | len = le16_to_cpu(ent->lde_namelen); | |
1275 | len = (len + align) & ~align; | |
1276 | lt = (void *)ent->lde_name + len; | |
1277 | type = IFTODT(le16_to_cpu(lt->lt_type)); | |
1278 | } | |
1279 | return type; | |
1280 | } | |
1281 | ||
1282 | /** | |
1283 | * build inode number from passed @fid */ | |
1284 | __u64 cl_fid_build_ino(const struct lu_fid *fid, int api32) | |
1285 | { | |
1286 | if (BITS_PER_LONG == 32 || api32) | |
1287 | RETURN(fid_flatten32(fid)); | |
1288 | else | |
1289 | RETURN(fid_flatten(fid)); | |
1290 | } | |
1291 | ||
1292 | /** | |
1293 | * build inode generation from passed @fid. If our FID overflows the 32-bit | |
1294 | * inode number then return a non-zero generation to distinguish them. */ | |
1295 | __u32 cl_fid_build_gen(const struct lu_fid *fid) | |
1296 | { | |
1297 | __u32 gen; | |
1298 | ENTRY; | |
1299 | ||
1300 | if (fid_is_igif(fid)) { | |
1301 | gen = lu_igif_gen(fid); | |
1302 | RETURN(gen); | |
1303 | } | |
1304 | ||
1305 | gen = (fid_flatten(fid) >> 32); | |
1306 | RETURN(gen); | |
1307 | } | |
1308 | ||
1309 | /* lsm is unreliable after hsm implementation as layout can be changed at | |
1310 | * any time. This is only to support old, non-clio-ized interfaces. It will | |
1311 | * cause deadlock if clio operations are called with this extra layout refcount | |
1312 | * because in case the layout changed during the IO, ll_layout_refresh() will | |
1313 | * have to wait for the refcount to become zero to destroy the older layout. | |
1314 | * | |
1315 | * Notice that the lsm returned by this function may not be valid unless called | |
1316 | * inside layout lock - MDS_INODELOCK_LAYOUT. */ | |
1317 | struct lov_stripe_md *ccc_inode_lsm_get(struct inode *inode) | |
1318 | { | |
1319 | return lov_lsm_get(cl_i2info(inode)->lli_clob); | |
1320 | } | |
1321 | ||
1322 | void inline ccc_inode_lsm_put(struct inode *inode, struct lov_stripe_md *lsm) | |
1323 | { | |
1324 | lov_lsm_put(cl_i2info(inode)->lli_clob, lsm); | |
1325 | } |