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 | * | |
1dc563a6 | 30 | * Copyright (c) 2011, 2015, Intel Corporation. |
d7e09d03 PT |
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 | ||
9fdaf8c0 | 44 | #include "../../include/linux/libcfs/libcfs.h" |
d7e09d03 PT |
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 | ||
67a235f5 GKH |
53 | #include "../include/obd.h" |
54 | #include "../include/obd_support.h" | |
55 | #include "../include/lustre_fid.h" | |
56 | #include "../include/lustre_lite.h" | |
57 | #include "../include/lustre_dlm.h" | |
58 | #include "../include/lustre_ver.h" | |
59 | #include "../include/lustre_mdc.h" | |
60 | #include "../include/cl_object.h" | |
d7e09d03 | 61 | |
67a235f5 | 62 | #include "../include/lclient.h" |
d7e09d03 PT |
63 | |
64 | #include "../llite/llite_internal.h" | |
65 | ||
2d95f10e | 66 | static const struct cl_req_operations ccc_req_ops; |
d7e09d03 PT |
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", | |
d0631921 | 82 | .ckd_size = sizeof(struct ccc_lock) |
d7e09d03 PT |
83 | }, |
84 | { | |
85 | .ckd_cache = &ccc_object_kmem, | |
86 | .ckd_name = "ccc_object_kmem", | |
d0631921 | 87 | .ckd_size = sizeof(struct ccc_object) |
d7e09d03 PT |
88 | }, |
89 | { | |
90 | .ckd_cache = &ccc_thread_kmem, | |
91 | .ckd_name = "ccc_thread_kmem", | |
d0631921 | 92 | .ckd_size = sizeof(struct ccc_thread_info), |
d7e09d03 PT |
93 | }, |
94 | { | |
95 | .ckd_cache = &ccc_session_kmem, | |
96 | .ckd_name = "ccc_session_kmem", | |
d0631921 | 97 | .ckd_size = sizeof(struct ccc_session) |
d7e09d03 PT |
98 | }, |
99 | { | |
100 | .ckd_cache = &ccc_req_kmem, | |
101 | .ckd_name = "ccc_req_kmem", | |
d0631921 | 102 | .ckd_size = sizeof(struct ccc_req) |
d7e09d03 PT |
103 | }, |
104 | { | |
105 | .ckd_cache = NULL | |
106 | } | |
107 | }; | |
108 | ||
109 | /***************************************************************************** | |
110 | * | |
111 | * Vvp device and device type functions. | |
112 | * | |
113 | */ | |
114 | ||
0be19afa | 115 | void *ccc_key_init(const struct lu_context *ctx, struct lu_context_key *key) |
d7e09d03 PT |
116 | { |
117 | struct ccc_thread_info *info; | |
118 | ||
3dd84289 | 119 | info = kmem_cache_zalloc(ccc_thread_kmem, GFP_NOFS); |
eb26ebb8 | 120 | if (!info) |
d7e09d03 PT |
121 | info = ERR_PTR(-ENOMEM); |
122 | return info; | |
123 | } | |
124 | ||
125 | void ccc_key_fini(const struct lu_context *ctx, | |
126 | struct lu_context_key *key, void *data) | |
127 | { | |
128 | struct ccc_thread_info *info = data; | |
772f34f5 | 129 | |
50d30362 | 130 | kmem_cache_free(ccc_thread_kmem, info); |
d7e09d03 PT |
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 | ||
3dd84289 | 138 | session = kmem_cache_zalloc(ccc_session_kmem, GFP_NOFS); |
eb26ebb8 | 139 | if (!session) |
d7e09d03 PT |
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; | |
772f34f5 | 148 | |
50d30362 | 149 | kmem_cache_free(ccc_session_kmem, session); |
d7e09d03 PT |
150 | } |
151 | ||
152 | struct lu_context_key ccc_key = { | |
153 | .lct_tags = LCT_CL_THREAD, | |
154 | .lct_init = ccc_key_init, | |
155 | .lct_fini = ccc_key_fini | |
156 | }; | |
157 | ||
158 | struct lu_context_key ccc_session_key = { | |
159 | .lct_tags = LCT_SESSION, | |
160 | .lct_init = ccc_session_key_init, | |
161 | .lct_fini = ccc_session_key_fini | |
162 | }; | |
163 | ||
d7e09d03 | 164 | /* type constructor/destructor: ccc_type_{init,fini,start,stop}(). */ |
5152bda8 | 165 | /* LU_TYPE_INIT_FINI(ccc, &ccc_key, &ccc_session_key); */ |
d7e09d03 PT |
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; | |
d7e09d03 PT |
172 | |
173 | vdv = lu2ccc_dev(d); | |
174 | vdv->cdv_next = lu2cl_dev(next); | |
175 | ||
eb26ebb8 | 176 | LASSERT(d->ld_site && next->ld_type); |
d7e09d03 PT |
177 | next->ld_site = d->ld_site; |
178 | rc = next->ld_type->ldt_ops->ldto_device_init( | |
179 | env, next, next->ld_type->ldt_name, NULL); | |
180 | if (rc == 0) { | |
181 | lu_device_get(next); | |
182 | lu_ref_add(&next->ld_reference, "lu-stack", &lu_site_init); | |
183 | } | |
0a3bdb00 | 184 | return rc; |
d7e09d03 PT |
185 | } |
186 | ||
187 | struct lu_device *ccc_device_fini(const struct lu_env *env, | |
188 | struct lu_device *d) | |
189 | { | |
190 | return cl2lu_dev(lu2ccc_dev(d)->cdv_next); | |
191 | } | |
192 | ||
193 | struct lu_device *ccc_device_alloc(const struct lu_env *env, | |
194 | struct lu_device_type *t, | |
195 | struct lustre_cfg *cfg, | |
196 | const struct lu_device_operations *luops, | |
197 | const struct cl_device_operations *clops) | |
198 | { | |
199 | struct ccc_device *vdv; | |
200 | struct lu_device *lud; | |
201 | struct cl_site *site; | |
202 | int rc; | |
d7e09d03 | 203 | |
00c40c16 | 204 | vdv = kzalloc(sizeof(*vdv), GFP_NOFS); |
c38ce354 | 205 | if (!vdv) |
0a3bdb00 | 206 | return ERR_PTR(-ENOMEM); |
d7e09d03 PT |
207 | |
208 | lud = &vdv->cdv_cl.cd_lu_dev; | |
209 | cl_device_init(&vdv->cdv_cl, t); | |
210 | ccc2lu_dev(vdv)->ld_ops = luops; | |
211 | vdv->cdv_cl.cd_ops = clops; | |
212 | ||
00c40c16 | 213 | site = kzalloc(sizeof(*site), GFP_NOFS); |
eb26ebb8 | 214 | if (site) { |
d7e09d03 PT |
215 | rc = cl_site_init(site, &vdv->cdv_cl); |
216 | if (rc == 0) | |
217 | rc = lu_site_init_finish(&site->cs_lu); | |
218 | else { | |
eb26ebb8 | 219 | LASSERT(!lud->ld_site); |
d7e09d03 | 220 | CERROR("Cannot init lu_site, rc %d.\n", rc); |
00c40c16 | 221 | kfree(site); |
d7e09d03 PT |
222 | } |
223 | } else | |
224 | rc = -ENOMEM; | |
225 | if (rc != 0) { | |
226 | ccc_device_free(env, lud); | |
227 | lud = ERR_PTR(rc); | |
228 | } | |
0a3bdb00 | 229 | return lud; |
d7e09d03 PT |
230 | } |
231 | ||
232 | struct lu_device *ccc_device_free(const struct lu_env *env, | |
233 | struct lu_device *d) | |
234 | { | |
235 | struct ccc_device *vdv = lu2ccc_dev(d); | |
236 | struct cl_site *site = lu2cl_site(d->ld_site); | |
237 | struct lu_device *next = cl2lu_dev(vdv->cdv_next); | |
238 | ||
eb26ebb8 | 239 | if (d->ld_site) { |
d7e09d03 | 240 | cl_site_fini(site); |
00c40c16 | 241 | kfree(site); |
d7e09d03 PT |
242 | } |
243 | cl_device_fini(lu2cl_dev(d)); | |
00c40c16 | 244 | kfree(vdv); |
d7e09d03 PT |
245 | return next; |
246 | } | |
247 | ||
248 | int ccc_req_init(const struct lu_env *env, struct cl_device *dev, | |
249 | struct cl_req *req) | |
250 | { | |
251 | struct ccc_req *vrq; | |
252 | int result; | |
253 | ||
3dd84289 | 254 | vrq = kmem_cache_zalloc(ccc_req_kmem, GFP_NOFS); |
eb26ebb8 | 255 | if (vrq) { |
d7e09d03 PT |
256 | cl_req_slice_add(req, &vrq->crq_cl, dev, &ccc_req_ops); |
257 | result = 0; | |
258 | } else | |
259 | result = -ENOMEM; | |
260 | return result; | |
261 | } | |
262 | ||
263 | /** | |
264 | * An `emergency' environment used by ccc_inode_fini() when cl_env_get() | |
265 | * fails. Access to this environment is serialized by ccc_inode_fini_guard | |
266 | * mutex. | |
267 | */ | |
847d47fa | 268 | static struct lu_env *ccc_inode_fini_env; |
d7e09d03 PT |
269 | |
270 | /** | |
271 | * A mutex serializing calls to slp_inode_fini() under extreme memory | |
272 | * pressure, when environments cannot be allocated. | |
273 | */ | |
274 | static DEFINE_MUTEX(ccc_inode_fini_guard); | |
275 | static int dummy_refcheck; | |
276 | ||
277 | int ccc_global_init(struct lu_device_type *device_type) | |
278 | { | |
279 | int result; | |
280 | ||
281 | result = lu_kmem_init(ccc_caches); | |
282 | if (result) | |
283 | return result; | |
284 | ||
285 | result = lu_device_type_init(device_type); | |
286 | if (result) | |
287 | goto out_kmem; | |
288 | ||
289 | ccc_inode_fini_env = cl_env_alloc(&dummy_refcheck, | |
290 | LCT_REMEMBER|LCT_NOREF); | |
291 | if (IS_ERR(ccc_inode_fini_env)) { | |
292 | result = PTR_ERR(ccc_inode_fini_env); | |
293 | goto out_device; | |
294 | } | |
295 | ||
296 | ccc_inode_fini_env->le_ctx.lc_cookie = 0x4; | |
297 | return 0; | |
298 | out_device: | |
299 | lu_device_type_fini(device_type); | |
300 | out_kmem: | |
301 | lu_kmem_fini(ccc_caches); | |
302 | return result; | |
303 | } | |
304 | ||
305 | void ccc_global_fini(struct lu_device_type *device_type) | |
306 | { | |
eb26ebb8 | 307 | if (ccc_inode_fini_env) { |
d7e09d03 PT |
308 | cl_env_put(ccc_inode_fini_env, &dummy_refcheck); |
309 | ccc_inode_fini_env = NULL; | |
310 | } | |
311 | lu_device_type_fini(device_type); | |
312 | lu_kmem_fini(ccc_caches); | |
313 | } | |
314 | ||
315 | /***************************************************************************** | |
316 | * | |
317 | * Object operations. | |
318 | * | |
319 | */ | |
320 | ||
321 | struct lu_object *ccc_object_alloc(const struct lu_env *env, | |
322 | const struct lu_object_header *unused, | |
323 | struct lu_device *dev, | |
324 | const struct cl_object_operations *clops, | |
325 | const struct lu_object_operations *luops) | |
326 | { | |
327 | struct ccc_object *vob; | |
328 | struct lu_object *obj; | |
329 | ||
3dd84289 | 330 | vob = kmem_cache_zalloc(ccc_object_kmem, GFP_NOFS); |
eb26ebb8 | 331 | if (vob) { |
d7e09d03 PT |
332 | struct cl_object_header *hdr; |
333 | ||
334 | obj = ccc2lu(vob); | |
335 | hdr = &vob->cob_header; | |
336 | cl_object_header_init(hdr); | |
337 | lu_object_init(obj, &hdr->coh_lu, dev); | |
338 | lu_object_add_top(&hdr->coh_lu, obj); | |
339 | ||
340 | vob->cob_cl.co_ops = clops; | |
341 | obj->lo_ops = luops; | |
342 | } else | |
343 | obj = NULL; | |
344 | return obj; | |
345 | } | |
346 | ||
347 | int ccc_object_init0(const struct lu_env *env, | |
348 | struct ccc_object *vob, | |
349 | const struct cl_object_conf *conf) | |
350 | { | |
351 | vob->cob_inode = conf->coc_inode; | |
352 | vob->cob_transient_pages = 0; | |
353 | cl_object_page_init(&vob->cob_cl, sizeof(struct ccc_page)); | |
354 | return 0; | |
355 | } | |
356 | ||
357 | int ccc_object_init(const struct lu_env *env, struct lu_object *obj, | |
358 | const struct lu_object_conf *conf) | |
359 | { | |
360 | struct ccc_device *dev = lu2ccc_dev(obj->lo_dev); | |
361 | struct ccc_object *vob = lu2ccc(obj); | |
362 | struct lu_object *below; | |
363 | struct lu_device *under; | |
364 | int result; | |
365 | ||
366 | under = &dev->cdv_next->cd_lu_dev; | |
367 | below = under->ld_ops->ldo_object_alloc(env, obj->lo_header, under); | |
eb26ebb8 | 368 | if (below) { |
d7e09d03 PT |
369 | const struct cl_object_conf *cconf; |
370 | ||
371 | cconf = lu2cl_conf(conf); | |
372 | INIT_LIST_HEAD(&vob->cob_pending_list); | |
373 | lu_object_add(obj, below); | |
374 | result = ccc_object_init0(env, vob, cconf); | |
375 | } else | |
376 | result = -ENOMEM; | |
377 | return result; | |
378 | } | |
379 | ||
380 | void ccc_object_free(const struct lu_env *env, struct lu_object *obj) | |
381 | { | |
382 | struct ccc_object *vob = lu2ccc(obj); | |
383 | ||
384 | lu_object_fini(obj); | |
385 | lu_object_header_fini(obj->lo_header); | |
50d30362 | 386 | kmem_cache_free(ccc_object_kmem, vob); |
d7e09d03 PT |
387 | } |
388 | ||
389 | int ccc_lock_init(const struct lu_env *env, | |
390 | struct cl_object *obj, struct cl_lock *lock, | |
391 | const struct cl_io *unused, | |
392 | const struct cl_lock_operations *lkops) | |
393 | { | |
394 | struct ccc_lock *clk; | |
395 | int result; | |
396 | ||
397 | CLOBINVRNT(env, obj, ccc_object_invariant(obj)); | |
398 | ||
3dd84289 | 399 | clk = kmem_cache_zalloc(ccc_lock_kmem, GFP_NOFS); |
eb26ebb8 | 400 | if (clk) { |
d7e09d03 PT |
401 | cl_lock_slice_add(lock, &clk->clk_cl, obj, lkops); |
402 | result = 0; | |
403 | } else | |
404 | result = -ENOMEM; | |
405 | return result; | |
406 | } | |
407 | ||
d7e09d03 PT |
408 | int ccc_object_glimpse(const struct lu_env *env, |
409 | const struct cl_object *obj, struct ost_lvb *lvb) | |
410 | { | |
411 | struct inode *inode = ccc_object_inode(obj); | |
412 | ||
d7e09d03 PT |
413 | lvb->lvb_mtime = cl_inode_mtime(inode); |
414 | lvb->lvb_atime = cl_inode_atime(inode); | |
415 | lvb->lvb_ctime = cl_inode_ctime(inode); | |
416 | /* | |
417 | * LU-417: Add dirty pages block count lest i_blocks reports 0, some | |
418 | * "cp" or "tar" on remote node may think it's a completely sparse file | |
419 | * and skip it. | |
420 | */ | |
421 | if (lvb->lvb_size > 0 && lvb->lvb_blocks == 0) | |
422 | lvb->lvb_blocks = dirty_cnt(inode); | |
0a3bdb00 | 423 | return 0; |
d7e09d03 PT |
424 | } |
425 | ||
d7e09d03 PT |
426 | static void ccc_object_size_lock(struct cl_object *obj) |
427 | { | |
428 | struct inode *inode = ccc_object_inode(obj); | |
429 | ||
abcf8080 | 430 | ll_inode_size_lock(inode); |
d7e09d03 PT |
431 | cl_object_attr_lock(obj); |
432 | } | |
433 | ||
434 | static void ccc_object_size_unlock(struct cl_object *obj) | |
435 | { | |
436 | struct inode *inode = ccc_object_inode(obj); | |
437 | ||
438 | cl_object_attr_unlock(obj); | |
abcf8080 | 439 | ll_inode_size_unlock(inode); |
d7e09d03 PT |
440 | } |
441 | ||
442 | /***************************************************************************** | |
443 | * | |
444 | * Page operations. | |
445 | * | |
446 | */ | |
447 | ||
448 | struct page *ccc_page_vmpage(const struct lu_env *env, | |
449 | const struct cl_page_slice *slice) | |
450 | { | |
451 | return cl2vm_page(slice); | |
452 | } | |
453 | ||
454 | int ccc_page_is_under_lock(const struct lu_env *env, | |
455 | const struct cl_page_slice *slice, | |
456 | struct cl_io *io) | |
457 | { | |
458 | struct ccc_io *cio = ccc_env_io(env); | |
459 | struct cl_lock_descr *desc = &ccc_env_info(env)->cti_descr; | |
460 | struct cl_page *page = slice->cpl_page; | |
461 | ||
462 | int result; | |
463 | ||
d7e09d03 PT |
464 | if (io->ci_type == CIT_READ || io->ci_type == CIT_WRITE || |
465 | io->ci_type == CIT_FAULT) { | |
466 | if (cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED) | |
467 | result = -EBUSY; | |
468 | else { | |
469 | desc->cld_start = page->cp_index; | |
470 | desc->cld_end = page->cp_index; | |
471 | desc->cld_obj = page->cp_obj; | |
472 | desc->cld_mode = CLM_READ; | |
473 | result = cl_queue_match(&io->ci_lockset.cls_done, | |
474 | desc) ? -EBUSY : 0; | |
475 | } | |
476 | } else | |
477 | result = 0; | |
0a3bdb00 | 478 | return result; |
d7e09d03 PT |
479 | } |
480 | ||
481 | int ccc_fail(const struct lu_env *env, const struct cl_page_slice *slice) | |
482 | { | |
483 | /* | |
484 | * Cached read? | |
485 | */ | |
486 | LBUG(); | |
487 | return 0; | |
488 | } | |
489 | ||
d7e09d03 PT |
490 | int ccc_transient_page_prep(const struct lu_env *env, |
491 | const struct cl_page_slice *slice, | |
492 | struct cl_io *unused) | |
493 | { | |
d7e09d03 | 494 | /* transient page should always be sent. */ |
0a3bdb00 | 495 | return 0; |
d7e09d03 PT |
496 | } |
497 | ||
498 | /***************************************************************************** | |
499 | * | |
500 | * Lock operations. | |
501 | * | |
502 | */ | |
503 | ||
504 | void ccc_lock_delete(const struct lu_env *env, | |
505 | const struct cl_lock_slice *slice) | |
506 | { | |
507 | CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj)); | |
508 | } | |
509 | ||
510 | void ccc_lock_fini(const struct lu_env *env, struct cl_lock_slice *slice) | |
511 | { | |
512 | struct ccc_lock *clk = cl2ccc_lock(slice); | |
772f34f5 | 513 | |
50d30362 | 514 | kmem_cache_free(ccc_lock_kmem, clk); |
d7e09d03 PT |
515 | } |
516 | ||
517 | int ccc_lock_enqueue(const struct lu_env *env, | |
518 | const struct cl_lock_slice *slice, | |
519 | struct cl_io *unused, __u32 enqflags) | |
4c309612 JX |
520 | { |
521 | CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj)); | |
522 | return 0; | |
523 | } | |
524 | ||
525 | int ccc_lock_use(const struct lu_env *env, const struct cl_lock_slice *slice) | |
d7e09d03 PT |
526 | { |
527 | CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj)); | |
528 | return 0; | |
529 | } | |
530 | ||
531 | int ccc_lock_unuse(const struct lu_env *env, const struct cl_lock_slice *slice) | |
532 | { | |
533 | CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj)); | |
534 | return 0; | |
535 | } | |
536 | ||
537 | int ccc_lock_wait(const struct lu_env *env, const struct cl_lock_slice *slice) | |
538 | { | |
539 | CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj)); | |
540 | return 0; | |
541 | } | |
542 | ||
543 | /** | |
544 | * Implementation of cl_lock_operations::clo_fits_into() methods for ccc | |
545 | * layer. This function is executed every time io finds an existing lock in | |
546 | * the lock cache while creating new lock. This function has to decide whether | |
547 | * cached lock "fits" into io. | |
548 | * | |
549 | * \param slice lock to be checked | |
550 | * \param io IO that wants a lock. | |
551 | * | |
552 | * \see lov_lock_fits_into(). | |
553 | */ | |
554 | int ccc_lock_fits_into(const struct lu_env *env, | |
555 | const struct cl_lock_slice *slice, | |
556 | const struct cl_lock_descr *need, | |
557 | const struct cl_io *io) | |
558 | { | |
559 | const struct cl_lock *lock = slice->cls_lock; | |
560 | const struct cl_lock_descr *descr = &lock->cll_descr; | |
561 | const struct ccc_io *cio = ccc_env_io(env); | |
562 | int result; | |
563 | ||
d7e09d03 PT |
564 | /* |
565 | * Work around DLM peculiarity: it assumes that glimpse | |
566 | * (LDLM_FL_HAS_INTENT) lock is always LCK_PR, and returns reads lock | |
567 | * when asked for LCK_PW lock with LDLM_FL_HAS_INTENT flag set. Make | |
568 | * sure that glimpse doesn't get CLM_WRITE top-lock, so that it | |
569 | * doesn't enqueue CLM_WRITE sub-locks. | |
570 | */ | |
571 | if (cio->cui_glimpse) | |
572 | result = descr->cld_mode != CLM_WRITE; | |
573 | ||
574 | /* | |
575 | * Also, don't match incomplete write locks for read, otherwise read | |
576 | * would enqueue missing sub-locks in the write mode. | |
577 | */ | |
578 | else if (need->cld_mode != descr->cld_mode) | |
579 | result = lock->cll_state >= CLS_ENQUEUED; | |
580 | else | |
581 | result = 1; | |
0a3bdb00 | 582 | return result; |
d7e09d03 PT |
583 | } |
584 | ||
585 | /** | |
586 | * Implements cl_lock_operations::clo_state() method for ccc layer, invoked | |
587 | * whenever lock state changes. Transfers object attributes, that might be | |
588 | * updated as a result of lock acquiring into inode. | |
589 | */ | |
590 | void ccc_lock_state(const struct lu_env *env, | |
591 | const struct cl_lock_slice *slice, | |
592 | enum cl_lock_state state) | |
593 | { | |
594 | struct cl_lock *lock = slice->cls_lock; | |
d7e09d03 PT |
595 | |
596 | /* | |
597 | * Refresh inode attributes when the lock is moving into CLS_HELD | |
598 | * state, and only when this is a result of real enqueue, rather than | |
599 | * of finding lock in the cache. | |
600 | */ | |
601 | if (state == CLS_HELD && lock->cll_state < CLS_HELD) { | |
602 | struct cl_object *obj; | |
603 | struct inode *inode; | |
604 | ||
605 | obj = slice->cls_obj; | |
606 | inode = ccc_object_inode(obj); | |
607 | ||
608 | /* vmtruncate() sets the i_size | |
609 | * under both a DLM lock and the | |
610 | * ll_inode_size_lock(). If we don't get the | |
611 | * ll_inode_size_lock() here we can match the DLM lock and | |
612 | * reset i_size. generic_file_write can then trust the | |
613 | * stale i_size when doing appending writes and effectively | |
614 | * cancel the result of the truncate. Getting the | |
615 | * ll_inode_size_lock() after the enqueue maintains the DLM | |
ec9a1ac2 OD |
616 | * -> ll_inode_size_lock() acquiring order. |
617 | */ | |
d7e09d03 PT |
618 | if (lock->cll_descr.cld_start == 0 && |
619 | lock->cll_descr.cld_end == CL_PAGE_EOF) | |
620 | cl_merge_lvb(env, inode); | |
621 | } | |
d7e09d03 PT |
622 | } |
623 | ||
624 | /***************************************************************************** | |
625 | * | |
626 | * io operations. | |
627 | * | |
628 | */ | |
629 | ||
d7e09d03 PT |
630 | int ccc_io_one_lock_index(const struct lu_env *env, struct cl_io *io, |
631 | __u32 enqflags, enum cl_lock_mode mode, | |
632 | pgoff_t start, pgoff_t end) | |
633 | { | |
634 | struct ccc_io *cio = ccc_env_io(env); | |
635 | struct cl_lock_descr *descr = &cio->cui_link.cill_descr; | |
636 | struct cl_object *obj = io->ci_obj; | |
637 | ||
638 | CLOBINVRNT(env, obj, ccc_object_invariant(obj)); | |
d7e09d03 PT |
639 | |
640 | CDEBUG(D_VFSTRACE, "lock: %d [%lu, %lu]\n", mode, start, end); | |
641 | ||
ec83e611 | 642 | memset(&cio->cui_link, 0, sizeof(cio->cui_link)); |
d7e09d03 PT |
643 | |
644 | if (cio->cui_fd && (cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) { | |
645 | descr->cld_mode = CLM_GROUP; | |
646 | descr->cld_gid = cio->cui_fd->fd_grouplock.cg_gid; | |
647 | } else { | |
648 | descr->cld_mode = mode; | |
649 | } | |
650 | descr->cld_obj = obj; | |
651 | descr->cld_start = start; | |
652 | descr->cld_end = end; | |
653 | descr->cld_enq_flags = enqflags; | |
654 | ||
655 | cl_io_lock_add(env, io, &cio->cui_link); | |
0a3bdb00 | 656 | return 0; |
d7e09d03 PT |
657 | } |
658 | ||
659 | void ccc_io_update_iov(const struct lu_env *env, | |
660 | struct ccc_io *cio, struct cl_io *io) | |
661 | { | |
d7e09d03 PT |
662 | size_t size = io->u.ci_rw.crw_count; |
663 | ||
eb26ebb8 | 664 | if (!cl_is_normalio(env, io) || !cio->cui_iter) |
d7e09d03 PT |
665 | return; |
666 | ||
b42b15fd | 667 | iov_iter_truncate(cio->cui_iter, size); |
d7e09d03 PT |
668 | } |
669 | ||
670 | int ccc_io_one_lock(const struct lu_env *env, struct cl_io *io, | |
671 | __u32 enqflags, enum cl_lock_mode mode, | |
672 | loff_t start, loff_t end) | |
673 | { | |
674 | struct cl_object *obj = io->ci_obj; | |
772f34f5 | 675 | |
d7e09d03 PT |
676 | return ccc_io_one_lock_index(env, io, enqflags, mode, |
677 | cl_index(obj, start), cl_index(obj, end)); | |
678 | } | |
679 | ||
680 | void ccc_io_end(const struct lu_env *env, const struct cl_io_slice *ios) | |
681 | { | |
682 | CLOBINVRNT(env, ios->cis_io->ci_obj, | |
683 | ccc_object_invariant(ios->cis_io->ci_obj)); | |
684 | } | |
685 | ||
686 | void ccc_io_advance(const struct lu_env *env, | |
687 | const struct cl_io_slice *ios, | |
688 | size_t nob) | |
689 | { | |
690 | struct ccc_io *cio = cl2ccc_io(env, ios); | |
691 | struct cl_io *io = ios->cis_io; | |
692 | struct cl_object *obj = ios->cis_io->ci_obj; | |
693 | ||
694 | CLOBINVRNT(env, obj, ccc_object_invariant(obj)); | |
695 | ||
696 | if (!cl_is_normalio(env, io)) | |
697 | return; | |
698 | ||
b42b15fd | 699 | iov_iter_reexpand(cio->cui_iter, cio->cui_tot_count -= nob); |
d7e09d03 PT |
700 | } |
701 | ||
702 | /** | |
703 | * Helper function that if necessary adjusts file size (inode->i_size), when | |
704 | * position at the offset \a pos is accessed. File size can be arbitrary stale | |
705 | * on a Lustre client, but client at least knows KMS. If accessed area is | |
706 | * inside [0, KMS], set file size to KMS, otherwise glimpse file size. | |
707 | * | |
708 | * Locking: cl_isize_lock is used to serialize changes to inode size and to | |
709 | * protect consistency between inode size and cl_object | |
710 | * attributes. cl_object_size_lock() protects consistency between cl_attr's of | |
711 | * top-object and sub-objects. | |
712 | */ | |
713 | int ccc_prep_size(const struct lu_env *env, struct cl_object *obj, | |
714 | struct cl_io *io, loff_t start, size_t count, int *exceed) | |
715 | { | |
716 | struct cl_attr *attr = ccc_env_thread_attr(env); | |
717 | struct inode *inode = ccc_object_inode(obj); | |
718 | loff_t pos = start + count - 1; | |
719 | loff_t kms; | |
720 | int result; | |
721 | ||
722 | /* | |
723 | * Consistency guarantees: following possibilities exist for the | |
724 | * relation between region being accessed and real file size at this | |
725 | * moment: | |
726 | * | |
727 | * (A): the region is completely inside of the file; | |
728 | * | |
729 | * (B-x): x bytes of region are inside of the file, the rest is | |
730 | * outside; | |
731 | * | |
732 | * (C): the region is completely outside of the file. | |
733 | * | |
734 | * This classification is stable under DLM lock already acquired by | |
735 | * the caller, because to change the class, other client has to take | |
736 | * DLM lock conflicting with our lock. Also, any updates to ->i_size | |
737 | * by other threads on this client are serialized by | |
738 | * ll_inode_size_lock(). This guarantees that short reads are handled | |
739 | * correctly in the face of concurrent writes and truncates. | |
740 | */ | |
741 | ccc_object_size_lock(obj); | |
742 | result = cl_object_attr_get(env, obj, attr); | |
743 | if (result == 0) { | |
744 | kms = attr->cat_kms; | |
745 | if (pos > kms) { | |
746 | /* | |
747 | * A glimpse is necessary to determine whether we | |
748 | * return a short read (B) or some zeroes at the end | |
749 | * of the buffer (C) | |
750 | */ | |
751 | ccc_object_size_unlock(obj); | |
752 | result = cl_glimpse_lock(env, io, inode, obj, 0); | |
eb26ebb8 | 753 | if (result == 0 && exceed) { |
d7e09d03 PT |
754 | /* If objective page index exceed end-of-file |
755 | * page index, return directly. Do not expect | |
756 | * kernel will check such case correctly. | |
757 | * linux-2.6.18-128.1.1 miss to do that. | |
ec9a1ac2 OD |
758 | * --bug 17336 |
759 | */ | |
d7e09d03 | 760 | loff_t size = cl_isize_read(inode); |
09cbfeaf | 761 | loff_t cur_index = start >> PAGE_SHIFT; |
159dc4bf | 762 | loff_t size_index = (size - 1) >> |
09cbfeaf | 763 | PAGE_SHIFT; |
d7e09d03 PT |
764 | |
765 | if ((size == 0 && cur_index != 0) || | |
81851d46 | 766 | size_index < cur_index) |
98e11370 | 767 | *exceed = 1; |
d7e09d03 PT |
768 | } |
769 | return result; | |
109a1624 AG |
770 | } |
771 | /* | |
772 | * region is within kms and, hence, within real file | |
773 | * size (A). We need to increase i_size to cover the | |
774 | * read region so that generic_file_read() will do its | |
775 | * job, but that doesn't mean the kms size is | |
776 | * _correct_, it is only the _minimum_ size. If | |
777 | * someone does a stat they will get the correct size | |
778 | * which will always be >= the kms value here. | |
779 | * b=11081 | |
780 | */ | |
781 | if (cl_isize_read(inode) < kms) { | |
782 | cl_isize_write_nolock(inode, kms); | |
783 | CDEBUG(D_VFSTRACE, | |
784 | DFID" updating i_size %llu\n", | |
785 | PFID(lu_object_fid(&obj->co_lu)), | |
786 | (__u64)cl_isize_read(inode)); | |
d7e09d03 | 787 | |
d7e09d03 PT |
788 | } |
789 | } | |
790 | ccc_object_size_unlock(obj); | |
791 | return result; | |
792 | } | |
793 | ||
794 | /***************************************************************************** | |
795 | * | |
796 | * Transfer operations. | |
797 | * | |
798 | */ | |
799 | ||
800 | void ccc_req_completion(const struct lu_env *env, | |
801 | const struct cl_req_slice *slice, int ioret) | |
802 | { | |
803 | struct ccc_req *vrq; | |
804 | ||
805 | if (ioret > 0) | |
806 | cl_stats_tally(slice->crs_dev, slice->crs_req->crq_type, ioret); | |
807 | ||
808 | vrq = cl2ccc_req(slice); | |
50d30362 | 809 | kmem_cache_free(ccc_req_kmem, vrq); |
d7e09d03 PT |
810 | } |
811 | ||
812 | /** | |
813 | * Implementation of struct cl_req_operations::cro_attr_set() for ccc | |
814 | * layer. ccc is responsible for | |
815 | * | |
816 | * - o_[mac]time | |
817 | * | |
818 | * - o_mode | |
819 | * | |
820 | * - o_parent_seq | |
821 | * | |
822 | * - o_[ug]id | |
823 | * | |
824 | * - o_parent_oid | |
825 | * | |
826 | * - o_parent_ver | |
827 | * | |
828 | * - o_ioepoch, | |
829 | * | |
d7e09d03 PT |
830 | */ |
831 | void ccc_req_attr_set(const struct lu_env *env, | |
832 | const struct cl_req_slice *slice, | |
833 | const struct cl_object *obj, | |
21aef7d9 | 834 | struct cl_req_attr *attr, u64 flags) |
d7e09d03 PT |
835 | { |
836 | struct inode *inode; | |
837 | struct obdo *oa; | |
21aef7d9 | 838 | u32 valid_flags; |
d7e09d03 PT |
839 | |
840 | oa = attr->cra_oa; | |
841 | inode = ccc_object_inode(obj); | |
842 | valid_flags = OBD_MD_FLTYPE; | |
843 | ||
d7e09d03 PT |
844 | if (slice->crs_req->crq_type == CRT_WRITE) { |
845 | if (flags & OBD_MD_FLEPOCH) { | |
846 | oa->o_valid |= OBD_MD_FLEPOCH; | |
847 | oa->o_ioepoch = cl_i2info(inode)->lli_ioepoch; | |
848 | valid_flags |= OBD_MD_FLMTIME | OBD_MD_FLCTIME | | |
849 | OBD_MD_FLUID | OBD_MD_FLGID; | |
850 | } | |
851 | } | |
852 | obdo_from_inode(oa, inode, valid_flags & flags); | |
853 | obdo_set_parent_fid(oa, &cl_i2info(inode)->lli_fid); | |
854 | memcpy(attr->cra_jobid, cl_i2info(inode)->lli_jobid, | |
855 | JOBSTATS_JOBID_SIZE); | |
856 | } | |
857 | ||
2d95f10e | 858 | static const struct cl_req_operations ccc_req_ops = { |
d7e09d03 PT |
859 | .cro_attr_set = ccc_req_attr_set, |
860 | .cro_completion = ccc_req_completion | |
861 | }; | |
862 | ||
ef2e0f55 | 863 | int cl_setattr_ost(struct inode *inode, const struct iattr *attr) |
d7e09d03 PT |
864 | { |
865 | struct lu_env *env; | |
866 | struct cl_io *io; | |
867 | int result; | |
868 | int refcheck; | |
869 | ||
d7e09d03 PT |
870 | env = cl_env_get(&refcheck); |
871 | if (IS_ERR(env)) | |
0a3bdb00 | 872 | return PTR_ERR(env); |
d7e09d03 PT |
873 | |
874 | io = ccc_env_thread_io(env); | |
875 | io->ci_obj = cl_i2info(inode)->lli_clob; | |
876 | ||
877 | io->u.ci_setattr.sa_attr.lvb_atime = LTIME_S(attr->ia_atime); | |
878 | io->u.ci_setattr.sa_attr.lvb_mtime = LTIME_S(attr->ia_mtime); | |
879 | io->u.ci_setattr.sa_attr.lvb_ctime = LTIME_S(attr->ia_ctime); | |
880 | io->u.ci_setattr.sa_attr.lvb_size = attr->ia_size; | |
881 | io->u.ci_setattr.sa_valid = attr->ia_valid; | |
d7e09d03 PT |
882 | |
883 | again: | |
884 | if (cl_io_init(env, io, CIT_SETATTR, io->ci_obj) == 0) { | |
885 | struct ccc_io *cio = ccc_env_io(env); | |
886 | ||
887 | if (attr->ia_valid & ATTR_FILE) | |
888 | /* populate the file descriptor for ftruncate to honor | |
ec9a1ac2 OD |
889 | * group lock - see LU-787 |
890 | */ | |
d7e09d03 PT |
891 | cio->cui_fd = cl_iattr2fd(inode, attr); |
892 | ||
893 | result = cl_io_loop(env, io); | |
894 | } else { | |
895 | result = io->ci_result; | |
896 | } | |
897 | cl_io_fini(env, io); | |
898 | if (unlikely(io->ci_need_restart)) | |
899 | goto again; | |
5ea17d6c JL |
900 | /* HSM import case: file is released, cannot be restored |
901 | * no need to fail except if restore registration failed | |
ec9a1ac2 OD |
902 | * with -ENODATA |
903 | */ | |
5ea17d6c JL |
904 | if (result == -ENODATA && io->ci_restore_needed && |
905 | io->ci_result != -ENODATA) | |
906 | result = 0; | |
d7e09d03 | 907 | cl_env_put(env, &refcheck); |
0a3bdb00 | 908 | return result; |
d7e09d03 PT |
909 | } |
910 | ||
911 | /***************************************************************************** | |
912 | * | |
913 | * Type conversions. | |
914 | * | |
915 | */ | |
916 | ||
917 | struct lu_device *ccc2lu_dev(struct ccc_device *vdv) | |
918 | { | |
919 | return &vdv->cdv_cl.cd_lu_dev; | |
920 | } | |
921 | ||
922 | struct ccc_device *lu2ccc_dev(const struct lu_device *d) | |
923 | { | |
924 | return container_of0(d, struct ccc_device, cdv_cl.cd_lu_dev); | |
925 | } | |
926 | ||
927 | struct ccc_device *cl2ccc_dev(const struct cl_device *d) | |
928 | { | |
929 | return container_of0(d, struct ccc_device, cdv_cl); | |
930 | } | |
931 | ||
932 | struct lu_object *ccc2lu(struct ccc_object *vob) | |
933 | { | |
934 | return &vob->cob_cl.co_lu; | |
935 | } | |
936 | ||
937 | struct ccc_object *lu2ccc(const struct lu_object *obj) | |
938 | { | |
939 | return container_of0(obj, struct ccc_object, cob_cl.co_lu); | |
940 | } | |
941 | ||
942 | struct ccc_object *cl2ccc(const struct cl_object *obj) | |
943 | { | |
944 | return container_of0(obj, struct ccc_object, cob_cl); | |
945 | } | |
946 | ||
947 | struct ccc_lock *cl2ccc_lock(const struct cl_lock_slice *slice) | |
948 | { | |
949 | return container_of(slice, struct ccc_lock, clk_cl); | |
950 | } | |
951 | ||
952 | struct ccc_io *cl2ccc_io(const struct lu_env *env, | |
953 | const struct cl_io_slice *slice) | |
954 | { | |
955 | struct ccc_io *cio; | |
956 | ||
957 | cio = container_of(slice, struct ccc_io, cui_cl); | |
958 | LASSERT(cio == ccc_env_io(env)); | |
959 | return cio; | |
960 | } | |
961 | ||
962 | struct ccc_req *cl2ccc_req(const struct cl_req_slice *slice) | |
963 | { | |
964 | return container_of0(slice, struct ccc_req, crq_cl); | |
965 | } | |
966 | ||
967 | struct page *cl2vm_page(const struct cl_page_slice *slice) | |
968 | { | |
969 | return cl2ccc_page(slice)->cpg_page; | |
970 | } | |
971 | ||
972 | /***************************************************************************** | |
973 | * | |
974 | * Accessors. | |
975 | * | |
976 | */ | |
977 | int ccc_object_invariant(const struct cl_object *obj) | |
978 | { | |
979 | struct inode *inode = ccc_object_inode(obj); | |
980 | struct cl_inode_info *lli = cl_i2info(inode); | |
981 | ||
982 | return (S_ISREG(cl_inode_mode(inode)) || | |
983 | /* i_mode of unlinked inode is zeroed. */ | |
984 | cl_inode_mode(inode) == 0) && lli->lli_clob == obj; | |
985 | } | |
986 | ||
987 | struct inode *ccc_object_inode(const struct cl_object *obj) | |
988 | { | |
989 | return cl2ccc(obj)->cob_inode; | |
990 | } | |
991 | ||
d7e09d03 PT |
992 | /** |
993 | * Initialize or update CLIO structures for regular files when new | |
994 | * meta-data arrives from the server. | |
995 | * | |
996 | * \param inode regular file inode | |
997 | * \param md new file metadata from MDS | |
998 | * - allocates cl_object if necessary, | |
999 | * - updated layout, if object was already here. | |
1000 | */ | |
1001 | int cl_file_inode_init(struct inode *inode, struct lustre_md *md) | |
1002 | { | |
1003 | struct lu_env *env; | |
1004 | struct cl_inode_info *lli; | |
1005 | struct cl_object *clob; | |
1006 | struct lu_site *site; | |
1007 | struct lu_fid *fid; | |
1008 | struct cl_object_conf conf = { | |
1009 | .coc_inode = inode, | |
1010 | .u = { | |
1011 | .coc_md = md | |
1012 | } | |
1013 | }; | |
1014 | int result = 0; | |
1015 | int refcheck; | |
1016 | ||
1017 | LASSERT(md->body->valid & OBD_MD_FLID); | |
1018 | LASSERT(S_ISREG(cl_inode_mode(inode))); | |
1019 | ||
1020 | env = cl_env_get(&refcheck); | |
1021 | if (IS_ERR(env)) | |
1022 | return PTR_ERR(env); | |
1023 | ||
1024 | site = cl_i2sbi(inode)->ll_site; | |
1025 | lli = cl_i2info(inode); | |
1026 | fid = &lli->lli_fid; | |
1027 | LASSERT(fid_is_sane(fid)); | |
1028 | ||
eb26ebb8 | 1029 | if (!lli->lli_clob) { |
d7e09d03 PT |
1030 | /* clob is slave of inode, empty lli_clob means for new inode, |
1031 | * there is no clob in cache with the given fid, so it is | |
1032 | * unnecessary to perform lookup-alloc-lookup-insert, just | |
ec9a1ac2 OD |
1033 | * alloc and insert directly. |
1034 | */ | |
d7e09d03 PT |
1035 | LASSERT(inode->i_state & I_NEW); |
1036 | conf.coc_lu.loc_flags = LOC_F_NEW; | |
1037 | clob = cl_object_find(env, lu2cl_dev(site->ls_top_dev), | |
1038 | fid, &conf); | |
1039 | if (!IS_ERR(clob)) { | |
1040 | /* | |
1041 | * No locking is necessary, as new inode is | |
1042 | * locked by I_NEW bit. | |
1043 | */ | |
1044 | lli->lli_clob = clob; | |
5dd16419 | 1045 | lli->lli_has_smd = lsm_has_objects(md->lsm); |
d7e09d03 PT |
1046 | lu_object_ref_add(&clob->co_lu, "inode", inode); |
1047 | } else | |
1048 | result = PTR_ERR(clob); | |
1049 | } else { | |
1050 | result = cl_conf_set(env, lli->lli_clob, &conf); | |
1051 | } | |
1052 | ||
1053 | cl_env_put(env, &refcheck); | |
1054 | ||
1055 | if (result != 0) | |
1056 | CERROR("Failure to initialize cl object "DFID": %d\n", | |
1057 | PFID(fid), result); | |
1058 | return result; | |
1059 | } | |
1060 | ||
1061 | /** | |
1062 | * Wait for others drop their references of the object at first, then we drop | |
1063 | * the last one, which will lead to the object be destroyed immediately. | |
1064 | * Must be called after cl_object_kill() against this object. | |
1065 | * | |
1066 | * The reason we want to do this is: destroying top object will wait for sub | |
1067 | * objects being destroyed first, so we can't let bottom layer (e.g. from ASTs) | |
1068 | * to initiate top object destroying which may deadlock. See bz22520. | |
1069 | */ | |
1070 | static void cl_object_put_last(struct lu_env *env, struct cl_object *obj) | |
1071 | { | |
1072 | struct lu_object_header *header = obj->co_lu.lo_header; | |
1073 | wait_queue_t waiter; | |
1074 | ||
1075 | if (unlikely(atomic_read(&header->loh_ref) != 1)) { | |
1076 | struct lu_site *site = obj->co_lu.lo_dev->ld_site; | |
1077 | struct lu_site_bkt_data *bkt; | |
1078 | ||
1079 | bkt = lu_site_bkt_from_fid(site, &header->loh_fid); | |
1080 | ||
9e795d35 | 1081 | init_waitqueue_entry(&waiter, current); |
d7e09d03 PT |
1082 | add_wait_queue(&bkt->lsb_marche_funebre, &waiter); |
1083 | ||
1084 | while (1) { | |
1085 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1086 | if (atomic_read(&header->loh_ref) == 1) | |
1087 | break; | |
b3669a7f | 1088 | schedule(); |
d7e09d03 PT |
1089 | } |
1090 | ||
1091 | set_current_state(TASK_RUNNING); | |
1092 | remove_wait_queue(&bkt->lsb_marche_funebre, &waiter); | |
1093 | } | |
1094 | ||
1095 | cl_object_put(env, obj); | |
1096 | } | |
1097 | ||
1098 | void cl_inode_fini(struct inode *inode) | |
1099 | { | |
1100 | struct lu_env *env; | |
1101 | struct cl_inode_info *lli = cl_i2info(inode); | |
1102 | struct cl_object *clob = lli->lli_clob; | |
1103 | int refcheck; | |
1104 | int emergency; | |
1105 | ||
eb26ebb8 | 1106 | if (clob) { |
d7e09d03 PT |
1107 | void *cookie; |
1108 | ||
1109 | cookie = cl_env_reenter(); | |
1110 | env = cl_env_get(&refcheck); | |
1111 | emergency = IS_ERR(env); | |
1112 | if (emergency) { | |
1113 | mutex_lock(&ccc_inode_fini_guard); | |
eb26ebb8 | 1114 | LASSERT(ccc_inode_fini_env); |
d7e09d03 PT |
1115 | cl_env_implant(ccc_inode_fini_env, &refcheck); |
1116 | env = ccc_inode_fini_env; | |
1117 | } | |
1118 | /* | |
1119 | * cl_object cache is a slave to inode cache (which, in turn | |
1120 | * is a slave to dentry cache), don't keep cl_object in memory | |
1121 | * when its master is evicted. | |
1122 | */ | |
1123 | cl_object_kill(env, clob); | |
1124 | lu_object_ref_del(&clob->co_lu, "inode", inode); | |
1125 | cl_object_put_last(env, clob); | |
1126 | lli->lli_clob = NULL; | |
1127 | if (emergency) { | |
1128 | cl_env_unplant(ccc_inode_fini_env, &refcheck); | |
1129 | mutex_unlock(&ccc_inode_fini_guard); | |
1130 | } else | |
1131 | cl_env_put(env, &refcheck); | |
1132 | cl_env_reexit(cookie); | |
1133 | } | |
1134 | } | |
1135 | ||
1136 | /** | |
1137 | * return IF_* type for given lu_dirent entry. | |
1138 | * IF_* flag shld be converted to particular OS file type in | |
1139 | * platform llite module. | |
1140 | */ | |
1141 | __u16 ll_dirent_type_get(struct lu_dirent *ent) | |
1142 | { | |
1143 | __u16 type = 0; | |
1144 | struct luda_type *lt; | |
1145 | int len = 0; | |
1146 | ||
1147 | if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) { | |
1148 | const unsigned align = sizeof(struct luda_type) - 1; | |
1149 | ||
1150 | len = le16_to_cpu(ent->lde_namelen); | |
1151 | len = (len + align) & ~align; | |
1152 | lt = (void *)ent->lde_name + len; | |
1153 | type = IFTODT(le16_to_cpu(lt->lt_type)); | |
1154 | } | |
1155 | return type; | |
1156 | } | |
1157 | ||
1158 | /** | |
ec9a1ac2 OD |
1159 | * build inode number from passed @fid |
1160 | */ | |
d7e09d03 PT |
1161 | __u64 cl_fid_build_ino(const struct lu_fid *fid, int api32) |
1162 | { | |
1163 | if (BITS_PER_LONG == 32 || api32) | |
0a3bdb00 | 1164 | return fid_flatten32(fid); |
d7e09d03 | 1165 | else |
0a3bdb00 | 1166 | return fid_flatten(fid); |
d7e09d03 PT |
1167 | } |
1168 | ||
1169 | /** | |
1170 | * build inode generation from passed @fid. If our FID overflows the 32-bit | |
ec9a1ac2 OD |
1171 | * inode number then return a non-zero generation to distinguish them. |
1172 | */ | |
d7e09d03 PT |
1173 | __u32 cl_fid_build_gen(const struct lu_fid *fid) |
1174 | { | |
1175 | __u32 gen; | |
d7e09d03 PT |
1176 | |
1177 | if (fid_is_igif(fid)) { | |
1178 | gen = lu_igif_gen(fid); | |
0a3bdb00 | 1179 | return gen; |
d7e09d03 PT |
1180 | } |
1181 | ||
159dc4bf | 1182 | gen = fid_flatten(fid) >> 32; |
0a3bdb00 | 1183 | return gen; |
d7e09d03 PT |
1184 | } |
1185 | ||
1186 | /* lsm is unreliable after hsm implementation as layout can be changed at | |
1187 | * any time. This is only to support old, non-clio-ized interfaces. It will | |
1188 | * cause deadlock if clio operations are called with this extra layout refcount | |
1189 | * because in case the layout changed during the IO, ll_layout_refresh() will | |
1190 | * have to wait for the refcount to become zero to destroy the older layout. | |
1191 | * | |
1192 | * Notice that the lsm returned by this function may not be valid unless called | |
ec9a1ac2 OD |
1193 | * inside layout lock - MDS_INODELOCK_LAYOUT. |
1194 | */ | |
d7e09d03 PT |
1195 | struct lov_stripe_md *ccc_inode_lsm_get(struct inode *inode) |
1196 | { | |
1197 | return lov_lsm_get(cl_i2info(inode)->lli_clob); | |
1198 | } | |
1199 | ||
60dc39b4 | 1200 | inline void ccc_inode_lsm_put(struct inode *inode, struct lov_stripe_md *lsm) |
d7e09d03 PT |
1201 | { |
1202 | lov_lsm_put(cl_i2info(inode)->lli_clob, lsm); | |
1203 | } |