4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
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
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * Implementation of cl_page for VVP layer.
38 * Author: Nikita Danilov <nikita.danilov@sun.com>
39 * Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
42 #define DEBUG_SUBSYSTEM S_LLITE
46 #include <lustre_lite.h>
48 #include "vvp_internal.h"
50 /*****************************************************************************
56 static void vvp_page_fini_common(struct ccc_page *cp)
58 struct page *vmpage = cp->cpg_page;
60 LASSERT(vmpage != NULL);
61 page_cache_release(vmpage);
64 static void vvp_page_fini(const struct lu_env *env,
65 struct cl_page_slice *slice)
67 struct ccc_page *cp = cl2ccc_page(slice);
68 struct page *vmpage = cp->cpg_page;
71 * vmpage->private was already cleared when page was moved into
74 LASSERT((struct cl_page *)vmpage->private != slice->cpl_page);
75 vvp_page_fini_common(cp);
78 static int vvp_page_own(const struct lu_env *env,
79 const struct cl_page_slice *slice, struct cl_io *io,
82 struct ccc_page *vpg = cl2ccc_page(slice);
83 struct page *vmpage = vpg->cpg_page;
85 LASSERT(vmpage != NULL);
87 if (!trylock_page(vmpage))
90 if (unlikely(PageWriteback(vmpage))) {
99 wait_on_page_writeback(vmpage);
103 static void vvp_page_assume(const struct lu_env *env,
104 const struct cl_page_slice *slice,
105 struct cl_io *unused)
107 struct page *vmpage = cl2vm_page(slice);
109 LASSERT(vmpage != NULL);
110 LASSERT(PageLocked(vmpage));
111 wait_on_page_writeback(vmpage);
114 static void vvp_page_unassume(const struct lu_env *env,
115 const struct cl_page_slice *slice,
116 struct cl_io *unused)
118 struct page *vmpage = cl2vm_page(slice);
120 LASSERT(vmpage != NULL);
121 LASSERT(PageLocked(vmpage));
124 static void vvp_page_disown(const struct lu_env *env,
125 const struct cl_page_slice *slice, struct cl_io *io)
127 struct page *vmpage = cl2vm_page(slice);
129 LASSERT(vmpage != NULL);
130 LASSERT(PageLocked(vmpage));
132 unlock_page(cl2vm_page(slice));
135 static void vvp_page_discard(const struct lu_env *env,
136 const struct cl_page_slice *slice,
137 struct cl_io *unused)
139 struct page *vmpage = cl2vm_page(slice);
140 struct ccc_page *cpg = cl2ccc_page(slice);
142 LASSERT(vmpage != NULL);
143 LASSERT(PageLocked(vmpage));
145 if (cpg->cpg_defer_uptodate && !cpg->cpg_ra_used)
146 ll_ra_stats_inc(vmpage->mapping->host, RA_STAT_DISCARDED);
148 ll_invalidate_page(vmpage);
151 static void vvp_page_delete(const struct lu_env *env,
152 const struct cl_page_slice *slice)
154 struct page *vmpage = cl2vm_page(slice);
155 struct inode *inode = vmpage->mapping->host;
156 struct cl_object *obj = slice->cpl_obj;
157 struct cl_page *page = slice->cpl_page;
160 LASSERT(PageLocked(vmpage));
161 LASSERT((struct cl_page *)vmpage->private == page);
162 LASSERT(inode == ccc_object_inode(obj));
164 vvp_write_complete(cl2ccc(obj), cl2ccc_page(slice));
166 /* Drop the reference count held in vvp_page_init */
167 refc = atomic_dec_return(&page->cp_ref);
168 LASSERTF(refc >= 1, "page = %p, refc = %d\n", page, refc);
170 ClearPageUptodate(vmpage);
171 ClearPagePrivate(vmpage);
174 * Reference from vmpage to cl_page is removed, but the reference back
175 * is still here. It is removed later in vvp_page_fini().
179 static void vvp_page_export(const struct lu_env *env,
180 const struct cl_page_slice *slice,
183 struct page *vmpage = cl2vm_page(slice);
185 LASSERT(vmpage != NULL);
186 LASSERT(PageLocked(vmpage));
188 SetPageUptodate(vmpage);
190 ClearPageUptodate(vmpage);
193 static int vvp_page_is_vmlocked(const struct lu_env *env,
194 const struct cl_page_slice *slice)
196 return PageLocked(cl2vm_page(slice)) ? -EBUSY : -ENODATA;
199 static int vvp_page_prep_read(const struct lu_env *env,
200 const struct cl_page_slice *slice,
201 struct cl_io *unused)
204 /* Skip the page already marked as PG_uptodate. */
205 RETURN(PageUptodate(cl2vm_page(slice)) ? -EALREADY : 0);
208 static int vvp_page_prep_write(const struct lu_env *env,
209 const struct cl_page_slice *slice,
210 struct cl_io *unused)
212 struct page *vmpage = cl2vm_page(slice);
214 LASSERT(PageLocked(vmpage));
215 LASSERT(!PageDirty(vmpage));
217 set_page_writeback(vmpage);
218 vvp_write_pending(cl2ccc(slice->cpl_obj), cl2ccc_page(slice));
224 * Handles page transfer errors at VM level.
226 * This takes inode as a separate argument, because inode on which error is to
227 * be set can be different from \a vmpage inode in case of direct-io.
229 static void vvp_vmpage_error(struct inode *inode, struct page *vmpage, int ioret)
231 struct ccc_object *obj = cl_inode2ccc(inode);
234 ClearPageError(vmpage);
235 obj->cob_discard_page_warned = 0;
237 SetPageError(vmpage);
238 if (ioret == -ENOSPC)
239 set_bit(AS_ENOSPC, &inode->i_mapping->flags);
241 set_bit(AS_EIO, &inode->i_mapping->flags);
243 if ((ioret == -ESHUTDOWN || ioret == -EINTR) &&
244 obj->cob_discard_page_warned == 0) {
245 obj->cob_discard_page_warned = 1;
246 ll_dirty_page_discard_warn(vmpage, ioret);
251 static void vvp_page_completion_read(const struct lu_env *env,
252 const struct cl_page_slice *slice,
255 struct ccc_page *cp = cl2ccc_page(slice);
256 struct page *vmpage = cp->cpg_page;
257 struct cl_page *page = slice->cpl_page;
258 struct inode *inode = ccc_object_inode(page->cp_obj);
261 LASSERT(PageLocked(vmpage));
262 CL_PAGE_HEADER(D_PAGE, env, page, "completing READ with %d\n", ioret);
264 if (cp->cpg_defer_uptodate)
265 ll_ra_count_put(ll_i2sbi(inode), 1);
268 if (!cp->cpg_defer_uptodate)
269 cl_page_export(env, page, 1);
271 cp->cpg_defer_uptodate = 0;
273 if (page->cp_sync_io == NULL)
279 static void vvp_page_completion_write(const struct lu_env *env,
280 const struct cl_page_slice *slice,
283 struct ccc_page *cp = cl2ccc_page(slice);
284 struct cl_page *pg = slice->cpl_page;
285 struct page *vmpage = cp->cpg_page;
288 LASSERT(ergo(pg->cp_sync_io != NULL, PageLocked(vmpage)));
289 LASSERT(PageWriteback(vmpage));
291 CL_PAGE_HEADER(D_PAGE, env, pg, "completing WRITE with %d\n", ioret);
294 * TODO: Actually it makes sense to add the page into oap pending
295 * list again and so that we don't need to take the page out from
296 * SoM write pending list, if we just meet a recoverable error,
298 * To implement this, we just need to return a non zero value in
299 * ->cpo_completion method. The underlying transfer should be notified
300 * and then re-add the page into pending transfer queue. -jay
303 cp->cpg_write_queued = 0;
304 vvp_write_complete(cl2ccc(slice->cpl_obj), cp);
307 * Only mark the page error only when it's an async write because
308 * applications won't wait for IO to finish.
310 if (pg->cp_sync_io == NULL)
311 vvp_vmpage_error(ccc_object_inode(pg->cp_obj), vmpage, ioret);
313 end_page_writeback(vmpage);
318 * Implements cl_page_operations::cpo_make_ready() method.
320 * This is called to yank a page from the transfer cache and to send it out as
321 * a part of transfer. This function try-locks the page. If try-lock failed,
322 * page is owned by some concurrent IO, and should be skipped (this is bad,
323 * but hopefully rare situation, as it usually results in transfer being
324 * shorter than possible).
326 * \retval 0 success, page can be placed into transfer
328 * \retval -EAGAIN page is either used by concurrent IO has been
329 * truncated. Skip it.
331 static int vvp_page_make_ready(const struct lu_env *env,
332 const struct cl_page_slice *slice)
334 struct page *vmpage = cl2vm_page(slice);
335 struct cl_page *pg = slice->cpl_page;
339 if (clear_page_dirty_for_io(vmpage)) {
340 LASSERT(pg->cp_state == CPS_CACHED);
341 /* This actually clears the dirty bit in the radix
343 set_page_writeback(vmpage);
344 vvp_write_pending(cl2ccc(slice->cpl_obj),
346 CL_PAGE_HEADER(D_PAGE, env, pg, "readied\n");
347 } else if (pg->cp_state == CPS_PAGEOUT) {
348 /* is it possible for osc_flush_async_page() to already
352 CL_PAGE_DEBUG(D_ERROR, env, pg, "Unexpecting page state %d.\n",
360 static int vvp_page_is_under_lock(const struct lu_env *env,
361 const struct cl_page_slice *slice,
362 struct cl_io *io, pgoff_t *max_index)
366 if (io->ci_type == CIT_READ || io->ci_type == CIT_WRITE ||
367 io->ci_type == CIT_FAULT) {
368 struct ccc_io *cio = ccc_env_io(env);
370 if (unlikely(cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED))
371 *max_index = CL_PAGE_EOF;
377 static int vvp_page_print(const struct lu_env *env,
378 const struct cl_page_slice *slice,
379 void *cookie, lu_printer_t printer)
381 struct ccc_page *vp = cl2ccc_page(slice);
382 struct page *vmpage = vp->cpg_page;
384 (*printer)(env, cookie, LUSTRE_VVP_NAME"-page@%p(%d:%d:%d) "
386 vp, vp->cpg_defer_uptodate, vp->cpg_ra_used,
387 vp->cpg_write_queued, vmpage);
388 if (vmpage != NULL) {
389 (*printer)(env, cookie, "%lx %d:%d %lx %lu %slru",
390 (long)vmpage->flags, page_count(vmpage),
391 page_mapcount(vmpage), vmpage->private,
393 list_empty(&vmpage->lru) ? "not-" : "");
395 (*printer)(env, cookie, "\n");
399 static const struct cl_page_operations vvp_page_ops = {
400 .cpo_own = vvp_page_own,
401 .cpo_assume = vvp_page_assume,
402 .cpo_unassume = vvp_page_unassume,
403 .cpo_disown = vvp_page_disown,
404 .cpo_discard = vvp_page_discard,
405 .cpo_delete = vvp_page_delete,
406 .cpo_export = vvp_page_export,
407 .cpo_is_vmlocked = vvp_page_is_vmlocked,
408 .cpo_fini = vvp_page_fini,
409 .cpo_print = vvp_page_print,
410 .cpo_is_under_lock = vvp_page_is_under_lock,
413 .cpo_prep = vvp_page_prep_read,
414 .cpo_completion = vvp_page_completion_read,
415 .cpo_make_ready = ccc_fail,
418 .cpo_prep = vvp_page_prep_write,
419 .cpo_completion = vvp_page_completion_write,
420 .cpo_make_ready = vvp_page_make_ready,
425 static void vvp_transient_page_verify(const struct cl_page *page)
429 static int vvp_transient_page_own(const struct lu_env *env,
430 const struct cl_page_slice *slice,
431 struct cl_io *unused, int nonblock)
433 vvp_transient_page_verify(slice->cpl_page);
437 static void vvp_transient_page_assume(const struct lu_env *env,
438 const struct cl_page_slice *slice,
439 struct cl_io *unused)
441 vvp_transient_page_verify(slice->cpl_page);
444 static void vvp_transient_page_unassume(const struct lu_env *env,
445 const struct cl_page_slice *slice,
446 struct cl_io *unused)
448 vvp_transient_page_verify(slice->cpl_page);
451 static void vvp_transient_page_disown(const struct lu_env *env,
452 const struct cl_page_slice *slice,
453 struct cl_io *unused)
455 vvp_transient_page_verify(slice->cpl_page);
458 static void vvp_transient_page_discard(const struct lu_env *env,
459 const struct cl_page_slice *slice,
460 struct cl_io *unused)
462 struct cl_page *page = slice->cpl_page;
464 vvp_transient_page_verify(slice->cpl_page);
467 * For transient pages, remove it from the radix tree.
469 cl_page_delete(env, page);
472 static int vvp_transient_page_is_vmlocked(const struct lu_env *env,
473 const struct cl_page_slice *slice)
475 struct inode *inode = ccc_object_inode(slice->cpl_obj);
478 locked = !mutex_trylock(&inode->i_mutex);
480 mutex_unlock(&inode->i_mutex);
481 return locked ? -EBUSY : -ENODATA;
485 vvp_transient_page_completion(const struct lu_env *env,
486 const struct cl_page_slice *slice,
489 vvp_transient_page_verify(slice->cpl_page);
492 static void vvp_transient_page_fini(const struct lu_env *env,
493 struct cl_page_slice *slice)
495 struct ccc_page *cp = cl2ccc_page(slice);
496 struct cl_page *clp = slice->cpl_page;
497 struct ccc_object *clobj = cl2ccc(clp->cp_obj);
499 vvp_page_fini_common(cp);
500 clobj->cob_transient_pages--;
503 static const struct cl_page_operations vvp_transient_page_ops = {
504 .cpo_own = vvp_transient_page_own,
505 .cpo_assume = vvp_transient_page_assume,
506 .cpo_unassume = vvp_transient_page_unassume,
507 .cpo_disown = vvp_transient_page_disown,
508 .cpo_discard = vvp_transient_page_discard,
509 .cpo_fini = vvp_transient_page_fini,
510 .cpo_is_vmlocked = vvp_transient_page_is_vmlocked,
511 .cpo_print = vvp_page_print,
512 .cpo_is_under_lock = vvp_page_is_under_lock,
515 .cpo_prep = ccc_transient_page_prep,
516 .cpo_completion = vvp_transient_page_completion,
519 .cpo_prep = ccc_transient_page_prep,
520 .cpo_completion = vvp_transient_page_completion,
525 int vvp_page_init(const struct lu_env *env, struct cl_object *obj,
526 struct cl_page *page, pgoff_t index)
528 struct ccc_page *cpg = cl_object_page_slice(obj, page);
529 struct page *vmpage = page->cp_vmpage;
531 CLOBINVRNT(env, obj, ccc_object_invariant(obj));
533 cpg->cpg_page = vmpage;
534 page_cache_get(vmpage);
536 INIT_LIST_HEAD(&cpg->cpg_pending_linkage);
537 if (page->cp_type == CPT_CACHEABLE) {
538 /* in cache, decref in vvp_page_delete */
539 atomic_inc(&page->cp_ref);
540 SetPagePrivate(vmpage);
541 vmpage->private = (unsigned long)page;
542 cl_page_slice_add(page, &cpg->cpg_cl, obj, index,
545 struct ccc_object *clobj = cl2ccc(obj);
547 cl_page_slice_add(page, &cpg->cpg_cl, obj, index,
548 &vvp_transient_page_ops);
549 clobj->cob_transient_pages++;