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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * Implementation of cl_page for VVP layer.
34 * Author: Nikita Danilov <nikita.danilov@sun.com>
35 * Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
38 #define DEBUG_SUBSYSTEM S_LLITE
40 #include <linux/atomic.h>
41 #include <linux/bitops.h>
43 #include <linux/mutex.h>
44 #include <linux/page-flags.h>
45 #include <linux/pagemap.h>
47 #include <libcfs/libcfs.h>
48 #include "llite_internal.h"
49 #include "vvp_internal.h"
51 /*****************************************************************************
57 static void vvp_page_fini_common(struct vvp_page *vpg)
59 struct page *vmpage = vpg->vpg_page;
61 LASSERT(vmpage != NULL);
65 static void vvp_page_fini(const struct lu_env *env,
66 struct cl_page_slice *slice)
68 struct vvp_page *vpg = cl2vvp_page(slice);
69 struct page *vmpage = vpg->vpg_page;
72 * vmpage->private was already cleared when page was moved into
75 LASSERT((struct cl_page *)vmpage->private != slice->cpl_page);
76 vvp_page_fini_common(vpg);
79 static int vvp_page_own(const struct lu_env *env,
80 const struct cl_page_slice *slice, struct cl_io *io,
83 struct vvp_page *vpg = cl2vvp_page(slice);
84 struct page *vmpage = vpg->vpg_page;
86 LASSERT(vmpage != NULL);
88 if (!trylock_page(vmpage))
91 if (unlikely(PageWriteback(vmpage))) {
100 wait_on_page_writeback(vmpage);
105 static void vvp_page_assume(const struct lu_env *env,
106 const struct cl_page_slice *slice,
107 struct cl_io *unused)
109 struct page *vmpage = cl2vm_page(slice);
111 LASSERT(vmpage != NULL);
112 LASSERT(PageLocked(vmpage));
113 wait_on_page_writeback(vmpage);
116 static void vvp_page_unassume(const struct lu_env *env,
117 const struct cl_page_slice *slice,
118 struct cl_io *unused)
120 struct page *vmpage = cl2vm_page(slice);
122 LASSERT(vmpage != NULL);
123 LASSERT(PageLocked(vmpage));
126 static void vvp_page_disown(const struct lu_env *env,
127 const struct cl_page_slice *slice, struct cl_io *io)
129 struct page *vmpage = cl2vm_page(slice);
131 LASSERT(vmpage != NULL);
132 LASSERT(PageLocked(vmpage));
134 unlock_page(cl2vm_page(slice));
137 static void vvp_page_discard(const struct lu_env *env,
138 const struct cl_page_slice *slice,
139 struct cl_io *unused)
141 struct page *vmpage = cl2vm_page(slice);
142 struct vvp_page *vpg = cl2vvp_page(slice);
144 LASSERT(vmpage != NULL);
145 LASSERT(PageLocked(vmpage));
147 if (vpg->vpg_defer_uptodate && !vpg->vpg_ra_used)
148 ll_ra_stats_inc(vmpage->mapping->host, RA_STAT_DISCARDED);
150 ll_invalidate_page(vmpage);
153 static void vvp_page_delete(const struct lu_env *env,
154 const struct cl_page_slice *slice)
156 struct page *vmpage = cl2vm_page(slice);
157 struct inode *inode = vmpage->mapping->host;
158 struct cl_object *obj = slice->cpl_obj;
159 struct cl_page *page = slice->cpl_page;
162 LASSERT(PageLocked(vmpage));
163 LASSERT((struct cl_page *)vmpage->private == page);
164 LASSERT(inode == vvp_object_inode(obj));
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);
223 * Handles page transfer errors at VM level.
225 * This takes inode as a separate argument, because inode on which error is to
226 * be set can be different from \a vmpage inode in case of direct-io.
228 static void vvp_vmpage_error(struct inode *inode, struct page *vmpage, int ioret)
230 struct vvp_object *obj = cl_inode2vvp(inode);
233 ClearPageError(vmpage);
234 obj->vob_discard_page_warned = 0;
236 SetPageError(vmpage);
237 if (ioret == -ENOSPC)
238 set_bit(AS_ENOSPC, &inode->i_mapping->flags);
240 set_bit(AS_EIO, &inode->i_mapping->flags);
242 if ((ioret == -ESHUTDOWN || ioret == -EINTR) &&
243 obj->vob_discard_page_warned == 0) {
244 obj->vob_discard_page_warned = 1;
245 ll_dirty_page_discard_warn(vmpage, ioret);
250 static void vvp_page_completion_read(const struct lu_env *env,
251 const struct cl_page_slice *slice,
254 struct vvp_page *vpg = cl2vvp_page(slice);
255 struct page *vmpage = vpg->vpg_page;
256 struct cl_page *page = slice->cpl_page;
257 struct inode *inode = vvp_object_inode(page->cp_obj);
260 LASSERT(PageLocked(vmpage));
261 CL_PAGE_HEADER(D_PAGE, env, page, "completing READ with %d\n", ioret);
263 if (vpg->vpg_defer_uptodate)
264 ll_ra_count_put(ll_i2sbi(inode), 1);
267 if (!vpg->vpg_defer_uptodate)
268 cl_page_export(env, page, 1);
270 vpg->vpg_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 vvp_page *vpg = cl2vvp_page(slice);
284 struct cl_page *pg = slice->cpl_page;
285 struct page *vmpage = vpg->vpg_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 * Only mark the page error only when it's an async write because
295 * applications won't wait for IO to finish.
297 if (pg->cp_sync_io == NULL)
298 vvp_vmpage_error(vvp_object_inode(pg->cp_obj), vmpage, ioret);
300 end_page_writeback(vmpage);
305 * Implements cl_page_operations::cpo_make_ready() method.
307 * This is called to yank a page from the transfer cache and to send it out as
308 * a part of transfer. This function try-locks the page. If try-lock failed,
309 * page is owned by some concurrent IO, and should be skipped (this is bad,
310 * but hopefully rare situation, as it usually results in transfer being
311 * shorter than possible).
313 * \retval 0 success, page can be placed into transfer
315 * \retval -EAGAIN page is either used by concurrent IO has been
316 * truncated. Skip it.
318 static int vvp_page_make_ready(const struct lu_env *env,
319 const struct cl_page_slice *slice)
321 struct page *vmpage = cl2vm_page(slice);
322 struct cl_page *pg = slice->cpl_page;
326 if (clear_page_dirty_for_io(vmpage)) {
327 LASSERT(pg->cp_state == CPS_CACHED);
328 /* This actually clears the dirty bit in the radix
330 set_page_writeback(vmpage);
331 CL_PAGE_HEADER(D_PAGE, env, pg, "readied\n");
332 } else if (pg->cp_state == CPS_PAGEOUT) {
333 /* is it possible for osc_flush_async_page() to already
337 CL_PAGE_DEBUG(D_ERROR, env, pg, "Unexpecting page state %d.\n",
345 static int vvp_page_print(const struct lu_env *env,
346 const struct cl_page_slice *slice,
347 void *cookie, lu_printer_t printer)
349 struct vvp_page *vpg = cl2vvp_page(slice);
350 struct page *vmpage = vpg->vpg_page;
352 (*printer)(env, cookie, LUSTRE_VVP_NAME"-page@%p(%d:%d) "
354 vpg, vpg->vpg_defer_uptodate, vpg->vpg_ra_used, vmpage);
356 if (vmpage != NULL) {
357 (*printer)(env, cookie, "%lx %d:%d %lx %lu %slru",
358 (long)vmpage->flags, page_count(vmpage),
359 page_mapcount(vmpage), vmpage->private,
361 list_empty(&vmpage->lru) ? "not-" : "");
364 (*printer)(env, cookie, "\n");
369 static int vvp_page_fail(const struct lu_env *env,
370 const struct cl_page_slice *slice)
380 static const struct cl_page_operations vvp_page_ops = {
381 .cpo_own = vvp_page_own,
382 .cpo_assume = vvp_page_assume,
383 .cpo_unassume = vvp_page_unassume,
384 .cpo_disown = vvp_page_disown,
385 .cpo_discard = vvp_page_discard,
386 .cpo_delete = vvp_page_delete,
387 .cpo_export = vvp_page_export,
388 .cpo_is_vmlocked = vvp_page_is_vmlocked,
389 .cpo_fini = vvp_page_fini,
390 .cpo_print = vvp_page_print,
393 .cpo_prep = vvp_page_prep_read,
394 .cpo_completion = vvp_page_completion_read,
395 .cpo_make_ready = vvp_page_fail,
398 .cpo_prep = vvp_page_prep_write,
399 .cpo_completion = vvp_page_completion_write,
400 .cpo_make_ready = vvp_page_make_ready,
405 static int vvp_transient_page_prep(const struct lu_env *env,
406 const struct cl_page_slice *slice,
407 struct cl_io *unused)
410 /* transient page should always be sent. */
414 static void vvp_transient_page_verify(const struct cl_page *page)
418 static int vvp_transient_page_own(const struct lu_env *env,
419 const struct cl_page_slice *slice,
420 struct cl_io *unused, int nonblock)
422 vvp_transient_page_verify(slice->cpl_page);
426 static void vvp_transient_page_assume(const struct lu_env *env,
427 const struct cl_page_slice *slice,
428 struct cl_io *unused)
430 vvp_transient_page_verify(slice->cpl_page);
433 static void vvp_transient_page_unassume(const struct lu_env *env,
434 const struct cl_page_slice *slice,
435 struct cl_io *unused)
437 vvp_transient_page_verify(slice->cpl_page);
440 static void vvp_transient_page_disown(const struct lu_env *env,
441 const struct cl_page_slice *slice,
442 struct cl_io *unused)
444 vvp_transient_page_verify(slice->cpl_page);
447 static void vvp_transient_page_discard(const struct lu_env *env,
448 const struct cl_page_slice *slice,
449 struct cl_io *unused)
451 struct cl_page *page = slice->cpl_page;
453 vvp_transient_page_verify(slice->cpl_page);
456 * For transient pages, remove it from the radix tree.
458 cl_page_delete(env, page);
461 static int vvp_transient_page_is_vmlocked(const struct lu_env *env,
462 const struct cl_page_slice *slice)
464 struct inode *inode = vvp_object_inode(slice->cpl_obj);
467 locked = !inode_trylock(inode);
470 return locked ? -EBUSY : -ENODATA;
474 vvp_transient_page_completion(const struct lu_env *env,
475 const struct cl_page_slice *slice,
478 vvp_transient_page_verify(slice->cpl_page);
481 static void vvp_transient_page_fini(const struct lu_env *env,
482 struct cl_page_slice *slice)
484 struct vvp_page *vpg = cl2vvp_page(slice);
485 struct cl_page *clp = slice->cpl_page;
486 struct vvp_object *clobj = cl2vvp(clp->cp_obj);
488 vvp_page_fini_common(vpg);
489 atomic_dec(&clobj->vob_transient_pages);
492 static const struct cl_page_operations vvp_transient_page_ops = {
493 .cpo_own = vvp_transient_page_own,
494 .cpo_assume = vvp_transient_page_assume,
495 .cpo_unassume = vvp_transient_page_unassume,
496 .cpo_disown = vvp_transient_page_disown,
497 .cpo_discard = vvp_transient_page_discard,
498 .cpo_fini = vvp_transient_page_fini,
499 .cpo_is_vmlocked = vvp_transient_page_is_vmlocked,
500 .cpo_print = vvp_page_print,
503 .cpo_prep = vvp_transient_page_prep,
504 .cpo_completion = vvp_transient_page_completion,
507 .cpo_prep = vvp_transient_page_prep,
508 .cpo_completion = vvp_transient_page_completion,
513 int vvp_page_init(const struct lu_env *env, struct cl_object *obj,
514 struct cl_page *page, pgoff_t index)
516 struct vvp_page *vpg = cl_object_page_slice(obj, page);
517 struct page *vmpage = page->cp_vmpage;
519 CLOBINVRNT(env, obj, vvp_object_invariant(obj));
521 vpg->vpg_page = vmpage;
524 if (page->cp_type == CPT_CACHEABLE) {
525 /* in cache, decref in vvp_page_delete */
526 atomic_inc(&page->cp_ref);
527 SetPagePrivate(vmpage);
528 vmpage->private = (unsigned long)page;
529 cl_page_slice_add(page, &vpg->vpg_cl, obj, index,
532 struct vvp_object *clobj = cl2vvp(obj);
534 cl_page_slice_add(page, &vpg->vpg_cl, obj, index,
535 &vvp_transient_page_ops);
536 atomic_inc(&clobj->vob_transient_pages);