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 "llite_internal.h"
48 #include "vvp_internal.h"
50 /*****************************************************************************
56 static void vvp_page_fini_common(struct vvp_page *vpg)
58 struct page *vmpage = vpg->vpg_page;
64 static void vvp_page_fini(const struct lu_env *env,
65 struct cl_page_slice *slice)
67 struct vvp_page *vpg = cl2vvp_page(slice);
68 struct page *vmpage = vpg->vpg_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(vpg);
78 static int vvp_page_own(const struct lu_env *env,
79 const struct cl_page_slice *slice, struct cl_io *io,
82 struct vvp_page *vpg = cl2vvp_page(slice);
83 struct page *vmpage = vpg->vpg_page;
87 if (!trylock_page(vmpage))
90 if (unlikely(PageWriteback(vmpage))) {
99 wait_on_page_writeback(vmpage);
104 static void vvp_page_assume(const struct lu_env *env,
105 const struct cl_page_slice *slice,
106 struct cl_io *unused)
108 struct page *vmpage = cl2vm_page(slice);
111 LASSERT(PageLocked(vmpage));
112 wait_on_page_writeback(vmpage);
115 static void vvp_page_unassume(const struct lu_env *env,
116 const struct cl_page_slice *slice,
117 struct cl_io *unused)
119 struct page *vmpage = cl2vm_page(slice);
122 LASSERT(PageLocked(vmpage));
125 static void vvp_page_disown(const struct lu_env *env,
126 const struct cl_page_slice *slice, struct cl_io *io)
128 struct page *vmpage = cl2vm_page(slice);
131 LASSERT(PageLocked(vmpage));
133 unlock_page(cl2vm_page(slice));
136 static void vvp_page_discard(const struct lu_env *env,
137 const struct cl_page_slice *slice,
138 struct cl_io *unused)
140 struct page *vmpage = cl2vm_page(slice);
141 struct vvp_page *vpg = cl2vvp_page(slice);
144 LASSERT(PageLocked(vmpage));
146 if (vpg->vpg_defer_uptodate && !vpg->vpg_ra_used)
147 ll_ra_stats_inc(vmpage->mapping->host, RA_STAT_DISCARDED);
149 ll_invalidate_page(vmpage);
152 static void vvp_page_delete(const struct lu_env *env,
153 const struct cl_page_slice *slice)
155 struct page *vmpage = cl2vm_page(slice);
156 struct inode *inode = vmpage->mapping->host;
157 struct cl_object *obj = slice->cpl_obj;
158 struct cl_page *page = slice->cpl_page;
161 LASSERT(PageLocked(vmpage));
162 LASSERT((struct cl_page *)vmpage->private == page);
163 LASSERT(inode == vvp_object_inode(obj));
165 /* Drop the reference count held in vvp_page_init */
166 refc = atomic_dec_return(&page->cp_ref);
167 LASSERTF(refc >= 1, "page = %p, refc = %d\n", page, refc);
169 ClearPagePrivate(vmpage);
172 * Reference from vmpage to cl_page is removed, but the reference back
173 * is still here. It is removed later in vvp_page_fini().
177 static void vvp_page_export(const struct lu_env *env,
178 const struct cl_page_slice *slice,
181 struct page *vmpage = cl2vm_page(slice);
184 LASSERT(PageLocked(vmpage));
186 SetPageUptodate(vmpage);
188 ClearPageUptodate(vmpage);
191 static int vvp_page_is_vmlocked(const struct lu_env *env,
192 const struct cl_page_slice *slice)
194 return PageLocked(cl2vm_page(slice)) ? -EBUSY : -ENODATA;
197 static int vvp_page_prep_read(const struct lu_env *env,
198 const struct cl_page_slice *slice,
199 struct cl_io *unused)
201 /* Skip the page already marked as PG_uptodate. */
202 return PageUptodate(cl2vm_page(slice)) ? -EALREADY : 0;
205 static int vvp_page_prep_write(const struct lu_env *env,
206 const struct cl_page_slice *slice,
207 struct cl_io *unused)
209 struct page *vmpage = cl2vm_page(slice);
210 struct cl_page *pg = slice->cpl_page;
212 LASSERT(PageLocked(vmpage));
213 LASSERT(!PageDirty(vmpage));
215 /* ll_writepage path is not a sync write, so need to set page writeback
219 set_page_writeback(vmpage);
225 * Handles page transfer errors at VM level.
227 * This takes inode as a separate argument, because inode on which error is to
228 * be set can be different from \a vmpage inode in case of direct-io.
230 static void vvp_vmpage_error(struct inode *inode, struct page *vmpage,
233 struct vvp_object *obj = cl_inode2vvp(inode);
236 ClearPageError(vmpage);
237 obj->vob_discard_page_warned = 0;
239 SetPageError(vmpage);
240 mapping_set_error(inode->i_mapping, ioret);
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);
259 LASSERT(PageLocked(vmpage));
260 CL_PAGE_HEADER(D_PAGE, env, page, "completing READ with %d\n", ioret);
262 if (vpg->vpg_defer_uptodate)
263 ll_ra_count_put(ll_i2sbi(inode), 1);
266 if (!vpg->vpg_defer_uptodate)
267 cl_page_export(env, page, 1);
269 vpg->vpg_defer_uptodate = 0;
272 if (!page->cp_sync_io)
276 static void vvp_page_completion_write(const struct lu_env *env,
277 const struct cl_page_slice *slice,
280 struct vvp_page *vpg = cl2vvp_page(slice);
281 struct cl_page *pg = slice->cpl_page;
282 struct page *vmpage = vpg->vpg_page;
284 CL_PAGE_HEADER(D_PAGE, env, pg, "completing WRITE with %d\n", ioret);
286 if (pg->cp_sync_io) {
287 LASSERT(PageLocked(vmpage));
288 LASSERT(!PageWriteback(vmpage));
290 LASSERT(PageWriteback(vmpage));
292 * Only mark the page error only when it's an async write
293 * because applications won't wait for IO to finish.
295 vvp_vmpage_error(vvp_object_inode(pg->cp_obj), vmpage, ioret);
297 end_page_writeback(vmpage);
302 * Implements cl_page_operations::cpo_make_ready() method.
304 * This is called to yank a page from the transfer cache and to send it out as
305 * a part of transfer. This function try-locks the page. If try-lock failed,
306 * page is owned by some concurrent IO, and should be skipped (this is bad,
307 * but hopefully rare situation, as it usually results in transfer being
308 * shorter than possible).
310 * \retval 0 success, page can be placed into transfer
312 * \retval -EAGAIN page is either used by concurrent IO has been
313 * truncated. Skip it.
315 static int vvp_page_make_ready(const struct lu_env *env,
316 const struct cl_page_slice *slice)
318 struct page *vmpage = cl2vm_page(slice);
319 struct cl_page *pg = slice->cpl_page;
323 if (clear_page_dirty_for_io(vmpage)) {
324 LASSERT(pg->cp_state == CPS_CACHED);
325 /* This actually clears the dirty bit in the radix tree. */
326 set_page_writeback(vmpage);
327 CL_PAGE_HEADER(D_PAGE, env, pg, "readied\n");
328 } else if (pg->cp_state == CPS_PAGEOUT) {
329 /* is it possible for osc_flush_async_page() to already
334 CL_PAGE_DEBUG(D_ERROR, env, pg, "Unexpecting page state %d.\n",
342 static int vvp_page_print(const struct lu_env *env,
343 const struct cl_page_slice *slice,
344 void *cookie, lu_printer_t printer)
346 struct vvp_page *vpg = cl2vvp_page(slice);
347 struct page *vmpage = vpg->vpg_page;
349 (*printer)(env, cookie, LUSTRE_VVP_NAME "-page@%p(%d:%d) vm@%p ",
350 vpg, vpg->vpg_defer_uptodate, vpg->vpg_ra_used, vmpage);
352 (*printer)(env, cookie, "%lx %d:%d %lx %lu %slru",
353 (long)vmpage->flags, page_count(vmpage),
354 page_mapcount(vmpage), vmpage->private,
356 list_empty(&vmpage->lru) ? "not-" : "");
359 (*printer)(env, cookie, "\n");
364 static int vvp_page_fail(const struct lu_env *env,
365 const struct cl_page_slice *slice)
375 static const struct cl_page_operations vvp_page_ops = {
376 .cpo_own = vvp_page_own,
377 .cpo_assume = vvp_page_assume,
378 .cpo_unassume = vvp_page_unassume,
379 .cpo_disown = vvp_page_disown,
380 .cpo_discard = vvp_page_discard,
381 .cpo_delete = vvp_page_delete,
382 .cpo_export = vvp_page_export,
383 .cpo_is_vmlocked = vvp_page_is_vmlocked,
384 .cpo_fini = vvp_page_fini,
385 .cpo_print = vvp_page_print,
388 .cpo_prep = vvp_page_prep_read,
389 .cpo_completion = vvp_page_completion_read,
390 .cpo_make_ready = vvp_page_fail,
393 .cpo_prep = vvp_page_prep_write,
394 .cpo_completion = vvp_page_completion_write,
395 .cpo_make_ready = vvp_page_make_ready,
400 static int vvp_transient_page_prep(const struct lu_env *env,
401 const struct cl_page_slice *slice,
402 struct cl_io *unused)
404 /* transient page should always be sent. */
408 static int vvp_transient_page_own(const struct lu_env *env,
409 const struct cl_page_slice *slice,
410 struct cl_io *unused, int nonblock)
415 static void vvp_transient_page_assume(const struct lu_env *env,
416 const struct cl_page_slice *slice,
417 struct cl_io *unused)
421 static void vvp_transient_page_unassume(const struct lu_env *env,
422 const struct cl_page_slice *slice,
423 struct cl_io *unused)
427 static void vvp_transient_page_disown(const struct lu_env *env,
428 const struct cl_page_slice *slice,
429 struct cl_io *unused)
433 static void vvp_transient_page_discard(const struct lu_env *env,
434 const struct cl_page_slice *slice,
435 struct cl_io *unused)
437 struct cl_page *page = slice->cpl_page;
440 * For transient pages, remove it from the radix tree.
442 cl_page_delete(env, page);
445 static int vvp_transient_page_is_vmlocked(const struct lu_env *env,
446 const struct cl_page_slice *slice)
448 struct inode *inode = vvp_object_inode(slice->cpl_obj);
451 locked = !inode_trylock(inode);
454 return locked ? -EBUSY : -ENODATA;
458 vvp_transient_page_completion(const struct lu_env *env,
459 const struct cl_page_slice *slice,
464 static void vvp_transient_page_fini(const struct lu_env *env,
465 struct cl_page_slice *slice)
467 struct vvp_page *vpg = cl2vvp_page(slice);
468 struct cl_page *clp = slice->cpl_page;
469 struct vvp_object *clobj = cl2vvp(clp->cp_obj);
471 vvp_page_fini_common(vpg);
472 atomic_dec(&clobj->vob_transient_pages);
475 static const struct cl_page_operations vvp_transient_page_ops = {
476 .cpo_own = vvp_transient_page_own,
477 .cpo_assume = vvp_transient_page_assume,
478 .cpo_unassume = vvp_transient_page_unassume,
479 .cpo_disown = vvp_transient_page_disown,
480 .cpo_discard = vvp_transient_page_discard,
481 .cpo_fini = vvp_transient_page_fini,
482 .cpo_is_vmlocked = vvp_transient_page_is_vmlocked,
483 .cpo_print = vvp_page_print,
486 .cpo_prep = vvp_transient_page_prep,
487 .cpo_completion = vvp_transient_page_completion,
490 .cpo_prep = vvp_transient_page_prep,
491 .cpo_completion = vvp_transient_page_completion,
496 int vvp_page_init(const struct lu_env *env, struct cl_object *obj,
497 struct cl_page *page, pgoff_t index)
499 struct vvp_page *vpg = cl_object_page_slice(obj, page);
500 struct page *vmpage = page->cp_vmpage;
502 CLOBINVRNT(env, obj, vvp_object_invariant(obj));
504 vpg->vpg_page = vmpage;
507 if (page->cp_type == CPT_CACHEABLE) {
508 /* in cache, decref in vvp_page_delete */
509 atomic_inc(&page->cp_ref);
510 SetPagePrivate(vmpage);
511 vmpage->private = (unsigned long)page;
512 cl_page_slice_add(page, &vpg->vpg_cl, obj, index,
515 struct vvp_object *clobj = cl2vvp(obj);
517 cl_page_slice_add(page, &vpg->vpg_cl, obj, index,
518 &vvp_transient_page_ops);
519 atomic_inc(&clobj->vob_transient_pages);