1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
30 * Use is subject to license terms.
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>
41 #define DEBUG_SUBSYSTEM S_LLITE
44 # error This file is kernel only.
48 #include <lustre_lite.h>
50 #include "vvp_internal.h"
52 /*****************************************************************************
58 static void vvp_page_fini_common(struct ccc_page *cp)
60 cfs_page_t *vmpage = cp->cpg_page;
62 LASSERT(vmpage != NULL);
63 page_cache_release(vmpage);
64 OBD_SLAB_FREE_PTR(cp, vvp_page_kmem);
67 static void vvp_page_fini(const struct lu_env *env,
68 struct cl_page_slice *slice)
70 struct ccc_page *cp = cl2ccc_page(slice);
71 cfs_page_t *vmpage = cp->cpg_page;
74 * vmpage->private was already cleared when page was moved into
77 LASSERT((struct cl_page *)vmpage->private != slice->cpl_page);
78 vvp_page_fini_common(cp);
81 static void vvp_page_own(const struct lu_env *env,
82 const struct cl_page_slice *slice, struct cl_io *_)
84 struct ccc_page *vpg = cl2ccc_page(slice);
85 cfs_page_t *vmpage = vpg->cpg_page;
87 LASSERT(vmpage != NULL);
89 wait_on_page_writeback(vmpage);
92 static void vvp_page_assume(const struct lu_env *env,
93 const struct cl_page_slice *slice, struct cl_io *_)
95 cfs_page_t *vmpage = cl2vm_page(slice);
97 LASSERT(vmpage != NULL);
98 LASSERT(PageLocked(vmpage));
99 wait_on_page_writeback(vmpage);
102 static void vvp_page_unassume(const struct lu_env *env,
103 const struct cl_page_slice *slice,
106 cfs_page_t *vmpage = cl2vm_page(slice);
108 LASSERT(vmpage != NULL);
109 LASSERT(PageLocked(vmpage));
112 static void vvp_page_disown(const struct lu_env *env,
113 const struct cl_page_slice *slice, struct cl_io *io)
115 cfs_page_t *vmpage = cl2vm_page(slice);
117 LASSERT(vmpage != NULL);
118 LASSERT(PageLocked(vmpage));
120 unlock_page(cl2vm_page(slice));
123 static void vvp_page_discard(const struct lu_env *env,
124 const struct cl_page_slice *slice, struct cl_io *_)
126 cfs_page_t *vmpage = cl2vm_page(slice);
127 struct address_space *mapping = vmpage->mapping;
128 struct ccc_page *cpg = cl2ccc_page(slice);
130 LASSERT(vmpage != NULL);
131 LASSERT(PageLocked(vmpage));
133 if (cpg->cpg_defer_uptodate && !cpg->cpg_ra_used)
134 ll_ra_stats_inc(mapping, RA_STAT_DISCARDED);
137 * truncate_complete_page() calls
138 * a_ops->invalidatepage()->cl_page_delete()->vvp_page_delete().
140 truncate_complete_page(mapping, vmpage);
143 static int vvp_page_unmap(const struct lu_env *env,
144 const struct cl_page_slice *slice, struct cl_io *_)
146 cfs_page_t *vmpage = cl2vm_page(slice);
147 __u64 offset = vmpage->index << CFS_PAGE_SHIFT;
149 LASSERT(vmpage != NULL);
150 LASSERT(PageLocked(vmpage));
152 * XXX is it safe to call this with the page lock held?
154 ll_teardown_mmaps(vmpage->mapping, offset, offset + CFS_PAGE_SIZE);
158 static void vvp_page_delete(const struct lu_env *env,
159 const struct cl_page_slice *slice)
161 cfs_page_t *vmpage = cl2vm_page(slice);
162 struct inode *inode = vmpage->mapping->host;
163 struct cl_object *obj = slice->cpl_obj;
165 LASSERT(PageLocked(vmpage));
166 LASSERT((struct cl_page *)vmpage->private == slice->cpl_page);
167 LASSERT(inode == ccc_object_inode(obj));
169 vvp_write_complete(cl2ccc(obj), cl2ccc_page(slice));
170 ClearPagePrivate(vmpage);
173 * Reference from vmpage to cl_page is removed, but the reference back
174 * is still here. It is removed later in vvp_page_fini().
178 static void vvp_page_export(const struct lu_env *env,
179 const struct cl_page_slice *slice)
181 cfs_page_t *vmpage = cl2vm_page(slice);
183 LASSERT(vmpage != NULL);
184 LASSERT(PageLocked(vmpage));
185 SetPageUptodate(vmpage);
188 static int vvp_page_is_vmlocked(const struct lu_env *env,
189 const struct cl_page_slice *slice)
191 return PageLocked(cl2vm_page(slice)) ? -EBUSY : -ENODATA;
194 static int vvp_page_prep_read(const struct lu_env *env,
195 const struct cl_page_slice *slice,
199 /* Skip the page already marked as PG_uptodate. */
200 RETURN(PageUptodate(cl2vm_page(slice)) ? -EALREADY : 0);
203 static int vvp_page_prep_write(const struct lu_env *env,
204 const struct cl_page_slice *slice,
207 cfs_page_t *vmpage = cl2vm_page(slice);
210 if (clear_page_dirty_for_io(vmpage)) {
211 set_page_writeback(vmpage);
219 * Handles page transfer errors at VM level.
221 * This takes inode as a separate argument, because inode on which error is to
222 * be set can be different from \a vmpage inode in case of direct-io.
224 static void vvp_vmpage_error(struct inode *inode, cfs_page_t *vmpage, int ioret)
227 ClearPageError(vmpage);
228 else if (ioret != -EINTR) {
229 SetPageError(vmpage);
230 if (ioret == -ENOSPC)
231 set_bit(AS_ENOSPC, &inode->i_mapping->flags);
233 set_bit(AS_EIO, &inode->i_mapping->flags);
237 static void vvp_page_completion_common(const struct lu_env *env,
238 struct ccc_page *cp, int ioret)
240 struct cl_page *clp = cp->cpg_cl.cpl_page;
241 cfs_page_t *vmpage = cp->cpg_page;
242 struct inode *inode = ccc_object_inode(clp->cp_obj);
243 struct cl_sync_io *anchor = cp->cpg_sync_io;
245 LINVRNT(cl_page_is_vmlocked(env, clp));
247 /* Don't assert the page writeback bit here because the lustre file
248 * may be as a backend of swap space. in this case, the page writeback
249 * is set by VM, and obvious we shouldn't clear it at all. Fortunately
250 * this type of pages are all TRANSIENT pages. */
251 KLASSERT(ergo(clp->cp_type == CPT_CACHEABLE, !PageWriteback(vmpage)));
253 vvp_vmpage_error(inode, vmpage, ioret);
255 if (anchor != NULL) {
256 cp->cpg_sync_io = NULL;
257 cl_sync_io_note(anchor, ioret);
258 } else if (clp->cp_type == CPT_CACHEABLE)
262 static void vvp_page_completion_read(const struct lu_env *env,
263 const struct cl_page_slice *slice,
266 struct ccc_page *cp = cl2ccc_page(slice);
267 struct cl_page *page = cl_page_top(slice->cpl_page);
268 struct inode *inode = ccc_object_inode(page->cp_obj);
271 CL_PAGE_HEADER(D_PAGE, env, page, "completing READ with %d\n", ioret);
273 if (cp->cpg_defer_uptodate)
274 ll_ra_count_put(ll_i2sbi(inode), 1);
277 /* XXX: do we need this for transient pages? */
278 if (!cp->cpg_defer_uptodate)
279 cl_page_export(env, page);
281 cp->cpg_defer_uptodate = 0;
282 vvp_page_completion_common(env, cp, ioret);
287 static void vvp_page_completion_write_common(const struct lu_env *env,
288 const struct cl_page_slice *slice,
291 struct ccc_page *cp = cl2ccc_page(slice);
294 cp->cpg_write_queued = 0;
296 * Only ioret == 0, write succeed, then this page could be
297 * deleted from the pending_writing count.
299 vvp_write_complete(cl2ccc(slice->cpl_obj), cp);
301 vvp_page_completion_common(env, cp, ioret);
304 static void vvp_page_completion_write(const struct lu_env *env,
305 const struct cl_page_slice *slice,
308 struct ccc_page *cp = cl2ccc_page(slice);
309 struct cl_page *pg = slice->cpl_page;
310 cfs_page_t *vmpage = cp->cpg_page;
314 LINVRNT(cl_page_is_vmlocked(env, pg));
315 LASSERT(PageWriteback(vmpage));
317 CL_PAGE_HEADER(D_PAGE, env, pg, "completing WRITE with %d\n", ioret);
319 end_page_writeback(vmpage);
320 LASSERT(!PageWriteback(vmpage));
322 vvp_page_completion_write_common(env, slice, ioret);
327 * Implements cl_page_operations::cpo_make_ready() method.
329 * This is called to yank a page from the transfer cache and to send it out as
330 * a part of transfer. This function try-locks the page. If try-lock failed,
331 * page is owned by some concurrent IO, and should be skipped (this is bad,
332 * but hopefully rare situation, as it usually results in transfer being
333 * shorter than possible).
335 * \retval 0 success, page can be placed into transfer
337 * \retval -EAGAIN page is either used by concurrent IO has been
338 * truncated. Skip it.
340 static int vvp_page_make_ready(const struct lu_env *env,
341 const struct cl_page_slice *slice)
343 cfs_page_t *vmpage = cl2vm_page(slice);
344 struct cl_page *pg = slice->cpl_page;
348 /* we're trying to write, but the page is locked.. come back later */
349 if (!TestSetPageLocked(vmpage)) {
350 if (pg->cp_state == CPS_CACHED) {
352 * We can cancel IO if page wasn't dirty after all.
354 clear_page_dirty_for_io(vmpage);
356 * This actually clears the dirty bit in the radix
359 set_page_writeback(vmpage);
361 CL_PAGE_HEADER(D_PAGE, env, pg, "readied\n");
365 * Page was concurrently truncated.
367 LASSERT(pg->cp_state == CPS_FREEING);
372 static int vvp_page_print(const struct lu_env *env,
373 const struct cl_page_slice *slice,
374 void *cookie, lu_printer_t printer)
376 struct ccc_page *vp = cl2ccc_page(slice);
377 cfs_page_t *vmpage = vp->cpg_page;
379 (*printer)(env, cookie, LUSTRE_VVP_NAME"-page@%p(%d:%d:%d) "
381 vp, vp->cpg_defer_uptodate, vp->cpg_ra_used,
382 vp->cpg_write_queued, vmpage);
383 if (vmpage != NULL) {
384 (*printer)(env, cookie, "%lx %d:%d %lx %lu %slru",
385 (long)vmpage->flags, page_count(vmpage),
386 page_mapcount(vmpage), vmpage->private,
388 list_empty(&vmpage->lru) ? "not-" : "");
390 (*printer)(env, cookie, "\n");
394 static const struct cl_page_operations vvp_page_ops = {
395 .cpo_own = vvp_page_own,
396 .cpo_assume = vvp_page_assume,
397 .cpo_unassume = vvp_page_unassume,
398 .cpo_disown = vvp_page_disown,
399 .cpo_vmpage = ccc_page_vmpage,
400 .cpo_discard = vvp_page_discard,
401 .cpo_delete = vvp_page_delete,
402 .cpo_unmap = vvp_page_unmap,
403 .cpo_export = vvp_page_export,
404 .cpo_is_vmlocked = vvp_page_is_vmlocked,
405 .cpo_fini = vvp_page_fini,
406 .cpo_print = vvp_page_print,
407 .cpo_is_under_lock = ccc_page_is_under_lock,
410 .cpo_prep = vvp_page_prep_read,
411 .cpo_completion = vvp_page_completion_read,
412 .cpo_make_ready = ccc_fail,
415 .cpo_prep = vvp_page_prep_write,
416 .cpo_completion = vvp_page_completion_write,
417 .cpo_make_ready = vvp_page_make_ready,
422 static void vvp_transient_page_verify(const struct cl_page *page)
424 struct inode *inode = ccc_object_inode(page->cp_obj);
426 LASSERT(!TRYLOCK_INODE_MUTEX(inode));
427 /* LASSERT_SEM_LOCKED(&inode->i_alloc_sem); */
430 static void vvp_transient_page_own(const struct lu_env *env,
431 const struct cl_page_slice *slice,
434 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,
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,
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,
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,
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 = !TRYLOCK_INODE_MUTEX(inode);
480 UNLOCK_INODE_MUTEX(inode);
481 return locked ? -EBUSY : -ENODATA;
485 vvp_transient_page_completion_write(const struct lu_env *env,
486 const struct cl_page_slice *slice,
489 vvp_transient_page_verify(slice->cpl_page);
490 vvp_page_completion_write_common(env, slice, ioret);
494 static void vvp_transient_page_fini(const struct lu_env *env,
495 struct cl_page_slice *slice)
497 struct ccc_page *cp = cl2ccc_page(slice);
498 struct cl_page *clp = slice->cpl_page;
499 struct ccc_object *clobj = cl2ccc(clp->cp_obj);
501 vvp_page_fini_common(cp);
502 LASSERT(!TRYLOCK_INODE_MUTEX(clobj->cob_inode));
503 clobj->cob_transient_pages--;
506 static const struct cl_page_operations vvp_transient_page_ops = {
507 .cpo_own = vvp_transient_page_own,
508 .cpo_assume = vvp_transient_page_assume,
509 .cpo_unassume = vvp_transient_page_unassume,
510 .cpo_disown = vvp_transient_page_disown,
511 .cpo_discard = vvp_transient_page_discard,
512 .cpo_vmpage = ccc_page_vmpage,
513 .cpo_fini = vvp_transient_page_fini,
514 .cpo_is_vmlocked = vvp_transient_page_is_vmlocked,
515 .cpo_print = vvp_page_print,
516 .cpo_is_under_lock = ccc_page_is_under_lock,
519 .cpo_prep = ccc_transient_page_prep,
520 .cpo_completion = vvp_page_completion_read,
523 .cpo_prep = ccc_transient_page_prep,
524 .cpo_completion = vvp_transient_page_completion_write,
529 struct cl_page *vvp_page_init(const struct lu_env *env, struct cl_object *obj,
530 struct cl_page *page, cfs_page_t *vmpage)
532 struct ccc_page *cpg;
535 CLOBINVRNT(env, obj, ccc_object_invariant(obj));
537 OBD_SLAB_ALLOC_PTR_GFP(cpg, vvp_page_kmem, CFS_ALLOC_IO);
539 cpg->cpg_page = vmpage;
540 page_cache_get(vmpage);
542 CFS_INIT_LIST_HEAD(&cpg->cpg_pending_linkage);
543 if (page->cp_type == CPT_CACHEABLE) {
544 SetPagePrivate(vmpage);
545 vmpage->private = (unsigned long)page;
546 cl_page_slice_add(page, &cpg->cpg_cl, obj,
549 struct ccc_object *clobj = cl2ccc(obj);
551 LASSERT(!TRYLOCK_INODE_MUTEX(clobj->cob_inode));
552 cl_page_slice_add(page, &cpg->cpg_cl, obj,
553 &vvp_transient_page_ops);
554 clobj->cob_transient_pages++;
559 return ERR_PTR(result);