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;
88 LASSERT(vmpage != NULL);
90 /* DEBUG CODE FOR #18881 */
91 while (TestSetPageLocked(vmpage)) {
92 cfs_schedule_timeout(CFS_TASK_INTERRUPTIBLE,
93 cfs_time_seconds(1)/10);
95 CL_PAGE_DEBUG(D_ERROR, env,
96 cl_page_top(slice->cpl_page),
97 "XXX page %p blocked on acquiring the"
98 " lock. process %s/%p, flags %lx,io %p\n",
99 vmpage, current->comm, current,
101 libcfs_debug_dumpstack(NULL);
102 LCONSOLE_WARN("Reproduced bug #18881,please contact:"
103 "jay <jinshan.xiong@sun.com>, thanks\n");
111 /* lock_page(vmpage); */
112 wait_on_page_writeback(vmpage);
115 static void vvp_page_assume(const struct lu_env *env,
116 const struct cl_page_slice *slice, struct cl_io *_)
118 cfs_page_t *vmpage = cl2vm_page(slice);
120 LASSERT(vmpage != NULL);
121 LASSERT(PageLocked(vmpage));
122 wait_on_page_writeback(vmpage);
125 static void vvp_page_unassume(const struct lu_env *env,
126 const struct cl_page_slice *slice,
129 cfs_page_t *vmpage = cl2vm_page(slice);
131 LASSERT(vmpage != NULL);
132 LASSERT(PageLocked(vmpage));
135 static void vvp_page_disown(const struct lu_env *env,
136 const struct cl_page_slice *slice, struct cl_io *io)
138 cfs_page_t *vmpage = cl2vm_page(slice);
140 LASSERT(vmpage != NULL);
141 LASSERT(PageLocked(vmpage));
143 unlock_page(cl2vm_page(slice));
146 static void vvp_page_discard(const struct lu_env *env,
147 const struct cl_page_slice *slice, struct cl_io *_)
149 cfs_page_t *vmpage = cl2vm_page(slice);
150 struct address_space *mapping = vmpage->mapping;
151 struct ccc_page *cpg = cl2ccc_page(slice);
153 LASSERT(vmpage != NULL);
154 LASSERT(PageLocked(vmpage));
156 if (cpg->cpg_defer_uptodate && !cpg->cpg_ra_used)
157 ll_ra_stats_inc(mapping, RA_STAT_DISCARDED);
160 * truncate_complete_page() calls
161 * a_ops->invalidatepage()->cl_page_delete()->vvp_page_delete().
163 truncate_complete_page(mapping, vmpage);
166 static int vvp_page_unmap(const struct lu_env *env,
167 const struct cl_page_slice *slice, struct cl_io *_)
169 cfs_page_t *vmpage = cl2vm_page(slice);
170 __u64 offset = vmpage->index << CFS_PAGE_SHIFT;
172 LASSERT(vmpage != NULL);
173 LASSERT(PageLocked(vmpage));
175 * XXX is it safe to call this with the page lock held?
177 ll_teardown_mmaps(vmpage->mapping, offset, offset + CFS_PAGE_SIZE);
181 static void vvp_page_delete(const struct lu_env *env,
182 const struct cl_page_slice *slice)
184 cfs_page_t *vmpage = cl2vm_page(slice);
185 struct inode *inode = vmpage->mapping->host;
186 struct cl_object *obj = slice->cpl_obj;
188 LASSERT(PageLocked(vmpage));
189 LASSERT((struct cl_page *)vmpage->private == slice->cpl_page);
190 LASSERT(inode == ccc_object_inode(obj));
192 vvp_write_complete(cl2ccc(obj), cl2ccc_page(slice));
193 ClearPagePrivate(vmpage);
196 * Reference from vmpage to cl_page is removed, but the reference back
197 * is still here. It is removed later in vvp_page_fini().
201 static void vvp_page_export(const struct lu_env *env,
202 const struct cl_page_slice *slice)
204 cfs_page_t *vmpage = cl2vm_page(slice);
206 LASSERT(vmpage != NULL);
207 LASSERT(PageLocked(vmpage));
208 SetPageUptodate(vmpage);
211 static int vvp_page_is_vmlocked(const struct lu_env *env,
212 const struct cl_page_slice *slice)
214 return PageLocked(cl2vm_page(slice)) ? -EBUSY : -ENODATA;
217 static int vvp_page_prep_read(const struct lu_env *env,
218 const struct cl_page_slice *slice,
222 /* Skip the page already marked as PG_uptodate. */
223 RETURN(PageUptodate(cl2vm_page(slice)) ? -EALREADY : 0);
226 static int vvp_page_prep_write(const struct lu_env *env,
227 const struct cl_page_slice *slice,
230 cfs_page_t *vmpage = cl2vm_page(slice);
233 if (clear_page_dirty_for_io(vmpage)) {
234 set_page_writeback(vmpage);
235 vvp_write_pending(cl2ccc(slice->cpl_obj), cl2ccc_page(slice));
243 * Handles page transfer errors at VM level.
245 * This takes inode as a separate argument, because inode on which error is to
246 * be set can be different from \a vmpage inode in case of direct-io.
248 static void vvp_vmpage_error(struct inode *inode, cfs_page_t *vmpage, int ioret)
251 ClearPageError(vmpage);
252 else if (ioret != -EINTR) {
253 SetPageError(vmpage);
254 if (ioret == -ENOSPC)
255 set_bit(AS_ENOSPC, &inode->i_mapping->flags);
257 set_bit(AS_EIO, &inode->i_mapping->flags);
261 static void vvp_page_completion_common(const struct lu_env *env,
262 struct ccc_page *cp, int ioret)
264 struct cl_page *clp = cp->cpg_cl.cpl_page;
265 cfs_page_t *vmpage = cp->cpg_page;
266 struct inode *inode = ccc_object_inode(clp->cp_obj);
267 struct cl_sync_io *anchor = cp->cpg_sync_io;
269 LINVRNT(cl_page_is_vmlocked(env, clp));
271 if (anchor != NULL) {
272 cp->cpg_sync_io = NULL;
273 cl_sync_io_note(anchor, ioret);
274 } else if (clp->cp_type == CPT_CACHEABLE) {
276 * Only mark the page error only when it's a cacheable page
279 * For sync IO and direct IO(CPT_TRANSIENT), the error is able
280 * to be seen by application, so we don't need to mark a page
283 vvp_vmpage_error(inode, vmpage, ioret);
288 static void vvp_page_completion_read(const struct lu_env *env,
289 const struct cl_page_slice *slice,
292 struct ccc_page *cp = cl2ccc_page(slice);
293 struct cl_page *page = cl_page_top(slice->cpl_page);
294 struct inode *inode = ccc_object_inode(page->cp_obj);
297 CL_PAGE_HEADER(D_PAGE, env, page, "completing READ with %d\n", ioret);
299 if (cp->cpg_defer_uptodate)
300 ll_ra_count_put(ll_i2sbi(inode), 1);
303 /* XXX: do we need this for transient pages? */
304 if (!cp->cpg_defer_uptodate)
305 cl_page_export(env, page);
307 cp->cpg_defer_uptodate = 0;
308 vvp_page_completion_common(env, cp, ioret);
313 static void vvp_page_completion_write_common(const struct lu_env *env,
314 const struct cl_page_slice *slice,
317 struct ccc_page *cp = cl2ccc_page(slice);
320 * TODO: Actually it makes sense to add the page into oap pending
321 * list again and so that we don't need to take the page out from
322 * SoM write pending list, if we just meet a recoverable error,
324 * To implement this, we just need to return a non zero value in
325 * ->cpo_completion method. The underlying transfer should be notified
326 * and then re-add the page into pending transfer queue. -jay
328 cp->cpg_write_queued = 0;
329 vvp_write_complete(cl2ccc(slice->cpl_obj), cp);
331 vvp_page_completion_common(env, cp, ioret);
334 static void vvp_page_completion_write(const struct lu_env *env,
335 const struct cl_page_slice *slice,
338 struct ccc_page *cp = cl2ccc_page(slice);
339 struct cl_page *pg = slice->cpl_page;
340 cfs_page_t *vmpage = cp->cpg_page;
344 LINVRNT(cl_page_is_vmlocked(env, pg));
345 LASSERT(PageWriteback(vmpage));
347 CL_PAGE_HEADER(D_PAGE, env, pg, "completing WRITE with %d\n", ioret);
349 vvp_page_completion_write_common(env, slice, ioret);
350 end_page_writeback(vmpage);
355 * Implements cl_page_operations::cpo_make_ready() method.
357 * This is called to yank a page from the transfer cache and to send it out as
358 * a part of transfer. This function try-locks the page. If try-lock failed,
359 * page is owned by some concurrent IO, and should be skipped (this is bad,
360 * but hopefully rare situation, as it usually results in transfer being
361 * shorter than possible).
363 * \retval 0 success, page can be placed into transfer
365 * \retval -EAGAIN page is either used by concurrent IO has been
366 * truncated. Skip it.
368 static int vvp_page_make_ready(const struct lu_env *env,
369 const struct cl_page_slice *slice)
371 cfs_page_t *vmpage = cl2vm_page(slice);
372 struct cl_page *pg = slice->cpl_page;
376 /* we're trying to write, but the page is locked.. come back later */
377 if (!TestSetPageLocked(vmpage)) {
378 if (pg->cp_state == CPS_CACHED) {
380 * We can cancel IO if page wasn't dirty after all.
382 clear_page_dirty_for_io(vmpage);
384 * This actually clears the dirty bit in the radix
387 set_page_writeback(vmpage);
388 vvp_write_pending(cl2ccc(slice->cpl_obj),
390 CL_PAGE_HEADER(D_PAGE, env, pg, "readied\n");
394 * Page was concurrently truncated.
396 LASSERT(pg->cp_state == CPS_FREEING);
401 static int vvp_page_print(const struct lu_env *env,
402 const struct cl_page_slice *slice,
403 void *cookie, lu_printer_t printer)
405 struct ccc_page *vp = cl2ccc_page(slice);
406 cfs_page_t *vmpage = vp->cpg_page;
408 (*printer)(env, cookie, LUSTRE_VVP_NAME"-page@%p(%d:%d:%d) "
410 vp, vp->cpg_defer_uptodate, vp->cpg_ra_used,
411 vp->cpg_write_queued, vmpage);
412 if (vmpage != NULL) {
413 (*printer)(env, cookie, "%lx %d:%d %lx %lu %slru",
414 (long)vmpage->flags, page_count(vmpage),
415 page_mapcount(vmpage), vmpage->private,
417 list_empty(&vmpage->lru) ? "not-" : "");
419 (*printer)(env, cookie, "\n");
423 static const struct cl_page_operations vvp_page_ops = {
424 .cpo_own = vvp_page_own,
425 .cpo_assume = vvp_page_assume,
426 .cpo_unassume = vvp_page_unassume,
427 .cpo_disown = vvp_page_disown,
428 .cpo_vmpage = ccc_page_vmpage,
429 .cpo_discard = vvp_page_discard,
430 .cpo_delete = vvp_page_delete,
431 .cpo_unmap = vvp_page_unmap,
432 .cpo_export = vvp_page_export,
433 .cpo_is_vmlocked = vvp_page_is_vmlocked,
434 .cpo_fini = vvp_page_fini,
435 .cpo_print = vvp_page_print,
436 .cpo_is_under_lock = ccc_page_is_under_lock,
439 .cpo_prep = vvp_page_prep_read,
440 .cpo_completion = vvp_page_completion_read,
441 .cpo_make_ready = ccc_fail,
444 .cpo_prep = vvp_page_prep_write,
445 .cpo_completion = vvp_page_completion_write,
446 .cpo_make_ready = vvp_page_make_ready,
451 static void vvp_transient_page_verify(const struct cl_page *page)
453 struct inode *inode = ccc_object_inode(page->cp_obj);
455 LASSERT(!TRYLOCK_INODE_MUTEX(inode));
456 /* LASSERT_SEM_LOCKED(&inode->i_alloc_sem); */
459 static void vvp_transient_page_own(const struct lu_env *env,
460 const struct cl_page_slice *slice,
463 vvp_transient_page_verify(slice->cpl_page);
466 static void vvp_transient_page_assume(const struct lu_env *env,
467 const struct cl_page_slice *slice,
470 vvp_transient_page_verify(slice->cpl_page);
473 static void vvp_transient_page_unassume(const struct lu_env *env,
474 const struct cl_page_slice *slice,
477 vvp_transient_page_verify(slice->cpl_page);
480 static void vvp_transient_page_disown(const struct lu_env *env,
481 const struct cl_page_slice *slice,
484 vvp_transient_page_verify(slice->cpl_page);
487 static void vvp_transient_page_discard(const struct lu_env *env,
488 const struct cl_page_slice *slice,
491 struct cl_page *page = slice->cpl_page;
493 vvp_transient_page_verify(slice->cpl_page);
496 * For transient pages, remove it from the radix tree.
498 cl_page_delete(env, page);
501 static int vvp_transient_page_is_vmlocked(const struct lu_env *env,
502 const struct cl_page_slice *slice)
504 struct inode *inode = ccc_object_inode(slice->cpl_obj);
507 locked = !TRYLOCK_INODE_MUTEX(inode);
509 UNLOCK_INODE_MUTEX(inode);
510 return locked ? -EBUSY : -ENODATA;
514 vvp_transient_page_completion_write(const struct lu_env *env,
515 const struct cl_page_slice *slice,
518 vvp_transient_page_verify(slice->cpl_page);
519 vvp_page_completion_write_common(env, slice, ioret);
523 static void vvp_transient_page_fini(const struct lu_env *env,
524 struct cl_page_slice *slice)
526 struct ccc_page *cp = cl2ccc_page(slice);
527 struct cl_page *clp = slice->cpl_page;
528 struct ccc_object *clobj = cl2ccc(clp->cp_obj);
530 vvp_page_fini_common(cp);
531 LASSERT(!TRYLOCK_INODE_MUTEX(clobj->cob_inode));
532 clobj->cob_transient_pages--;
535 static const struct cl_page_operations vvp_transient_page_ops = {
536 .cpo_own = vvp_transient_page_own,
537 .cpo_assume = vvp_transient_page_assume,
538 .cpo_unassume = vvp_transient_page_unassume,
539 .cpo_disown = vvp_transient_page_disown,
540 .cpo_discard = vvp_transient_page_discard,
541 .cpo_vmpage = ccc_page_vmpage,
542 .cpo_fini = vvp_transient_page_fini,
543 .cpo_is_vmlocked = vvp_transient_page_is_vmlocked,
544 .cpo_print = vvp_page_print,
545 .cpo_is_under_lock = ccc_page_is_under_lock,
548 .cpo_prep = ccc_transient_page_prep,
549 .cpo_completion = vvp_page_completion_read,
552 .cpo_prep = ccc_transient_page_prep,
553 .cpo_completion = vvp_transient_page_completion_write,
558 struct cl_page *vvp_page_init(const struct lu_env *env, struct cl_object *obj,
559 struct cl_page *page, cfs_page_t *vmpage)
561 struct ccc_page *cpg;
564 CLOBINVRNT(env, obj, ccc_object_invariant(obj));
566 OBD_SLAB_ALLOC_PTR_GFP(cpg, vvp_page_kmem, CFS_ALLOC_IO);
568 cpg->cpg_page = vmpage;
569 page_cache_get(vmpage);
571 CFS_INIT_LIST_HEAD(&cpg->cpg_pending_linkage);
572 if (page->cp_type == CPT_CACHEABLE) {
573 SetPagePrivate(vmpage);
574 vmpage->private = (unsigned long)page;
575 cl_page_slice_add(page, &cpg->cpg_cl, obj,
578 struct ccc_object *clobj = cl2ccc(obj);
580 LASSERT(!TRYLOCK_INODE_MUTEX(clobj->cob_inode));
581 cl_page_slice_add(page, &cpg->cpg_cl, obj,
582 &vvp_transient_page_ops);
583 clobj->cob_transient_pages++;
588 return ERR_PTR(result);