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, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * Implementation of cl_page for VVP layer.
33 * Author: Nikita Danilov <nikita.danilov@sun.com>
34 * Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
37 #define DEBUG_SUBSYSTEM S_LLITE
39 #include <linux/atomic.h>
40 #include <linux/bitops.h>
42 #include <linux/mutex.h>
43 #include <linux/page-flags.h>
44 #include <linux/pagemap.h>
46 #include <libcfs/libcfs.h>
47 #include "llite_internal.h"
48 #include "vvp_internal.h"
50 /*****************************************************************************
56 static void vvp_page_fini_common(struct vvp_page *vpg, struct pagevec *pvec)
58 struct page *vmpage = vpg->vpg_page;
60 LASSERT(vmpage != NULL);
62 if (!pagevec_add(pvec, vmpage))
63 pagevec_release(pvec);
69 static void vvp_page_fini(const struct lu_env *env,
70 struct cl_page_slice *slice,
73 struct vvp_page *vpg = cl2vvp_page(slice);
74 struct page *vmpage = vpg->vpg_page;
77 * vmpage->private was already cleared when page was moved into
80 LASSERT((struct cl_page *)vmpage->private != slice->cpl_page);
81 vvp_page_fini_common(vpg, pvec);
84 static int vvp_page_own(const struct lu_env *env,
85 const struct cl_page_slice *slice, struct cl_io *io,
88 struct vvp_page *vpg = cl2vvp_page(slice);
89 struct page *vmpage = vpg->vpg_page;
93 LASSERT(vmpage != NULL);
95 if (!trylock_page(vmpage))
98 if (unlikely(PageWriteback(vmpage))) {
107 wait_on_page_writeback(vmpage);
112 static void vvp_page_assume(const struct lu_env *env,
113 const struct cl_page_slice *slice,
114 struct cl_io *unused)
116 struct page *vmpage = cl2vm_page(slice);
118 LASSERT(vmpage != NULL);
119 LASSERT(PageLocked(vmpage));
120 wait_on_page_writeback(vmpage);
123 static void vvp_page_unassume(const struct lu_env *env,
124 const struct cl_page_slice *slice,
125 struct cl_io *unused)
127 struct page *vmpage = cl2vm_page(slice);
129 LASSERT(vmpage != NULL);
130 LASSERT(PageLocked(vmpage));
133 static void vvp_page_disown(const struct lu_env *env,
134 const struct cl_page_slice *slice, struct cl_io *io)
136 struct page *vmpage = cl2vm_page(slice);
140 LASSERT(vmpage != NULL);
141 LASSERT(PageLocked(vmpage));
143 unlock_page(cl2vm_page(slice));
148 static void vvp_page_discard(const struct lu_env *env,
149 const struct cl_page_slice *slice,
150 struct cl_io *unused)
152 struct page *vmpage = cl2vm_page(slice);
153 struct vvp_page *vpg = cl2vvp_page(slice);
155 LASSERT(vmpage != NULL);
156 LASSERT(PageLocked(vmpage));
158 if (vpg->vpg_defer_uptodate && !vpg->vpg_ra_used && vmpage->mapping)
159 ll_ra_stats_inc(vmpage->mapping->host, RA_STAT_DISCARDED);
161 generic_error_remove_page(vmpage->mapping, vmpage);
164 static void vvp_page_delete(const struct lu_env *env,
165 const struct cl_page_slice *slice)
167 struct page *vmpage = cl2vm_page(slice);
168 struct cl_page *page = slice->cpl_page;
171 LASSERT(PageLocked(vmpage));
172 LASSERT((struct cl_page *)vmpage->private == page);
175 /* Drop the reference count held in vvp_page_init */
176 refc = atomic_dec_return(&page->cp_ref);
177 LASSERTF(refc >= 1, "page = %p, refc = %d\n", page, refc);
179 ClearPagePrivate(vmpage);
183 * Vmpage might not be released due page refcount != 2,
184 * clear Page uptodate here to avoid stale data.
186 ClearPageUptodate(vmpage);
188 * Reference from vmpage to cl_page is removed, but the reference back
189 * is still here. It is removed later in vvp_page_fini().
193 static void vvp_page_export(const struct lu_env *env,
194 const struct cl_page_slice *slice,
197 struct page *vmpage = cl2vm_page(slice);
199 LASSERT(vmpage != NULL);
200 LASSERT(PageLocked(vmpage));
202 SetPageUptodate(vmpage);
204 ClearPageUptodate(vmpage);
207 static int vvp_page_is_vmlocked(const struct lu_env *env,
208 const struct cl_page_slice *slice)
210 return PageLocked(cl2vm_page(slice)) ? -EBUSY : -ENODATA;
213 static int vvp_page_prep_read(const struct lu_env *env,
214 const struct cl_page_slice *slice,
215 struct cl_io *unused)
218 /* Skip the page already marked as PG_uptodate. */
219 RETURN(PageUptodate(cl2vm_page(slice)) ? -EALREADY : 0);
222 static int vvp_page_prep_write(const struct lu_env *env,
223 const struct cl_page_slice *slice,
224 struct cl_io *unused)
226 struct page *vmpage = cl2vm_page(slice);
227 struct cl_page *pg = slice->cpl_page;
229 LASSERT(PageLocked(vmpage));
230 LASSERT(!PageDirty(vmpage));
232 /* ll_writepage path is not a sync write, so need to set page writeback
235 if (pg->cp_sync_io == NULL)
236 set_page_writeback(vmpage);
242 * Handles page transfer errors at VM level.
244 * This takes inode as a separate argument, because inode on which error is to
245 * be set can be different from \a vmpage inode in case of direct-io.
247 static void vvp_vmpage_error(struct inode *inode, struct page *vmpage,
250 struct vvp_object *obj = cl_inode2vvp(inode);
253 ClearPageError(vmpage);
254 obj->vob_discard_page_warned = 0;
256 SetPageError(vmpage);
257 if (ioret == -ENOSPC)
258 set_bit(AS_ENOSPC, &inode->i_mapping->flags);
260 set_bit(AS_EIO, &inode->i_mapping->flags);
262 if ((ioret == -ESHUTDOWN || ioret == -EINTR ||
263 ioret == -EIO) && obj->vob_discard_page_warned == 0) {
264 obj->vob_discard_page_warned = 1;
265 ll_dirty_page_discard_warn(vmpage, ioret);
270 static void vvp_page_completion_read(const struct lu_env *env,
271 const struct cl_page_slice *slice,
274 struct vvp_page *vpg = cl2vvp_page(slice);
275 struct page *vmpage = vpg->vpg_page;
276 struct cl_page *page = slice->cpl_page;
277 struct inode *inode = vvp_object_inode(page->cp_obj);
280 LASSERT(PageLocked(vmpage));
281 CL_PAGE_HEADER(D_PAGE, env, page, "completing READ with %d\n", ioret);
283 if (vpg->vpg_defer_uptodate)
284 ll_ra_count_put(ll_i2sbi(inode), 1);
287 if (!vpg->vpg_defer_uptodate)
288 cl_page_export(env, page, 1);
289 } else if (vpg->vpg_defer_uptodate) {
290 vpg->vpg_defer_uptodate = 0;
291 if (ioret == -EAGAIN) {
292 /* mirror read failed, it needs to destroy the page
293 * because subpage would be from wrong osc when trying
294 * to read from a new mirror
296 generic_error_remove_page(vmpage->mapping, vmpage);
300 if (page->cp_sync_io == NULL)
306 static void vvp_page_completion_write(const struct lu_env *env,
307 const struct cl_page_slice *slice,
310 struct vvp_page *vpg = cl2vvp_page(slice);
311 struct cl_page *pg = slice->cpl_page;
312 struct page *vmpage = vpg->vpg_page;
315 CL_PAGE_HEADER(D_PAGE, env, pg, "completing WRITE with %d\n", ioret);
317 if (pg->cp_sync_io != NULL) {
318 LASSERT(PageLocked(vmpage));
319 LASSERT(!PageWriteback(vmpage));
321 LASSERT(PageWriteback(vmpage));
323 * Only mark the page error only when it's an async write
324 * because applications won't wait for IO to finish.
326 vvp_vmpage_error(vvp_object_inode(pg->cp_obj), vmpage, ioret);
328 end_page_writeback(vmpage);
334 * Implements cl_page_operations::cpo_make_ready() method.
336 * This is called to yank a page from the transfer cache and to send it out as
337 * a part of transfer. This function try-locks the page. If try-lock failed,
338 * page is owned by some concurrent IO, and should be skipped (this is bad,
339 * but hopefully rare situation, as it usually results in transfer being
340 * shorter than possible).
342 * \retval 0 success, page can be placed into transfer
344 * \retval -EAGAIN page is either used by concurrent IO has been
345 * truncated. Skip it.
347 static int vvp_page_make_ready(const struct lu_env *env,
348 const struct cl_page_slice *slice)
350 struct page *vmpage = cl2vm_page(slice);
351 struct cl_page *pg = slice->cpl_page;
355 if (clear_page_dirty_for_io(vmpage)) {
356 LASSERT(pg->cp_state == CPS_CACHED);
357 /* This actually clears the dirty bit in the radix
360 set_page_writeback(vmpage);
361 CL_PAGE_HEADER(D_PAGE, env, pg, "readied\n");
362 } else if (pg->cp_state == CPS_PAGEOUT) {
363 /* is it possible for osc_flush_async_page() to already
368 CL_PAGE_DEBUG(D_ERROR, env, pg, "Unexpecting page state %d.\n",
376 static int vvp_page_print(const struct lu_env *env,
377 const struct cl_page_slice *slice,
378 void *cookie, lu_printer_t printer)
380 struct vvp_page *vpg = cl2vvp_page(slice);
381 struct page *vmpage = vpg->vpg_page;
383 (*printer)(env, cookie,
384 LUSTRE_VVP_NAME"-page@%p(%d:%d) vm@%p ",
385 vpg, vpg->vpg_defer_uptodate, vpg->vpg_ra_used, vmpage);
387 if (vmpage != NULL) {
388 (*printer)(env, cookie, "%lx %d:%d %lx %lu %slru",
389 (long)vmpage->flags, page_count(vmpage),
390 page_mapcount(vmpage), vmpage->private,
392 list_empty(&vmpage->lru) ? "not-" : "");
395 (*printer)(env, cookie, "\n");
400 static int vvp_page_fail(const struct lu_env *env,
401 const struct cl_page_slice *slice)
411 static const struct cl_page_operations vvp_page_ops = {
412 .cpo_own = vvp_page_own,
413 .cpo_assume = vvp_page_assume,
414 .cpo_unassume = vvp_page_unassume,
415 .cpo_disown = vvp_page_disown,
416 .cpo_discard = vvp_page_discard,
417 .cpo_delete = vvp_page_delete,
418 .cpo_export = vvp_page_export,
419 .cpo_is_vmlocked = vvp_page_is_vmlocked,
420 .cpo_fini = vvp_page_fini,
421 .cpo_print = vvp_page_print,
424 .cpo_prep = vvp_page_prep_read,
425 .cpo_completion = vvp_page_completion_read,
426 .cpo_make_ready = vvp_page_fail,
429 .cpo_prep = vvp_page_prep_write,
430 .cpo_completion = vvp_page_completion_write,
431 .cpo_make_ready = vvp_page_make_ready,
436 static void vvp_transient_page_discard(const struct lu_env *env,
437 const struct cl_page_slice *slice,
438 struct cl_io *unused)
440 struct cl_page *page = slice->cpl_page;
443 * For transient pages, remove it from the radix tree.
445 cl_page_delete(env, page);
448 static int vvp_transient_page_is_vmlocked(const struct lu_env *env,
449 const struct cl_page_slice *slice)
454 static void vvp_transient_page_fini(const struct lu_env *env,
455 struct cl_page_slice *slice,
456 struct pagevec *pvec)
458 struct vvp_page *vpg = cl2vvp_page(slice);
460 vvp_page_fini_common(vpg, pvec);
463 static const struct cl_page_operations vvp_transient_page_ops = {
464 .cpo_discard = vvp_transient_page_discard,
465 .cpo_fini = vvp_transient_page_fini,
466 .cpo_is_vmlocked = vvp_transient_page_is_vmlocked,
467 .cpo_print = vvp_page_print,
470 int vvp_page_init(const struct lu_env *env, struct cl_object *obj,
471 struct cl_page *page, pgoff_t index)
473 struct vvp_page *vpg = cl_object_page_slice(obj, page);
474 struct page *vmpage = page->cp_vmpage;
476 CLOBINVRNT(env, obj, vvp_object_invariant(obj));
478 vpg->vpg_page = vmpage;
481 if (page->cp_type == CPT_TRANSIENT) {
482 /* DIO pages are referenced by userspace, we don't need to take
483 * a reference on them. (contrast with get_page() call above)
485 cl_page_slice_add(page, &vpg->vpg_cl, obj,
486 &vvp_transient_page_ops);
488 /* in cache, decref in vvp_page_delete */
489 atomic_inc(&page->cp_ref);
490 SetPagePrivate(vmpage);
491 vmpage->private = (unsigned long)page;
492 cl_page_slice_add(page, &vpg->vpg_cl, obj,