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 * lustre/llite/llite_lib.c
38 * Lustre Light Super operations
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include <linux/module.h>
44 #include <linux/types.h>
45 #include <linux/random.h>
46 #include <linux/version.h>
48 #include <lustre_lite.h>
49 #include <lustre_ha.h>
50 #include <lustre_dlm.h>
51 #include <lprocfs_status.h>
52 #include <lustre_disk.h>
53 #include <lustre_param.h>
54 #include <lustre_cache.h>
55 #include "llite_internal.h"
57 cfs_mem_cache_t *ll_file_data_slab;
59 LIST_HEAD(ll_super_blocks);
60 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
62 extern struct address_space_operations ll_aops;
63 extern struct address_space_operations ll_dir_aops;
66 #define log2(n) ffz(~(n))
69 static inline void ll_pglist_fini(struct ll_sb_info *sbi)
74 if (sbi->ll_pglist == NULL)
77 for_each_possible_cpu(i) {
78 page = sbi->ll_pglist[i]->llpd_page;
80 sbi->ll_pglist[i] = NULL;
85 OBD_FREE(sbi->ll_pglist, sizeof(void *)*num_possible_cpus());
86 sbi->ll_pglist = NULL;
89 static inline int ll_pglist_init(struct ll_sb_info *sbi)
91 struct ll_pglist_data *pd;
96 OBD_ALLOC(sbi->ll_pglist, sizeof(void *) * num_possible_cpus());
97 if (sbi->ll_pglist == NULL)
100 budget = sbi->ll_async_page_max / num_online_cpus();
101 for_each_possible_cpu(i) {
102 struct page *page = alloc_pages_node(cpu_to_node(i),
109 if (color + L1_CACHE_ALIGN(sizeof(*pd)) > PAGE_SIZE)
112 pd = (struct ll_pglist_data *)(page_address(page) + color);
113 memset(pd, 0, sizeof(*pd));
114 spin_lock_init(&pd->llpd_lock);
115 INIT_LIST_HEAD(&pd->llpd_list);
117 pd->llpd_budget = budget;
119 pd->llpd_page = page;
120 atomic_set(&pd->llpd_sample_count, 0);
121 sbi->ll_pglist[i] = pd;
122 color += L1_CACHE_ALIGN(sizeof(*pd));
128 static struct ll_sb_info *ll_init_sbi(void)
130 struct ll_sb_info *sbi = NULL;
137 OBD_ALLOC(sbi, sizeof(*sbi));
141 OBD_ALLOC(sbi->ll_async_page_sample, sizeof(long)*num_possible_cpus());
142 if (sbi->ll_async_page_sample == NULL)
145 spin_lock_init(&sbi->ll_lock);
146 spin_lock_init(&sbi->ll_lco.lco_lock);
147 spin_lock_init(&sbi->ll_pp_extent_lock);
148 spin_lock_init(&sbi->ll_process_lock);
149 sbi->ll_rw_stats_on = 0;
152 pages = si.totalram - si.totalhigh;
153 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
154 #ifdef HAVE_BGL_SUPPORT
155 sbi->ll_async_page_max = pages / 4;
157 sbi->ll_async_page_max = pages / 2;
160 sbi->ll_async_page_max = (pages / 4) * 3;
163 lcounter_init(&sbi->ll_async_page_count);
164 spin_lock_init(&sbi->ll_async_page_reblnc_lock);
165 sbi->ll_async_page_sample_max = 64 * num_online_cpus();
166 sbi->ll_async_page_reblnc_count = 0;
167 sbi->ll_async_page_clock_hand = 0;
168 if (ll_pglist_init(sbi))
171 sbi->ll_ra_info.ra_max_pages = min(pages / 32,
172 SBI_DEFAULT_READAHEAD_MAX);
173 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
174 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
175 sbi->ll_contention_time = SBI_DEFAULT_CONTENTION_SECONDS;
176 sbi->ll_lockless_truncate_enable = SBI_DEFAULT_LOCKLESS_TRUNCATE_ENABLE;
177 INIT_LIST_HEAD(&sbi->ll_conn_chain);
178 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
180 ll_generate_random_uuid(uuid);
181 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
182 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
184 spin_lock(&ll_sb_lock);
185 list_add_tail(&sbi->ll_list, &ll_super_blocks);
186 spin_unlock(&ll_sb_lock);
188 #ifdef ENABLE_CHECKSUM
189 sbi->ll_flags |= LL_SBI_DATA_CHECKSUM;
191 #ifdef ENABLE_LLITE_CHECKSUM
192 sbi->ll_flags |= LL_SBI_LLITE_CHECKSUM;
195 #ifdef HAVE_LRU_RESIZE_SUPPORT
196 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
199 #ifdef HAVE_EXPORT___IGET
200 INIT_LIST_HEAD(&sbi->ll_deathrow);
201 spin_lock_init(&sbi->ll_deathrow_lock);
203 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
204 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
205 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
208 /* metadata statahead is enabled by default */
209 sbi->ll_sa_max = LL_SA_RPC_DEF;
214 if (sbi->ll_async_page_sample)
215 OBD_FREE(sbi->ll_async_page_sample,
216 sizeof(long) * num_possible_cpus());
218 OBD_FREE(sbi, sizeof(*sbi));
222 void ll_free_sbi(struct super_block *sb)
224 struct ll_sb_info *sbi = ll_s2sbi(sb);
229 spin_lock(&ll_sb_lock);
230 list_del(&sbi->ll_list);
231 spin_unlock(&ll_sb_lock);
232 lcounter_destroy(&sbi->ll_async_page_count);
233 OBD_FREE(sbi->ll_async_page_sample,
234 sizeof(long) * num_possible_cpus());
235 OBD_FREE(sbi, sizeof(*sbi));
240 static struct dentry_operations ll_d_root_ops = {
241 #ifdef DCACHE_LUSTRE_INVALID
242 .d_compare = ll_dcompare,
246 static int client_common_fill_super(struct super_block *sb,
247 char *mdc, char *osc)
249 struct inode *root = 0;
250 struct ll_sb_info *sbi = ll_s2sbi(sb);
251 struct obd_device *obd;
252 struct ll_fid rootfid;
253 struct obd_statfs osfs;
254 struct ptlrpc_request *request = NULL;
255 struct lustre_handle osc_conn = {0, };
256 struct lustre_handle mdc_conn = {0, };
258 struct obd_connect_data *data = NULL;
262 obd = class_name2obd(mdc);
264 CERROR("MDC %s: not setup or attached\n", mdc);
268 OBD_ALLOC(data, sizeof(*data));
272 if (proc_lustre_fs_root) {
273 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
276 CERROR("could not register mount in /proc/fs/lustre\n");
279 /* indicate the features supported by this client */
280 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_IBITS |
281 OBD_CONNECT_JOIN | OBD_CONNECT_ATTRFID |
282 OBD_CONNECT_NODEVOH | OBD_CONNECT_CANCELSET |
283 OBD_CONNECT_AT | OBD_CONNECT_FID |
284 OBD_CONNECT_VBR | OBD_CONNECT_LOV_V3;
285 #ifdef HAVE_LRU_RESIZE_SUPPORT
286 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
287 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
289 #ifdef CONFIG_FS_POSIX_ACL
290 data->ocd_connect_flags |= OBD_CONNECT_ACL;
292 data->ocd_ibits_known = MDS_INODELOCK_FULL;
293 data->ocd_version = LUSTRE_VERSION_CODE;
295 if (sb->s_flags & MS_RDONLY)
296 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
297 if (sbi->ll_flags & LL_SBI_USER_XATTR)
298 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
300 #ifdef HAVE_MS_FLOCK_LOCK
301 /* force vfs to use lustre handler for flock() calls - bug 10743 */
302 sb->s_flags |= MS_FLOCK_LOCK;
305 if (sbi->ll_flags & LL_SBI_FLOCK)
306 sbi->ll_fop = &ll_file_operations_flock;
307 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
308 sbi->ll_fop = &ll_file_operations;
310 sbi->ll_fop = &ll_file_operations_noflock;
312 err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid, data, NULL);
314 LCONSOLE_ERROR_MSG(0x14f, "An MDT (mdc %s) is performing "
315 "recovery, of which this client is not a "
316 "part. Please wait for recovery to complete,"
317 " abort, or time out.\n", mdc);
320 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
323 sbi->ll_mdc_exp = class_conn2export(&mdc_conn);
324 err = obd_fid_init(sbi->ll_mdc_exp);
328 err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
332 /* MDC connect is surely finished by now because we actually sent
333 * a statfs RPC, otherwise obd_connect() is asynchronous. */
334 *data = class_exp2cliimp(sbi->ll_mdc_exp)->imp_connect_data;
336 LASSERT(osfs.os_bsize);
337 sb->s_blocksize = osfs.os_bsize;
338 sb->s_blocksize_bits = log2(osfs.os_bsize);
339 sb->s_magic = LL_SUPER_MAGIC;
341 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
342 * retval = in_file->f_op->sendfile(...);
344 * retval = -EOVERFLOW;
346 * it will check if *ppos is greater than max. However, max equals to
347 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
348 * has been defined as a signed long long ineger in linux kernel. */
349 #if BITS_PER_LONG == 64
350 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
352 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
354 sbi->ll_namelen = osfs.os_namelen;
355 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
357 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
358 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
359 LCONSOLE_INFO("Disabling user_xattr feature because "
360 "it is not supported on the server\n");
361 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
364 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
366 sb->s_flags |= MS_POSIXACL;
368 sbi->ll_flags |= LL_SBI_ACL;
370 sbi->ll_flags &= ~LL_SBI_ACL;
372 if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
373 sbi->ll_flags |= LL_SBI_JOIN;
375 obd = class_name2obd(osc);
377 CERROR("OSC %s: not setup or attached\n", osc);
378 GOTO(out_mdc, err = -ENODEV);
381 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_GRANT |
382 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
383 OBD_CONNECT_SRVLOCK | OBD_CONNECT_CANCELSET|
384 OBD_CONNECT_AT | OBD_CONNECT_FID |
385 OBD_CONNECT_VBR | OBD_CONNECT_TRUNCLOCK;
387 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
388 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
389 * disabled by default, because it can still be enabled on the
390 * fly via /proc. As a consequence, we still need to come to an
391 * agreement on the supported algorithms at connect time */
392 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
394 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
395 data->ocd_cksum_types = OBD_CKSUM_ADLER;
397 /* send the list of supported checksum types */
398 data->ocd_cksum_types = OBD_CKSUM_ALL;
401 #ifdef HAVE_LRU_RESIZE_SUPPORT
402 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
403 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
406 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
407 "ocd_grant: %d\n", data->ocd_connect_flags,
408 data->ocd_version, data->ocd_grant);
410 obd->obd_upcall.onu_owner = &sbi->ll_lco;
411 obd->obd_upcall.onu_upcall = ll_ocd_update;
412 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
414 err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid, data, NULL);
416 LCONSOLE_ERROR_MSG(0x150, "An OST (osc %s) is performing "
417 "recovery, of which this client is not a "
418 "part. Please wait for recovery to "
419 "complete, abort, or time out.\n", osc);
422 CERROR("cannot connect to %s: rc = %d\n", osc, err);
425 sbi->ll_osc_exp = class_conn2export(&osc_conn);
426 spin_lock(&sbi->ll_lco.lco_lock);
427 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
428 sbi->ll_lco.lco_mdc_exp = sbi->ll_mdc_exp;
429 sbi->ll_lco.lco_osc_exp = sbi->ll_osc_exp;
430 spin_unlock(&sbi->ll_lco.lco_lock);
432 err = obd_register_page_removal_cb(sbi->ll_osc_exp,
436 CERROR("cannot register page removal callback: rc = %d\n",err);
439 err = obd_register_lock_cancel_cb(sbi->ll_osc_exp,
440 ll_extent_lock_cancel_cb);
442 CERROR("cannot register lock cancel callback: rc = %d\n", err);
443 GOTO(out_page_rm_cb, err);
446 err = mdc_init_ea_size(sbi->ll_mdc_exp, sbi->ll_osc_exp);
448 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
449 GOTO(out_lock_cn_cb, err);
452 err = obd_prep_async_page(sbi->ll_osc_exp, NULL, NULL, NULL,
453 0, NULL, NULL, NULL, 0, NULL);
455 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
456 "filesystem. There must be at least one "
457 "active OST for a client to start.\n");
458 GOTO(out_lock_cn_cb, err);
461 if (!ll_async_page_slab) {
462 ll_async_page_slab_size =
463 size_round(sizeof(struct ll_async_page)) + err;
464 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
465 ll_async_page_slab_size,
467 if (!ll_async_page_slab)
468 GOTO(out_lock_cn_cb, -ENOMEM);
471 err = mdc_getstatus(sbi->ll_mdc_exp, &rootfid);
473 CERROR("cannot mds_connect: rc = %d\n", err);
474 GOTO(out_lock_cn_cb, err);
476 CDEBUG(D_SUPER, "rootfid "LPU64":"DFID"\n", rootfid.id,
477 PFID((struct lu_fid*)&rootfid));
478 sbi->ll_rootino = rootfid.id;
480 sb->s_op = &lustre_super_operations;
481 #if THREAD_SIZE >= 8192
482 /* Disable the NFS export because of stack overflow
483 * when THREAD_SIZE < 8192. Please refer to 17630. */
484 sb->s_export_op = &lustre_export_operations;
488 * XXX: move this to after cbd setup? */
489 err = mdc_getattr(sbi->ll_mdc_exp, &rootfid,
490 OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS |
491 (sbi->ll_flags & LL_SBI_ACL ? OBD_MD_FLACL : 0),
494 CERROR("mdc_getattr failed for root: rc = %d\n", err);
495 GOTO(out_lock_cn_cb, err);
498 err = mdc_req2lustre_md(request, REPLY_REC_OFF, sbi->ll_osc_exp, &md);
500 CERROR("failed to understand root inode md: rc = %d\n",err);
501 ptlrpc_req_finished (request);
502 GOTO(out_lock_cn_cb, err);
505 LASSERT(sbi->ll_rootino != 0);
506 root = ll_iget(sb, ll_fid_build_ino(sbi, &rootfid), &md);
508 ptlrpc_req_finished(request);
510 if (root == NULL || is_bad_inode(root)) {
511 mdc_free_lustre_md(sbi->ll_osc_exp, &md);
512 CERROR("lustre_lite: bad iget4 for root\n");
513 GOTO(out_root, err = -EBADF);
516 err = ll_close_thread_start(&sbi->ll_lcq);
518 CERROR("cannot start close thread: rc %d\n", err);
522 checksum = sbi->ll_flags & LL_SBI_DATA_CHECKSUM;
523 err = obd_set_info_async(sbi->ll_osc_exp, sizeof(KEY_CHECKSUM),
524 KEY_CHECKSUM, sizeof(checksum),
527 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
528 backing dev info assigned to inode mapping is used for
529 determining maximal readahead. */
530 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
531 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
532 /* bug 2805 - set VM readahead to zero */
533 vm_max_readahead = vm_min_readahead = 0;
536 sb->s_root = d_alloc_root(root);
538 OBD_FREE(data, sizeof(*data));
539 sb->s_root->d_op = &ll_d_root_ops;
541 sbi->ll_sdev_orig = sb->s_dev;
542 /* We set sb->s_dev equal on all lustre clients in order to support
543 * NFS export clustering. NFSD requires that the FSID be the same
545 /* s_dev is also used in lt_compare() to compare two fs, but that is
546 * only a node-local comparison. */
547 sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
548 strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
556 obd_unregister_lock_cancel_cb(sbi->ll_osc_exp,
557 ll_extent_lock_cancel_cb);
559 obd_unregister_page_removal_cb(sbi->ll_osc_exp,
562 obd_disconnect(sbi->ll_osc_exp);
563 sbi->ll_osc_exp = NULL;
565 obd_fid_fini(sbi->ll_mdc_exp);
566 obd_disconnect(sbi->ll_mdc_exp);
567 sbi->ll_mdc_exp = NULL;
570 OBD_FREE(data, sizeof(*data));
571 lprocfs_unregister_mountpoint(sbi);
575 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
579 *lmmsize = obd_size_diskmd(sbi->ll_osc_exp, NULL);
581 rc = obd_get_info(sbi->ll_mdc_exp, sizeof(KEY_MAX_EASIZE),
582 KEY_MAX_EASIZE, &size, lmmsize, NULL);
584 CERROR("Get max mdsize error rc %d \n", rc);
589 void ll_dump_inode(struct inode *inode)
591 struct list_head *tmp;
592 int dentry_count = 0;
594 LASSERT(inode != NULL);
596 list_for_each(tmp, &inode->i_dentry)
599 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
600 inode, ll_i2mdcexp(inode)->exp_obd->obd_name, inode->i_ino,
601 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
604 void lustre_dump_dentry(struct dentry *dentry, int recur)
606 struct list_head *tmp;
609 LASSERT(dentry != NULL);
611 list_for_each(tmp, &dentry->d_subdirs)
614 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
615 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
616 dentry->d_name.len, dentry->d_name.name,
617 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
618 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
619 dentry->d_flags, dentry->d_fsdata, subdirs);
620 if (dentry->d_inode != NULL)
621 ll_dump_inode(dentry->d_inode);
626 list_for_each(tmp, &dentry->d_subdirs) {
627 struct dentry *d = list_entry(tmp, struct dentry, d_child);
628 lustre_dump_dentry(d, recur - 1);
632 #ifdef HAVE_EXPORT___IGET
633 static void prune_dir_dentries(struct inode *inode)
635 struct dentry *dentry, *prev = NULL;
637 /* due to lustre specific logic, a directory
638 * can have few dentries - a bug from VFS POV */
640 spin_lock(&dcache_lock);
641 if (!list_empty(&inode->i_dentry)) {
642 dentry = list_entry(inode->i_dentry.prev,
643 struct dentry, d_alias);
644 /* in order to prevent infinite loops we
645 * break if previous dentry is busy */
646 if (dentry != prev) {
649 spin_unlock(&dcache_lock);
651 /* try to kill all child dentries */
652 shrink_dcache_parent(dentry);
655 /* now try to get rid of current dentry */
656 d_prune_aliases(inode);
660 spin_unlock(&dcache_lock);
663 static void prune_deathrow_one(struct ll_inode_info *lli)
665 struct inode *inode = ll_info2i(lli);
667 /* first, try to drop any dentries - they hold a ref on the inode */
668 if (S_ISDIR(inode->i_mode))
669 prune_dir_dentries(inode);
671 d_prune_aliases(inode);
674 /* if somebody still uses it, leave it */
675 LASSERT(atomic_read(&inode->i_count) > 0);
676 if (atomic_read(&inode->i_count) > 1)
679 CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
680 inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
682 /* seems nobody uses it anymore */
690 static void prune_deathrow(struct ll_sb_info *sbi, int try)
692 struct ll_inode_info *lli;
696 if (need_resched() && try)
700 if (!spin_trylock(&sbi->ll_deathrow_lock))
703 spin_lock(&sbi->ll_deathrow_lock);
708 if (!list_empty(&sbi->ll_deathrow)) {
709 lli = list_entry(sbi->ll_deathrow.next,
710 struct ll_inode_info,
712 list_del_init(&lli->lli_dead_list);
713 if (!list_empty(&sbi->ll_deathrow))
716 spin_unlock(&sbi->ll_deathrow_lock);
719 prune_deathrow_one(lli);
721 } while (empty == 0);
723 #else /* !HAVE_EXPORT___IGET */
724 #define prune_deathrow(sbi, try) do {} while (0)
725 #endif /* HAVE_EXPORT___IGET */
727 void client_common_put_super(struct super_block *sb)
729 struct ll_sb_info *sbi = ll_s2sbi(sb);
732 ll_close_thread_shutdown(sbi->ll_lcq);
734 lprocfs_unregister_mountpoint(sbi);
736 /* destroy inodes in deathrow */
737 prune_deathrow(sbi, 0);
739 list_del(&sbi->ll_conn_chain);
741 /* callbacks is cleared after disconnect each target */
742 obd_disconnect(sbi->ll_osc_exp);
743 sbi->ll_osc_exp = NULL;
745 obd_fid_fini(sbi->ll_mdc_exp);
746 obd_disconnect(sbi->ll_mdc_exp);
747 sbi->ll_mdc_exp = NULL;
752 void ll_kill_super(struct super_block *sb)
754 struct ll_sb_info *sbi;
759 if (!(sb->s_flags & MS_ACTIVE))
763 /* we need restore s_dev from changed for clustred NFS before put_super
764 * because new kernels have cached s_dev and change sb->s_dev in
765 * put_super not affected real removing devices */
767 sb->s_dev = sbi->ll_sdev_orig;
771 char *ll_read_opt(const char *opt, char *data)
777 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
778 if (strncmp(opt, data, strlen(opt)))
780 if ((value = strchr(data, '=')) == NULL)
784 OBD_ALLOC(retval, strlen(value) + 1);
786 CERROR("out of memory!\n");
790 memcpy(retval, value, strlen(value)+1);
791 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
795 static inline int ll_set_opt(const char *opt, char *data, int fl)
797 if (strncmp(opt, data, strlen(opt)) != 0)
803 /* non-client-specific mount options are parsed in lmd_parse */
804 static int ll_options(char *options, int *flags)
807 char *s1 = options, *s2;
813 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
816 CDEBUG(D_SUPER, "next opt=%s\n", s1);
817 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
822 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
827 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
832 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
837 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
842 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
847 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
849 /* Ignore deprecated mount option. The client will
850 * always try to mount with ACL support, whether this
851 * is used depends on whether server supports it. */
852 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
853 "mount option 'acl'.\n");
856 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
858 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
859 "mount option 'noacl'.\n");
863 tmp = ll_set_opt("checksum", s1, LL_SBI_DATA_CHECKSUM);
868 tmp = ll_set_opt("nochecksum", s1, LL_SBI_DATA_CHECKSUM);
874 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
879 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
884 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
890 s2 = strchr(s1, ',');
898 void ll_lli_init(struct ll_inode_info *lli)
900 lli->lli_inode_magic = LLI_INODE_MAGIC;
901 sema_init(&lli->lli_size_sem, 1);
902 sema_init(&lli->lli_write_sem, 1);
904 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
905 spin_lock_init(&lli->lli_lock);
906 sema_init(&lli->lli_och_sem, 1);
907 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
908 lli->lli_mds_exec_och = NULL;
909 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
910 lli->lli_open_fd_exec_count = 0;
911 INIT_LIST_HEAD(&lli->lli_dead_list);
912 #ifdef HAVE_CLOSE_THREAD
913 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
918 #define MDCDEV "mdc_dev"
919 static int old_lustre_process_log(struct super_block *sb, char *newprofile,
920 struct config_llog_instance *cfg)
922 struct lustre_sb_info *lsi = s2lsi(sb);
923 struct obd_device *obd;
924 struct lustre_handle mdc_conn = {0, };
925 struct obd_export *exp;
926 char *ptr, *mdt, *profile;
927 char niduuid[10] = "mdtnid0";
929 struct obd_uuid mdc_uuid;
930 struct llog_ctxt *ctxt;
931 struct obd_connect_data ocd = { 0 };
933 int i, rc = 0, recov_bk = 1, failnodes = 0;
936 ll_generate_random_uuid(uuid);
937 class_uuid_unparse(uuid, &mdc_uuid);
938 CDEBUG(D_HA, "generated uuid: %s\n", mdc_uuid.uuid);
940 /* Figure out the old mdt and profile name from new-style profile
941 ("lustre" from "mds/lustre-client") */
943 profile = strchr(mdt, '/');
944 if (profile == NULL) {
945 CDEBUG(D_CONFIG, "Can't find MDT name in %s\n", newprofile);
946 GOTO(out, rc = -EINVAL);
950 ptr = strrchr(profile, '-');
952 CDEBUG(D_CONFIG, "Can't find client name in %s\n", newprofile);
953 GOTO(out, rc = -EINVAL);
957 LCONSOLE_WARN("This looks like an old mount command; I will try to "
958 "contact MDT '%s' for profile '%s'\n", mdt, profile);
960 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
962 ptr = lsi->lsi_lmd->lmd_dev;
963 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
964 rc = do_lcfg(MDCDEV, nid, LCFG_ADD_UUID, niduuid, 0,0,0);
966 /* Stop at the first failover nid */
971 CERROR("No valid MDT nids found.\n");
972 GOTO(out, rc = -EINVAL);
976 rc = do_lcfg(MDCDEV, 0, LCFG_ATTACH, LUSTRE_MDC_NAME,mdc_uuid.uuid,0,0);
978 GOTO(out_del_uuid, rc);
980 rc = do_lcfg(MDCDEV, 0, LCFG_SETUP, mdt, niduuid, 0, 0);
982 LCONSOLE_ERROR_MSG(0x153, "I couldn't establish a connection "
983 "with the MDT. Check that the MDT host NID "
984 "is correct and the networks are up.\n");
985 GOTO(out_detach, rc);
988 obd = class_name2obd(MDCDEV);
990 GOTO(out_cleanup, rc = -EINVAL);
992 /* Add any failover nids */
993 while (*ptr == ':') {
994 /* New failover node */
995 sprintf(niduuid, "mdtnid%d", failnodes);
997 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
999 rc = do_lcfg(MDCDEV, nid, LCFG_ADD_UUID, niduuid,0,0,0);
1001 CERROR("Add uuid for %s failed %d\n",
1002 libcfs_nid2str(nid), rc);
1007 rc = do_lcfg(MDCDEV, 0, LCFG_ADD_CONN, niduuid, 0, 0,0);
1009 CERROR("Add conn for %s failed %d\n",
1010 libcfs_nid2str(nid), rc);
1018 /* Try all connections, but only once. */
1019 rc = obd_set_info_async(obd->obd_self_export,
1020 sizeof(KEY_INIT_RECOV_BACKUP), KEY_INIT_RECOV_BACKUP,
1021 sizeof(recov_bk), &recov_bk, NULL);
1023 GOTO(out_cleanup, rc);
1025 /* If we don't have this then an ACL MDS will refuse the connection */
1026 ocd.ocd_connect_flags = OBD_CONNECT_ACL;
1028 rc = obd_connect(&mdc_conn, obd, &mdc_uuid, &ocd, NULL);
1030 CERROR("cannot connect to %s: rc = %d\n", mdt, rc);
1031 GOTO(out_cleanup, rc);
1034 exp = class_conn2export(&mdc_conn);
1036 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1038 cfg->cfg_flags |= CFG_F_COMPAT146;
1041 rc = class_config_parse_llog(ctxt, profile, cfg);
1044 * For debugging, it's useful to just dump the log
1046 rc = class_config_dump_llog(ctxt, profile, cfg);
1048 llog_ctxt_put(ctxt);
1051 /* Set the caller's profile name to the old-style */
1052 memcpy(newprofile, profile, strlen(profile) + 1);
1056 LCONSOLE_ERROR_MSG(0x154, "%s: The configuration '%s' could not"
1057 " be read from the MDT '%s'. Make sure this"
1058 " client and the MDT are running compatible "
1059 "versions of Lustre.\n",
1060 obd->obd_name, profile, mdt);
1063 LCONSOLE_ERROR_MSG(0x155, "%s: The configuration '%s' could not"
1064 " be read from the MDT '%s'. This may be "
1065 "the result of communication errors between "
1066 "the client and the MDT, or if the MDT is "
1067 "not running.\n", obd->obd_name, profile,
1072 /* We don't so much care about errors in cleaning up the config llog
1073 * connection, as we have already read the config by this point. */
1074 obd_disconnect(exp);
1077 do_lcfg(MDCDEV, 0, LCFG_CLEANUP, 0, 0, 0, 0);
1080 do_lcfg(MDCDEV, 0, LCFG_DETACH, 0, 0, 0, 0);
1083 /* class_add_uuid adds a nid even if the same uuid exists; we might
1084 delete any copy here. So they all better match. */
1085 for (i = 0; i < failnodes; i++) {
1086 sprintf(niduuid, "mdtnid%d", i);
1087 do_lcfg(MDCDEV, 0, LCFG_DEL_UUID, niduuid, 0, 0, 0);
1089 /* class_import_put will get rid of the additional connections */
1093 /* end COMPAT_146 */
1095 int ll_fill_super(struct super_block *sb)
1097 struct lustre_profile *lprof = NULL;
1098 struct lustre_sb_info *lsi = s2lsi(sb);
1099 struct ll_sb_info *sbi;
1100 char *osc = NULL, *mdc = NULL;
1101 char *profilenm = get_profile_name(sb);
1102 struct config_llog_instance cfg = {0, };
1103 char ll_instance[sizeof(sb) * 2 + 1];
1106 char pseudo[32] = { 0 };
1109 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1113 /* client additional sb info */
1114 lsi->lsi_llsbi = sbi = ll_init_sbi();
1120 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1122 GOTO(out_free, err);
1124 /* Generate a string unique to this super, in case some joker tries
1125 to mount the same fs at two mount points.
1126 Use the address of the super itself.*/
1127 sprintf(ll_instance, "%p", sb);
1128 cfg.cfg_instance = ll_instance;
1129 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1132 /* set up client obds */
1133 if (strchr(profilenm, '/') != NULL) /* COMPAT_146 */
1134 err = -EINVAL; /* skip error messages, use old config code */
1136 err = lustre_process_log(sb, profilenm, &cfg);
1141 oldnamelen = strlen(profilenm) + 1;
1142 /* Temp storage for 1.4.6 profile name */
1143 OBD_ALLOC(oldname, oldnamelen);
1145 memcpy(oldname, profilenm, oldnamelen);
1146 rc = old_lustre_process_log(sb, oldname, &cfg);
1148 /* That worked - update the profile name
1151 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1152 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1153 OBD_ALLOC(lsi->lsi_lmd->lmd_profile,
1154 strlen(oldname) + 1);
1155 if (!lsi->lsi_lmd->lmd_profile) {
1156 OBD_FREE(oldname, oldnamelen);
1157 GOTO(out_free, err = -ENOMEM);
1159 memcpy(lsi->lsi_lmd->lmd_profile, oldname,
1160 strlen(oldname) + 1);
1161 profilenm = get_profile_name(sb);
1162 /* Don't ever try to recover the MGS */
1163 rc = ptlrpc_set_import_active(
1164 lsi->lsi_mgc->u.cli.cl_import, 0);
1166 OBD_FREE(oldname, oldnamelen);
1169 /* end COMPAT_146 */
1171 CERROR("Unable to process log: %d\n", err);
1172 GOTO(out_free, err);
1175 lprof = class_get_profile(profilenm);
1176 if (lprof == NULL) {
1177 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1178 " read from the MGS. Does that filesystem "
1179 "exist?\n", profilenm);
1180 GOTO(out_free, err = -EINVAL);
1184 * The configuration for 1.8 client and 2.0 client are different.
1185 * 2.0 introduces lmv, but 1.8 directly uses mdc.
1186 * Here, we will hack to use proper name for mdc if needed.
1192 save = lprof->lp_mdc;
1193 fsname_end = strrchr(save, '-');
1195 namelen = fsname_end - save;
1196 if (strcmp(fsname_end, "-clilmv") == 0) {
1197 strncpy(pseudo, save, namelen);
1198 strcat(pseudo, "-MDT0000-mdc");
1199 lprof->lp_mdc = pseudo;
1200 CDEBUG(D_INFO, "1.8.x connecting to 2.0: lmv=%s"
1201 " new mdc=%s\n", save, pseudo);
1206 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1207 lprof->lp_mdc, lprof->lp_osc);
1209 OBD_ALLOC(osc, strlen(lprof->lp_osc) +
1210 strlen(ll_instance) + 2);
1212 GOTO(out_free, err = -ENOMEM);
1213 sprintf(osc, "%s-%s", lprof->lp_osc, ll_instance);
1215 OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
1216 strlen(ll_instance) + 2);
1218 GOTO(out_free, err = -ENOMEM);
1219 sprintf(mdc, "%s-%s", lprof->lp_mdc, ll_instance);
1221 /* connections, registrations, sb setup */
1222 err = client_common_fill_super(sb, mdc, osc);
1226 lprof->lp_mdc = save;
1228 OBD_FREE(mdc, strlen(mdc) + 1);
1230 OBD_FREE(osc, strlen(osc) + 1);
1234 LCONSOLE_WARN("Client %s has started\n", profilenm);
1237 } /* ll_fill_super */
1240 void ll_put_super(struct super_block *sb)
1242 struct config_llog_instance cfg;
1243 char ll_instance[sizeof(sb) * 2 + 1];
1244 struct obd_device *obd;
1245 struct lustre_sb_info *lsi = s2lsi(sb);
1246 struct ll_sb_info *sbi = ll_s2sbi(sb);
1247 char *profilenm = get_profile_name(sb);
1248 int force = 1, next;
1251 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1253 sprintf(ll_instance, "%p", sb);
1254 cfg.cfg_instance = ll_instance;
1255 lustre_end_log(sb, NULL, &cfg);
1257 if (sbi->ll_mdc_exp) {
1258 obd = class_exp2obd(sbi->ll_mdc_exp);
1260 force = obd->obd_force;
1263 /* We need to set force before the lov_disconnect in
1264 lustre_common_put_super, since l_d cleans up osc's as well. */
1267 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1269 obd->obd_force = force;
1274 /* Only if client_common_fill_super succeeded */
1275 client_common_put_super(sb);
1279 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1280 class_manual_cleanup(obd);
1284 class_del_profile(profilenm);
1287 lsi->lsi_llsbi = NULL;
1289 lustre_common_put_super(sb);
1291 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1296 } /* client_put_super */
1298 #if defined(HAVE_REGISTER_CACHE) || defined(HAVE_SHRINKER_CACHE)
1300 #if defined(HAVE_CACHE_RETURN_INT)
1305 ll_shrink_cache(int priority, unsigned int gfp_mask)
1307 struct ll_sb_info *sbi;
1310 list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1311 count += llap_shrink_cache(sbi, priority);
1313 #if defined(HAVE_CACHE_RETURN_INT)
1318 struct cache_definition ll_cache_definition = {
1319 .name = "llap_cache",
1320 .shrink = ll_shrink_cache
1322 #endif /* HAVE_REGISTER_CACHE || HAVE_SHRINKER_CACHE */
1324 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1326 struct inode *inode = NULL;
1327 /* NOTE: we depend on atomic igrab() -bzzz */
1328 lock_res_and_lock(lock);
1329 if (lock->l_ast_data) {
1330 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1331 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1332 inode = igrab(lock->l_ast_data);
1334 inode = lock->l_ast_data;
1335 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1337 lock, __FILE__, __func__, __LINE__,
1338 "l_ast_data %p is bogus: magic %08x",
1339 lock->l_ast_data, lli->lli_inode_magic);
1343 unlock_res_and_lock(lock);
1347 static int null_if_equal(struct ldlm_lock *lock, void *data)
1349 if (data == lock->l_ast_data) {
1350 lock->l_ast_data = NULL;
1352 if (lock->l_req_mode != lock->l_granted_mode)
1353 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1356 return LDLM_ITER_CONTINUE;
1359 void ll_clear_inode(struct inode *inode)
1362 struct ll_inode_info *lli = ll_i2info(inode);
1363 struct ll_sb_info *sbi = ll_i2sbi(inode);
1366 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1367 inode->i_generation, inode);
1369 if (S_ISDIR(inode->i_mode)) {
1370 /* these should have been cleared in ll_file_release */
1371 LASSERT(lli->lli_sai == NULL);
1372 LASSERT(lli->lli_opendir_key == NULL);
1373 LASSERT(lli->lli_opendir_pid == 0);
1376 ll_inode2fid(&fid, inode);
1377 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1378 mdc_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
1380 LASSERT(!lli->lli_open_fd_write_count);
1381 LASSERT(!lli->lli_open_fd_read_count);
1382 LASSERT(!lli->lli_open_fd_exec_count);
1384 if (lli->lli_mds_write_och)
1385 ll_mdc_real_close(inode, FMODE_WRITE);
1386 if (lli->lli_mds_exec_och) {
1388 CERROR("No FMODE exec, bug exec och is present for "
1389 "inode %ld\n", inode->i_ino);
1390 ll_mdc_real_close(inode, FMODE_EXEC);
1392 if (lli->lli_mds_read_och)
1393 ll_mdc_real_close(inode, FMODE_READ);
1397 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
1398 null_if_equal, inode);
1400 obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
1401 lli->lli_smd = NULL;
1404 if (lli->lli_symlink_name) {
1405 OBD_FREE(lli->lli_symlink_name,
1406 strlen(lli->lli_symlink_name) + 1);
1407 lli->lli_symlink_name = NULL;
1410 #ifdef CONFIG_FS_POSIX_ACL
1411 if (lli->lli_posix_acl) {
1412 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1413 posix_acl_release(lli->lli_posix_acl);
1414 lli->lli_posix_acl = NULL;
1418 lli->lli_inode_magic = LLI_INODE_DEAD;
1420 #ifdef HAVE_EXPORT___IGET
1421 spin_lock(&sbi->ll_deathrow_lock);
1422 list_del_init(&lli->lli_dead_list);
1423 spin_unlock(&sbi->ll_deathrow_lock);
1428 static int ll_setattr_do_truncate(struct inode *inode, loff_t new_size)
1430 struct ll_sb_info *sbi = ll_i2sbi(inode);
1431 struct ll_inode_info *lli = ll_i2info(inode);
1432 struct lov_stripe_md *lsm = lli->lli_smd;
1434 ldlm_policy_data_t policy = { .l_extent = {new_size,
1436 struct lustre_handle lockh = { 0 };
1437 int local_lock = 0; /* 0 - no local lock;
1438 * 1 - lock taken by lock_extent;
1439 * 2 - by obd_match*/
1444 UNLOCK_INODE_MUTEX(inode);
1445 UP_WRITE_I_ALLOC_SEM(inode);
1447 if (sbi->ll_lockless_truncate_enable &&
1448 (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK)) {
1449 ast_flags = LDLM_FL_BLOCK_GRANTED;
1450 rc = obd_match(sbi->ll_osc_exp, lsm, LDLM_EXTENT,
1451 &policy, LCK_PW, &ast_flags, inode, &lockh);
1455 } else if (rc == 0) {
1456 rc = ll_file_punch(inode, new_size, 1);
1459 /* XXX when we fix the AST intents to pass the discard-range
1460 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1462 ast_flags = (new_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
1463 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
1465 if (likely(rc == 0))
1469 LOCK_INODE_MUTEX(inode);
1470 DOWN_WRITE_I_ALLOC_SEM(inode);
1471 if (likely(rc == 0)) {
1472 /* Only ll_inode_size_lock is taken at this level.
1473 * lov_stripe_lock() is grabbed by ll_truncate() only over
1474 * call to obd_adjust_kms(). If vmtruncate returns 0, then
1475 * ll_truncate dropped ll_inode_size_lock() */
1476 ll_inode_size_lock(inode, 0);
1478 set_bit(LLI_F_SRVLOCK, &lli->lli_flags);
1479 rc = vmtruncate(inode, new_size);
1480 clear_bit(LLI_F_SRVLOCK, &lli->lli_flags);
1482 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1483 ll_inode_size_unlock(inode, 0);
1487 if (local_lock == 2)
1488 err = obd_cancel(sbi->ll_osc_exp, lsm, LCK_PW, &lockh);
1490 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1491 if (unlikely(err != 0)){
1492 CERROR("extent unlock failed: err=%d,"
1493 " unlock method =%d\n", err, local_lock);
1501 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1502 * object(s) determine the file size and mtime. Otherwise, the MDS will
1503 * keep these values until such a time that objects are allocated for it.
1504 * We do the MDS operations first, as it is checking permissions for us.
1505 * We don't to the MDS RPC if there is nothing that we want to store there,
1506 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1507 * going to do an RPC anyways.
1509 * If we are doing a truncate, we will send the mtime and ctime updates
1510 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1511 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1514 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1516 struct ll_inode_info *lli = ll_i2info(inode);
1517 struct lov_stripe_md *lsm = lli->lli_smd;
1518 struct ll_sb_info *sbi = ll_i2sbi(inode);
1519 struct ptlrpc_request *request = NULL;
1520 struct mdc_op_data op_data = { { 0 } };
1521 struct lustre_md md;
1522 int ia_valid = attr->ia_valid;
1526 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1528 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1530 if (ia_valid & ATTR_SIZE) {
1531 if (attr->ia_size > ll_file_maxbytes(inode)) {
1532 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1533 attr->ia_size, ll_file_maxbytes(inode));
1537 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1540 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1541 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1542 if (current->fsuid != inode->i_uid &&
1543 !cfs_capable(CFS_CAP_FOWNER))
1547 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1548 if (attr->ia_valid & ATTR_CTIME) {
1549 attr->ia_ctime = CURRENT_TIME;
1550 attr->ia_valid |= ATTR_CTIME_SET;
1552 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1553 attr->ia_atime = CURRENT_TIME;
1554 attr->ia_valid |= ATTR_ATIME_SET;
1556 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1557 attr->ia_mtime = CURRENT_TIME;
1558 attr->ia_valid |= ATTR_MTIME_SET;
1561 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1562 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1563 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1566 /* NB: ATTR_SIZE will only be set after this point if the size
1567 * resides on the MDS, ie, this file has no objects. */
1569 attr->ia_valid &= ~ATTR_SIZE;
1571 /* We always do an MDS RPC, even if we're only changing the size;
1572 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1573 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0, NULL);
1575 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
1576 attr, NULL, 0, NULL, 0, &request);
1579 ptlrpc_req_finished(request);
1580 if (rc == -ENOENT) {
1582 /* Unlinked special device node? Or just a race?
1583 * Pretend we done everything. */
1584 if (!S_ISREG(inode->i_mode) &&
1585 !S_ISDIR(inode->i_mode))
1586 rc = inode_setattr(inode, attr);
1587 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY)
1588 CERROR("mdc_setattr fails: rc = %d\n", rc);
1592 rc = mdc_req2lustre_md(request, REPLY_REC_OFF, sbi->ll_osc_exp, &md);
1594 ptlrpc_req_finished(request);
1598 /* We call inode_setattr to adjust timestamps.
1599 * If there is at least some data in file, we cleared ATTR_SIZE above to
1600 * avoid invoking vmtruncate, otherwise it is important to call
1601 * vmtruncate in inode_setattr to update inode->i_size (bug 6196) */
1602 rc = inode_setattr(inode, attr);
1604 ll_update_inode(inode, &md);
1605 ptlrpc_req_finished(request);
1607 if (!lsm || !S_ISREG(inode->i_mode)) {
1608 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1612 /* We really need to get our PW lock before we change inode->i_size.
1613 * If we don't we can race with other i_size updaters on our node, like
1614 * ll_file_read. We can also race with i_size propogation to other
1615 * nodes through dirtying and writeback of final cached pages. This
1616 * last one is especially bad for racing o_append users on other
1618 if (ia_valid & ATTR_SIZE) {
1619 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1620 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1621 struct obd_info oinfo = { { { 0 } } };
1623 struct lustre_handle lockh = { 0 };
1626 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1627 inode->i_ino, LTIME_S(attr->ia_mtime));
1631 oa->o_id = lsm->lsm_object_id;
1632 oa->o_gr = lsm->lsm_object_gr;
1633 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1635 valid = OBD_MD_FLTYPE | OBD_MD_FLFID | OBD_MD_FLGENER;
1637 if (LTIME_S(attr->ia_mtime) < LTIME_S(attr->ia_ctime)){
1638 struct ost_lvb xtimes;
1640 /* setting mtime to past is performed under PW
1641 * EOF extent lock */
1642 oinfo.oi_policy.l_extent.start = 0;
1643 oinfo.oi_policy.l_extent.end = OBD_OBJECT_EOF;
1644 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW,
1650 /* setattr under locks
1652 * 1. restore inode's timestamps which
1653 * are about to be set as long as
1654 * concurrent stat (via
1655 * ll_glimpse_size) might bring
1658 * 2. update lsm so that next stat
1659 * (via ll_glimpse_size) could get
1660 * correct values in lsm */
1661 lov_stripe_lock(lli->lli_smd);
1662 if (ia_valid & ATTR_ATIME) {
1663 LTIME_S(inode->i_atime) =
1665 LTIME_S(attr->ia_atime);
1666 valid |= OBD_MD_FLATIME;
1668 if (ia_valid & ATTR_MTIME) {
1669 LTIME_S(inode->i_mtime) =
1671 LTIME_S(attr->ia_mtime);
1672 valid |= OBD_MD_FLMTIME;
1674 if (ia_valid & ATTR_CTIME) {
1675 LTIME_S(inode->i_ctime) =
1677 LTIME_S(attr->ia_ctime);
1678 valid |= OBD_MD_FLCTIME;
1681 obd_update_lvb(ll_i2obdexp(inode), lli->lli_smd,
1683 lov_stripe_unlock(lli->lli_smd);
1687 * 1. do not use inode's timestamps
1688 * because concurrent stat might fill
1689 * the inode with out-of-date times,
1690 * send values from attr instead
1692 * 2.do no update lsm, as long as stat
1693 * (via ll_glimpse_size) will bring
1694 * attributes from osts anyway */
1695 if (ia_valid & ATTR_ATIME) {
1696 oa->o_atime = LTIME_S(attr->ia_atime);
1697 oa->o_valid |= OBD_MD_FLATIME;
1699 if (ia_valid & ATTR_MTIME) {
1700 oa->o_mtime = LTIME_S(attr->ia_mtime);
1701 oa->o_valid |= OBD_MD_FLMTIME;
1703 if (ia_valid & ATTR_CTIME) {
1704 oa->o_ctime = LTIME_S(attr->ia_ctime);
1705 oa->o_valid |= OBD_MD_FLCTIME;
1709 obdo_from_inode(oa, inode, valid);
1714 rc = obd_setattr_rqset(sbi->ll_osc_exp, &oinfo, NULL);
1716 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1718 if (LTIME_S(attr->ia_mtime) < LTIME_S(attr->ia_ctime)){
1721 err = ll_extent_unlock(NULL, inode, lsm,
1723 if (unlikely(err != 0)) {
1724 CERROR("extent unlock failed: "
1738 int ll_setattr(struct dentry *de, struct iattr *attr)
1742 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1743 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1744 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1745 if ((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1746 (ATTR_SIZE|ATTR_MODE)) {
1747 mode = de->d_inode->i_mode;
1748 if (((mode & S_ISUID) && (!(attr->ia_mode & S_ISUID))) ||
1749 ((mode & S_ISGID) && (mode & S_IXGRP) &&
1750 (!(attr->ia_mode & S_ISGID))))
1751 attr->ia_valid |= ATTR_FORCE;
1754 return ll_setattr_raw(de->d_inode, attr);
1757 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1758 __u64 max_age, __u32 flags)
1760 struct ll_sb_info *sbi = ll_s2sbi(sb);
1761 struct obd_statfs obd_osfs;
1765 rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age, flags);
1767 CERROR("mdc_statfs fails: rc = %d\n", rc);
1771 osfs->os_type = sb->s_magic;
1773 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1774 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1776 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_osc_exp),
1777 &obd_osfs, max_age, flags);
1779 CERROR("obd_statfs fails: rc = %d\n", rc);
1783 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1784 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1787 osfs->os_bsize = obd_osfs.os_bsize;
1788 osfs->os_blocks = obd_osfs.os_blocks;
1789 osfs->os_bfree = obd_osfs.os_bfree;
1790 osfs->os_bavail = obd_osfs.os_bavail;
1792 /* If we don't have as many objects free on the OST as inodes
1793 * on the MDS, we reduce the total number of inodes to
1794 * compensate, so that the "inodes in use" number is correct.
1796 if (obd_osfs.os_ffree < osfs->os_ffree) {
1797 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1799 osfs->os_ffree = obd_osfs.os_ffree;
1804 #ifndef HAVE_STATFS_DENTRY_PARAM
1805 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1808 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1810 struct super_block *sb = de->d_sb;
1812 struct obd_statfs osfs;
1815 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1816 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1818 /* For now we will always get up-to-date statfs values, but in the
1819 * future we may allow some amount of caching on the client (e.g.
1820 * from QOS or lprocfs updates). */
1821 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1825 statfs_unpack(sfs, &osfs);
1827 /* We need to downshift for all 32-bit kernels, because we can't
1828 * tell if the kernel is being called via sys_statfs64() or not.
1829 * Stop before overflowing f_bsize - in which case it is better
1830 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1831 if (sizeof(long) < 8) {
1832 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1835 osfs.os_blocks >>= 1;
1836 osfs.os_bfree >>= 1;
1837 osfs.os_bavail >>= 1;
1841 sfs->f_blocks = osfs.os_blocks;
1842 sfs->f_bfree = osfs.os_bfree;
1843 sfs->f_bavail = osfs.os_bavail;
1848 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1850 struct ll_inode_info *lli;
1851 struct lov_stripe_md *lsm;
1853 lli = ll_i2info(inode);
1854 LASSERT(lli->lli_size_sem_owner != current);
1855 down(&lli->lli_size_sem);
1856 LASSERT(lli->lli_size_sem_owner == NULL);
1857 lli->lli_size_sem_owner = current;
1859 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1862 lov_stripe_lock(lsm);
1865 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1867 struct ll_inode_info *lli;
1868 struct lov_stripe_md *lsm;
1870 lli = ll_i2info(inode);
1872 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1875 lov_stripe_unlock(lsm);
1876 LASSERT(lli->lli_size_sem_owner == current);
1877 lli->lli_size_sem_owner = NULL;
1878 up(&lli->lli_size_sem);
1881 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1883 struct ll_inode_info *lli = ll_i2info(inode);
1885 dump_lsm(D_INODE, lsm);
1886 dump_lsm(D_INODE, lli->lli_smd);
1887 LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1888 "lsm must be joined lsm %p\n", lsm);
1889 obd_free_memmd(ll_i2obdexp(inode), &lli->lli_smd);
1890 CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1891 lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1893 lli->lli_maxbytes = lsm->lsm_maxbytes;
1894 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1895 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1898 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1900 struct ll_inode_info *lli = ll_i2info(inode);
1901 struct mds_body *body = md->body;
1902 struct lov_stripe_md *lsm = md->lsm;
1903 struct ll_sb_info *sbi = ll_i2sbi(inode);
1906 CDEBUG(D_INODE, "body->valid = "LPX64"\n", body->valid);
1908 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1910 if (lli->lli_smd == NULL) {
1911 if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1912 lsm->lsm_magic != LOV_MAGIC_V3 &&
1913 lsm->lsm_magic != LOV_MAGIC_JOIN) {
1914 dump_lsm(D_ERROR, lsm);
1917 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1918 lsm, inode->i_ino, inode->i_generation, inode);
1919 /* ll_inode_size_lock() requires it is only called
1920 * with lli_smd != NULL or lock_lsm == 0 or we can
1921 * race between lock/unlock. bug 9547 */
1923 lli->lli_maxbytes = lsm->lsm_maxbytes;
1924 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1925 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1927 if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1928 lli->lli_smd->lsm_stripe_count ==
1929 lsm->lsm_stripe_count) {
1930 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1931 CERROR("lsm mismatch for inode %ld\n",
1933 CERROR("lli_smd:\n");
1934 dump_lsm(D_ERROR, lli->lli_smd);
1936 dump_lsm(D_ERROR, lsm);
1940 ll_replace_lsm(inode, lsm);
1942 if (lli->lli_smd != lsm)
1943 obd_free_memmd(ll_i2obdexp(inode), &lsm);
1946 #ifdef CONFIG_FS_POSIX_ACL
1947 LASSERT(!md->posix_acl || (body->valid & OBD_MD_FLACL));
1948 if (body->valid & OBD_MD_FLACL) {
1949 spin_lock(&lli->lli_lock);
1950 if (lli->lli_posix_acl)
1951 posix_acl_release(lli->lli_posix_acl);
1952 lli->lli_posix_acl = md->posix_acl;
1953 spin_unlock(&lli->lli_lock);
1957 inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1958 if (body->valid & OBD_MD_FLGENER)
1959 inode->i_generation = body->generation;
1961 if (body->valid & OBD_MD_FLATIME) {
1962 if (body->atime > LTIME_S(inode->i_atime))
1963 LTIME_S(inode->i_atime) = body->atime;
1964 lli->lli_lvb.lvb_atime = body->atime;
1966 if (body->valid & OBD_MD_FLMTIME) {
1967 if (body->mtime > LTIME_S(inode->i_mtime)) {
1968 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1969 "to "LPU64"\n", inode->i_ino,
1970 LTIME_S(inode->i_mtime), body->mtime);
1971 LTIME_S(inode->i_mtime) = body->mtime;
1973 lli->lli_lvb.lvb_mtime = body->mtime;
1975 if (body->valid & OBD_MD_FLCTIME) {
1976 if (body->ctime > LTIME_S(inode->i_ctime))
1977 LTIME_S(inode->i_ctime) = body->ctime;
1978 lli->lli_lvb.lvb_ctime = body->ctime;
1980 if (body->valid & OBD_MD_FLMODE)
1981 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1982 if (body->valid & OBD_MD_FLTYPE)
1983 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1984 if (S_ISREG(inode->i_mode)) {
1985 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS+1, LL_MAX_BLKSIZE_BITS);
1987 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1989 #ifdef HAVE_INODE_BLKSIZE
1990 inode->i_blksize = 1<<inode->i_blkbits;
1992 if (body->valid & OBD_MD_FLUID)
1993 inode->i_uid = body->uid;
1994 if (body->valid & OBD_MD_FLGID)
1995 inode->i_gid = body->gid;
1996 if (body->valid & OBD_MD_FLFLAGS)
1997 inode->i_flags = ll_ext_to_inode_flags(body->flags |
1998 MDS_BFLAG_EXT_FLAGS);
1999 if (body->valid & OBD_MD_FLNLINK)
2000 inode->i_nlink = body->nlink;
2002 if (body->valid & OBD_MD_FLRDEV)
2003 inode->i_rdev = old_decode_dev(body->rdev);
2004 if (body->valid & OBD_MD_FLSIZE) {
2005 #if 0 /* Can't block ll_test_inode->ll_update_inode, b=14326*/
2006 ll_inode_size_lock(inode, 0);
2007 i_size_write(inode, body->size);
2008 ll_inode_size_unlock(inode, 0);
2010 inode->i_size = body->size;
2013 if (body->valid & OBD_MD_FLBLOCKS)
2014 inode->i_blocks = body->blocks;
2016 if (body->valid & OBD_MD_FLSIZE)
2017 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
2021 static struct backing_dev_info ll_backing_dev_info = {
2022 .ra_pages = 0, /* No readahead */
2023 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
2024 .capabilities = 0, /* Does contribute to dirty memory */
2026 .memory_backed = 0, /* Does contribute to dirty memory */
2030 void ll_read_inode2(struct inode *inode, void *opaque)
2032 struct lustre_md *md = opaque;
2033 struct ll_inode_info *lli = ll_i2info(inode);
2036 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
2037 inode->i_generation, inode);
2041 LASSERT(!lli->lli_smd);
2043 /* Core attributes from the MDS first. This is a new inode, and
2044 * the VFS doesn't zero times in the core inode so we have to do
2045 * it ourselves. They will be overwritten by either MDS or OST
2046 * attributes - we just need to make sure they aren't newer. */
2047 LTIME_S(inode->i_mtime) = 0;
2048 LTIME_S(inode->i_atime) = 0;
2049 LTIME_S(inode->i_ctime) = 0;
2051 ll_update_inode(inode, md);
2053 /* OIDEBUG(inode); */
2055 if (S_ISREG(inode->i_mode)) {
2056 struct ll_sb_info *sbi = ll_i2sbi(inode);
2057 inode->i_op = &ll_file_inode_operations;
2058 inode->i_fop = sbi->ll_fop;
2059 inode->i_mapping->a_ops = &ll_aops;
2061 } else if (S_ISDIR(inode->i_mode)) {
2062 inode->i_op = &ll_dir_inode_operations;
2063 inode->i_fop = &ll_dir_operations;
2064 inode->i_mapping->a_ops = &ll_dir_aops;
2066 } else if (S_ISLNK(inode->i_mode)) {
2067 inode->i_op = &ll_fast_symlink_inode_operations;
2070 inode->i_op = &ll_special_inode_operations;
2071 init_special_inode(inode, inode->i_mode,
2072 kdev_t_to_nr(inode->i_rdev));
2073 /* initializing backing dev info. */
2074 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2079 int ll_iocontrol(struct inode *inode, struct file *file,
2080 unsigned int cmd, unsigned long arg)
2082 struct ll_sb_info *sbi = ll_i2sbi(inode);
2083 struct ptlrpc_request *req = NULL;
2088 case EXT3_IOC_GETFLAGS: {
2090 struct mds_body *body;
2092 ll_inode2fid(&fid, inode);
2093 rc = mdc_getattr(sbi->ll_mdc_exp, &fid, OBD_MD_FLFLAGS,0,&req);
2095 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2099 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
2102 /* We want to return EXT3_*_FL flags to the caller via this
2103 * ioctl. An older MDS may be sending S_* flags, fix it up. */
2104 flags = ll_inode_to_ext_flags(body->flags,
2105 MDS_BFLAG_EXT_FLAGS);
2106 ptlrpc_req_finished (req);
2108 RETURN(put_user(flags, (int *)arg));
2110 case EXT3_IOC_SETFLAGS: {
2111 struct mdc_op_data op_data = { { 0 } };
2112 struct ll_iattr_struct attr;
2113 struct obd_info oinfo = { { { 0 } } };
2114 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2116 if (get_user(flags, (int *)arg))
2120 OBDO_ALLOC(oinfo.oi_oa);
2124 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0, NULL);
2126 memset(&attr, 0, sizeof(attr));
2127 attr.ia_attr_flags = flags;
2128 ((struct iattr *)&attr)->ia_valid |= ATTR_ATTR_FLAG;
2130 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
2131 (struct iattr *)&attr, NULL, 0, NULL, 0, &req);
2132 ptlrpc_req_finished(req);
2133 if (rc || lsm == NULL) {
2134 OBDO_FREE(oinfo.oi_oa);
2138 oinfo.oi_oa->o_id = lsm->lsm_object_id;
2139 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
2140 oinfo.oi_oa->o_flags = flags;
2141 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP |
2144 obdo_from_inode(oinfo.oi_oa, inode,
2145 OBD_MD_FLFID | OBD_MD_FLGENER);
2146 rc = obd_setattr_rqset(sbi->ll_osc_exp, &oinfo, NULL);
2147 OBDO_FREE(oinfo.oi_oa);
2149 if (rc != -EPERM && rc != -EACCES)
2150 CERROR("mdc_setattr_async fails: rc = %d\n", rc);
2154 inode->i_flags = ll_ext_to_inode_flags(flags |
2155 MDS_BFLAG_EXT_FLAGS);
2165 /* umount -f client means force down, don't save state */
2166 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2167 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2169 struct super_block *sb = vfsmnt->mnt_sb;
2171 void ll_umount_begin(struct super_block *sb)
2174 struct lustre_sb_info *lsi = s2lsi(sb);
2175 struct ll_sb_info *sbi = ll_s2sbi(sb);
2176 struct obd_device *obd;
2177 struct obd_ioctl_data ioc_data = { 0 };
2180 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2181 if (!(flags & MNT_FORCE)) {
2187 /* Tell the MGC we got umount -f */
2188 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2190 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2191 sb->s_count, atomic_read(&sb->s_active));
2193 obd = class_exp2obd(sbi->ll_mdc_exp);
2195 CERROR("Invalid MDC connection handle "LPX64"\n",
2196 sbi->ll_mdc_exp->exp_handle.h_cookie);
2201 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
2204 obd = class_exp2obd(sbi->ll_osc_exp);
2206 CERROR("Invalid LOV connection handle "LPX64"\n",
2207 sbi->ll_osc_exp->exp_handle.h_cookie);
2213 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
2216 /* Really, we'd like to wait until there are no requests outstanding,
2217 * and then continue. For now, we just invalidate the requests,
2218 * schedule() and sleep one second if needed, and hope.
2221 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2222 if (atomic_read(&vfsmnt->mnt_count) > 2) {
2223 cfs_schedule_timeout(CFS_TASK_INTERRUPTIBLE,
2224 cfs_time_seconds(1));
2225 if (atomic_read(&vfsmnt->mnt_count) > 2)
2226 LCONSOLE_WARN("Mount still busy with %d refs! You "
2227 "may try to umount it a bit later\n",
2228 atomic_read(&vfsmnt->mnt_count));
2235 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2237 struct ll_sb_info *sbi = ll_s2sbi(sb);
2241 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2242 read_only = *flags & MS_RDONLY;
2243 err = obd_set_info_async(sbi->ll_mdc_exp, sizeof(KEY_READONLY),
2244 KEY_READONLY, sizeof(read_only),
2247 /* MDS might have expected a different ro key value, b=17493 */
2248 if (err == -EINVAL) {
2249 CDEBUG(D_CONFIG, "Retrying remount with 1.6.6 ro key\n");
2250 err = obd_set_info_async(sbi->ll_mdc_exp,
2251 sizeof(KEY_READONLY_166COMPAT),
2252 KEY_READONLY_166COMPAT,
2258 CERROR("Failed to change the read-only flag during "
2259 "remount: %d\n", err);
2264 sb->s_flags |= MS_RDONLY;
2266 sb->s_flags &= ~MS_RDONLY;
2271 int ll_prep_inode(struct obd_export *exp, struct inode **inode,
2272 struct ptlrpc_request *req, int offset,struct super_block *sb)
2274 struct lustre_md md;
2275 struct ll_sb_info *sbi = NULL;
2279 LASSERT(*inode || sb);
2280 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2281 prune_deathrow(sbi, 1);
2283 rc = mdc_req2lustre_md(req, offset, exp, &md);
2288 ll_update_inode(*inode, &md);
2291 /** hashing VFS inode by FIDs.
2292 * IGIF will be used for for compatibility if needed.
2294 *inode =ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2295 if (*inode == NULL || is_bad_inode(*inode)) {
2296 mdc_free_lustre_md(exp, &md);
2298 CERROR("new_inode -fatal: rc %d\n", rc);
2303 rc = obd_checkmd(exp, ll_i2mdcexp(*inode),
2304 ll_i2info(*inode)->lli_smd);
2309 char *llap_origins[] = {
2310 [LLAP_ORIGIN_UNKNOWN] = "--",
2311 [LLAP_ORIGIN_READPAGE] = "rp",
2312 [LLAP_ORIGIN_READAHEAD] = "ra",
2313 [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2314 [LLAP_ORIGIN_WRITEPAGE] = "wp",
2315 [LLAP_ORIGIN_LOCKLESS_IO] = "ls"
2318 struct ll_async_page *llite_pglist_next_llap(struct list_head *head,
2319 struct list_head *list)
2321 struct ll_async_page *llap;
2322 struct list_head *pos;
2324 list_for_each(pos, list) {
2327 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2328 if (llap->llap_page == NULL)
2336 int ll_obd_statfs(struct inode *inode, void *arg)
2338 struct ll_sb_info *sbi = NULL;
2339 struct obd_device *client_obd = NULL, *lov_obd = NULL;
2340 struct lov_obd *lov = NULL;
2341 struct obd_statfs stat_buf = {0};
2343 struct obd_ioctl_data *data = NULL;
2347 if (!inode || !(sbi = ll_i2sbi(inode)))
2348 GOTO(out_statfs, rc = -EINVAL);
2350 rc = obd_ioctl_getdata(&buf, &len, arg);
2352 GOTO(out_statfs, rc);
2355 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2356 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2357 GOTO(out_statfs, rc = -EINVAL);
2359 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2360 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2362 if (type == LL_STATFS_MDC) {
2364 GOTO(out_statfs, rc = -ENODEV);
2365 client_obd = class_exp2obd(sbi->ll_mdc_exp);
2366 } else if (type == LL_STATFS_LOV) {
2367 lov_obd = class_exp2obd(sbi->ll_osc_exp);
2368 lov = &lov_obd->u.lov;
2370 if (index >= lov->desc.ld_tgt_count)
2371 GOTO(out_statfs, rc = -ENODEV);
2373 if (!lov->lov_tgts[index])
2374 /* Try again with the next index */
2375 GOTO(out_statfs, rc = -EAGAIN);
2377 client_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2378 if (!lov->lov_tgts[index]->ltd_active)
2379 GOTO(out_uuid, rc = -ENODATA);
2383 GOTO(out_statfs, rc = -EINVAL);
2385 rc = obd_statfs(client_obd, &stat_buf, cfs_time_current_64() - HZ, 1);
2387 GOTO(out_statfs, rc);
2389 if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2390 GOTO(out_statfs, rc = -EFAULT);
2393 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(client_obd),
2399 obd_ioctl_freedata(buf, len);
2403 int ll_process_config(struct lustre_cfg *lcfg)
2407 struct lprocfs_static_vars lvars;
2411 lprocfs_llite_init_vars(&lvars);
2413 /* The instance name contains the sb: lustre-client-aacfe000 */
2414 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2415 if (!ptr || !*(++ptr))
2417 if (sscanf(ptr, "%lx", &x) != 1)
2420 /* This better be a real Lustre superblock! */
2421 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2423 /* Note we have not called client_common_fill_super yet, so
2424 proc fns must be able to handle that! */
2425 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2430 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2432 struct ll_sb_info *sbi;
2434 LASSERT((seq != NULL) && (vfs != NULL));
2435 sbi = ll_s2sbi(vfs->mnt_sb);
2437 if (sbi->ll_flags & LL_SBI_NOLCK)
2438 seq_puts(seq, ",nolock");
2440 if (sbi->ll_flags & LL_SBI_FLOCK)
2441 seq_puts(seq, ",flock");
2443 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2444 seq_puts(seq, ",localflock");
2446 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2447 seq_puts(seq, ",user_xattr");
2449 if (sbi->ll_flags & LL_SBI_ACL)
2450 seq_puts(seq, ",acl");