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 <obd_support.h>
56 #include "llite_internal.h"
58 cfs_mem_cache_t *ll_file_data_slab;
60 LIST_HEAD(ll_super_blocks);
61 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
63 extern struct address_space_operations ll_aops;
64 extern struct address_space_operations ll_dir_aops;
67 #define log2(n) ffz(~(n))
70 static void ll_pglist_fini(struct ll_sb_info *sbi)
75 if (sbi->ll_pglist == NULL)
78 for_each_possible_cpu(i) {
79 page = sbi->ll_pglist[i]->llpd_page;
81 sbi->ll_pglist[i] = NULL;
86 OBD_FREE(sbi->ll_pglist, sizeof(void *)*num_possible_cpus());
87 sbi->ll_pglist = NULL;
90 static int ll_pglist_init(struct ll_sb_info *sbi)
92 struct ll_pglist_data *pd;
97 OBD_ALLOC(sbi->ll_pglist, sizeof(void *) * num_possible_cpus());
98 if (sbi->ll_pglist == NULL)
101 budget = sbi->ll_async_page_max / num_online_cpus();
102 for_each_possible_cpu(i) {
103 struct page *page = alloc_pages_node(cpu_to_node(i),
110 if (color + L1_CACHE_ALIGN(sizeof(*pd)) > PAGE_SIZE)
113 pd = (struct ll_pglist_data *)(page_address(page) + color);
114 memset(pd, 0, sizeof(*pd));
115 spin_lock_init(&pd->llpd_lock);
116 INIT_LIST_HEAD(&pd->llpd_list);
118 pd->llpd_budget = budget;
120 pd->llpd_page = page;
121 atomic_set(&pd->llpd_sample_count, 0);
122 sbi->ll_pglist[i] = pd;
123 color += L1_CACHE_ALIGN(sizeof(*pd));
129 static struct ll_sb_info *ll_init_sbi(void)
131 struct ll_sb_info *sbi = NULL;
138 OBD_ALLOC(sbi, sizeof(*sbi));
142 OBD_ALLOC(sbi->ll_async_page_sample, sizeof(long)*num_possible_cpus());
143 if (sbi->ll_async_page_sample == NULL)
146 spin_lock_init(&sbi->ll_lock);
147 init_mutex(&sbi->ll_lco.lco_lock);
148 spin_lock_init(&sbi->ll_pp_extent_lock);
149 spin_lock_init(&sbi->ll_process_lock);
150 sbi->ll_rw_stats_on = 0;
153 pages = si.totalram - si.totalhigh;
154 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
155 #ifdef HAVE_BGL_SUPPORT
156 sbi->ll_async_page_max = pages / 4;
158 sbi->ll_async_page_max = pages / 2;
161 sbi->ll_async_page_max = (pages / 4) * 3;
164 lcounter_init(&sbi->ll_async_page_count);
165 spin_lock_init(&sbi->ll_async_page_reblnc_lock);
166 sbi->ll_async_page_sample_max = 64 * num_online_cpus();
167 sbi->ll_async_page_reblnc_count = 0;
168 sbi->ll_async_page_clock_hand = 0;
169 if (ll_pglist_init(sbi))
172 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
173 SBI_DEFAULT_READAHEAD_MAX);
174 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
175 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
176 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
177 sbi->ll_contention_time = SBI_DEFAULT_CONTENTION_SECONDS;
178 sbi->ll_lockless_truncate_enable = SBI_DEFAULT_LOCKLESS_TRUNCATE_ENABLE;
179 INIT_LIST_HEAD(&sbi->ll_conn_chain);
180 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
182 ll_generate_random_uuid(uuid);
183 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
184 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
186 spin_lock(&ll_sb_lock);
187 list_add_tail(&sbi->ll_list, &ll_super_blocks);
188 spin_unlock(&ll_sb_lock);
190 #ifdef ENABLE_CHECKSUM
191 sbi->ll_flags |= LL_SBI_DATA_CHECKSUM;
193 #ifdef ENABLE_LLITE_CHECKSUM
194 sbi->ll_flags |= LL_SBI_LLITE_CHECKSUM;
197 #ifdef HAVE_LRU_RESIZE_SUPPORT
198 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
201 #ifdef HAVE_EXPORT___IGET
202 INIT_LIST_HEAD(&sbi->ll_deathrow);
203 spin_lock_init(&sbi->ll_deathrow_lock);
205 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
206 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
207 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
210 /* metadata statahead is enabled by default */
211 sbi->ll_sa_max = LL_SA_RPC_DEF;
216 if (sbi->ll_async_page_sample)
217 OBD_FREE(sbi->ll_async_page_sample,
218 sizeof(long) * num_possible_cpus());
220 OBD_FREE(sbi, sizeof(*sbi));
224 void ll_free_sbi(struct super_block *sb)
226 struct ll_sb_info *sbi = ll_s2sbi(sb);
230 spin_lock(&ll_sb_lock);
231 list_del(&sbi->ll_list);
232 spin_unlock(&ll_sb_lock);
233 /* dont allow find cache via sb list first */
235 lcounter_destroy(&sbi->ll_async_page_count);
236 OBD_FREE(sbi->ll_async_page_sample,
237 sizeof(long) * num_possible_cpus());
238 OBD_FREE(sbi, sizeof(*sbi));
243 static struct dentry_operations ll_d_root_ops = {
244 #ifdef DCACHE_LUSTRE_INVALID
245 .d_compare = ll_dcompare,
247 .d_revalidate = ll_revalidate_nd,
250 static int client_common_fill_super(struct super_block *sb,
251 char *mdc, char *osc)
253 struct inode *root = 0;
254 struct ll_sb_info *sbi = ll_s2sbi(sb);
255 struct obd_device *obd;
256 struct ll_fid rootfid;
257 struct obd_statfs osfs;
258 struct ptlrpc_request *request = NULL;
259 struct lustre_handle osc_conn = {0, };
260 struct lustre_handle mdc_conn = {0, };
262 struct obd_connect_data *data = NULL;
266 obd = class_name2obd(mdc);
268 CERROR("MDC %s: not setup or attached\n", mdc);
272 OBD_ALLOC(data, sizeof(*data));
276 if (proc_lustre_fs_root) {
277 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
280 CERROR("could not register mount in /proc/fs/lustre\n");
283 /* indicate the features supported by this client */
284 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_IBITS |
285 OBD_CONNECT_JOIN | OBD_CONNECT_ATTRFID |
286 OBD_CONNECT_NODEVOH | OBD_CONNECT_CANCELSET |
287 OBD_CONNECT_AT | OBD_CONNECT_FID |
288 OBD_CONNECT_VBR | OBD_CONNECT_LOV_V3;
289 #ifdef HAVE_LRU_RESIZE_SUPPORT
290 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
291 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
293 #ifdef CONFIG_FS_POSIX_ACL
294 data->ocd_connect_flags |= OBD_CONNECT_ACL;
296 data->ocd_ibits_known = MDS_INODELOCK_FULL;
297 data->ocd_version = LUSTRE_VERSION_CODE;
299 if (sb->s_flags & MS_RDONLY)
300 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
301 if (sbi->ll_flags & LL_SBI_USER_XATTR)
302 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
304 #ifdef HAVE_MS_FLOCK_LOCK
305 /* force vfs to use lustre handler for flock() calls - bug 10743 */
306 sb->s_flags |= MS_FLOCK_LOCK;
309 if (sbi->ll_flags & LL_SBI_FLOCK)
310 sbi->ll_fop = &ll_file_operations_flock;
311 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
312 sbi->ll_fop = &ll_file_operations;
314 sbi->ll_fop = &ll_file_operations_noflock;
317 err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid, data, &sbi->ll_mdc_exp);
319 LCONSOLE_ERROR_MSG(0x14f, "An MDT (mdc %s) is performing "
320 "recovery, of which this client is not a "
321 "part. Please wait for recovery to complete,"
322 " abort, or time out.\n", mdc);
325 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
329 err = obd_fid_init(sbi->ll_mdc_exp);
333 err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
337 /* MDC connect is surely finished by now because we actually sent
338 * a statfs RPC, otherwise obd_connect() is asynchronous. */
339 *data = class_exp2cliimp(sbi->ll_mdc_exp)->imp_connect_data;
341 LASSERT(osfs.os_bsize);
342 sb->s_blocksize = osfs.os_bsize;
343 sb->s_blocksize_bits = log2(osfs.os_bsize);
344 sb->s_magic = LL_SUPER_MAGIC;
346 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
347 * retval = in_file->f_op->sendfile(...);
349 * retval = -EOVERFLOW;
351 * it will check if *ppos is greater than max. However, max equals to
352 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
353 * has been defined as a signed long long ineger in linux kernel. */
354 #if BITS_PER_LONG == 64
355 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
357 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
359 sbi->ll_namelen = osfs.os_namelen;
360 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
362 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
363 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
364 LCONSOLE_INFO("Disabling user_xattr feature because "
365 "it is not supported on the server\n");
366 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
369 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
371 sb->s_flags |= MS_POSIXACL;
373 sbi->ll_flags |= LL_SBI_ACL;
375 sbi->ll_flags &= ~LL_SBI_ACL;
377 if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
378 sbi->ll_flags |= LL_SBI_JOIN;
380 obd = class_name2obd(osc);
382 CERROR("OSC %s: not setup or attached\n", osc);
383 GOTO(out_mdc, err = -ENODEV);
386 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_GRANT |
387 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
388 OBD_CONNECT_SRVLOCK | OBD_CONNECT_CANCELSET|
389 OBD_CONNECT_AT | OBD_CONNECT_FID |
390 OBD_CONNECT_VBR | OBD_CONNECT_TRUNCLOCK|
391 OBD_CONNECT_GRANT_SHRINK;
393 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
394 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
395 * disabled by default, because it can still be enabled on the
396 * fly via /proc. As a consequence, we still need to come to an
397 * agreement on the supported algorithms at connect time */
398 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
400 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
401 data->ocd_cksum_types = OBD_CKSUM_ADLER;
403 /* send the list of supported checksum types */
404 data->ocd_cksum_types = OBD_CKSUM_ALL;
407 #ifdef HAVE_LRU_RESIZE_SUPPORT
408 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
409 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
412 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
413 "ocd_grant: %d\n", data->ocd_connect_flags,
414 data->ocd_version, data->ocd_grant);
416 obd->obd_upcall.onu_owner = &sbi->ll_lco;
417 obd->obd_upcall.onu_upcall = ll_ocd_update;
418 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
420 obd_register_lock_cancel_cb(obd, ll_extent_lock_cancel_cb);
421 obd_register_page_removal_cb(obd, ll_page_removal_cb, ll_pin_extent_cb);
423 err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid, data, &sbi->ll_osc_exp);
425 LCONSOLE_ERROR_MSG(0x150, "An OST (osc %s) is performing "
426 "recovery, of which this client is not a "
427 "part. Please wait for recovery to "
428 "complete, abort, or time out.\n", osc);
431 CERROR("cannot connect to %s: rc = %d\n", osc, err);
435 mutex_down(&sbi->ll_lco.lco_lock);
436 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
437 sbi->ll_lco.lco_mdc_exp = sbi->ll_mdc_exp;
438 sbi->ll_lco.lco_osc_exp = sbi->ll_osc_exp;
439 mutex_up(&sbi->ll_lco.lco_lock);
441 err = mdc_init_ea_size(sbi->ll_mdc_exp, sbi->ll_osc_exp);
443 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
447 err = obd_prep_async_page(sbi->ll_osc_exp, NULL, NULL, NULL,
448 0, NULL, NULL, NULL, 0, NULL);
450 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
451 "filesystem. There must be at least one "
452 "active OST for a client to start.\n");
456 if (!ll_async_page_slab) {
457 ll_async_page_slab_size =
458 size_round(sizeof(struct ll_async_page)) + err;
459 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
460 ll_async_page_slab_size,
462 if (!ll_async_page_slab)
463 GOTO(out_osc, err = -ENOMEM);
466 err = mdc_getstatus(sbi->ll_mdc_exp, &rootfid);
468 CERROR("cannot mds_connect: rc = %d\n", err);
471 CDEBUG(D_SUPER, "rootfid "LPU64":"DFID"\n", rootfid.id,
472 PFID((struct lu_fid*)&rootfid));
473 sbi->ll_rootino = rootfid.id;
475 sb->s_op = &lustre_super_operations;
476 #if THREAD_SIZE >= 8192
477 /* Disable the NFS export because of stack overflow
478 * when THREAD_SIZE < 8192. Please refer to 17630. */
479 sb->s_export_op = &lustre_export_operations;
483 * XXX: move this to after cbd setup? */
484 err = mdc_getattr(sbi->ll_mdc_exp, &rootfid,
485 OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS |
486 (sbi->ll_flags & LL_SBI_ACL ? OBD_MD_FLACL : 0),
489 CERROR("mdc_getattr failed for root: rc = %d\n", err);
493 err = mdc_req2lustre_md(request, REPLY_REC_OFF, sbi->ll_osc_exp, &md);
495 CERROR("failed to understand root inode md: rc = %d\n",err);
496 ptlrpc_req_finished (request);
500 LASSERT(sbi->ll_rootino != 0);
501 root = ll_iget(sb, ll_fid_build_ino(sbi, &rootfid), &md);
503 ptlrpc_req_finished(request);
505 if (root == NULL || is_bad_inode(root)) {
506 mdc_free_lustre_md(sbi->ll_osc_exp, &md);
507 CERROR("lustre_lite: bad iget4 for root\n");
508 GOTO(out_root, err = -EBADF);
511 err = ll_close_thread_start(&sbi->ll_lcq);
513 CERROR("cannot start close thread: rc %d\n", err);
517 checksum = sbi->ll_flags & LL_SBI_DATA_CHECKSUM;
518 err = obd_set_info_async(sbi->ll_osc_exp, sizeof(KEY_CHECKSUM),
519 KEY_CHECKSUM, sizeof(checksum),
522 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
523 backing dev info assigned to inode mapping is used for
524 determining maximal readahead. */
525 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
526 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
527 /* bug 2805 - set VM readahead to zero */
528 vm_max_readahead = vm_min_readahead = 0;
531 sb->s_root = d_alloc_root(root);
533 OBD_FREE(data, sizeof(*data));
534 sb->s_root->d_op = &ll_d_root_ops;
536 sbi->ll_sdev_orig = sb->s_dev;
537 /* We set sb->s_dev equal on all lustre clients in order to support
538 * NFS export clustering. NFSD requires that the FSID be the same
540 /* s_dev is also used in lt_compare() to compare two fs, but that is
541 * only a node-local comparison. */
542 sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
543 strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
551 obd_disconnect(sbi->ll_osc_exp);
552 sbi->ll_osc_exp = NULL;
554 obd = class_name2obd(osc);
555 obd_unregister_lock_cancel_cb(obd, ll_extent_lock_cancel_cb);
556 obd_unregister_page_removal_cb(obd, ll_page_removal_cb);
558 obd_fid_fini(sbi->ll_mdc_exp);
559 obd_disconnect(sbi->ll_mdc_exp);
560 sbi->ll_mdc_exp = NULL;
563 OBD_FREE(data, sizeof(*data));
564 lprocfs_unregister_mountpoint(sbi);
568 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
572 *lmmsize = obd_size_diskmd(sbi->ll_osc_exp, NULL);
574 rc = obd_get_info(sbi->ll_mdc_exp, sizeof(KEY_MAX_EASIZE),
575 KEY_MAX_EASIZE, &size, lmmsize, NULL);
577 CERROR("Get max mdsize error rc %d \n", rc);
582 void ll_dump_inode(struct inode *inode)
584 struct list_head *tmp;
585 int dentry_count = 0;
587 LASSERT(inode != NULL);
589 list_for_each(tmp, &inode->i_dentry)
592 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
593 inode, ll_i2mdcexp(inode)->exp_obd->obd_name, inode->i_ino,
594 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
597 void lustre_dump_dentry(struct dentry *dentry, int recur)
599 struct list_head *tmp;
602 LASSERT(dentry != NULL);
604 list_for_each(tmp, &dentry->d_subdirs)
607 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
608 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
609 dentry->d_name.len, dentry->d_name.name,
610 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
611 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
612 dentry->d_flags, dentry->d_fsdata, subdirs);
613 if (dentry->d_inode != NULL)
614 ll_dump_inode(dentry->d_inode);
619 list_for_each(tmp, &dentry->d_subdirs) {
620 struct dentry *d = list_entry(tmp, struct dentry, d_child);
621 lustre_dump_dentry(d, recur - 1);
625 #ifdef HAVE_EXPORT___IGET
626 static void prune_dir_dentries(struct inode *inode)
628 struct dentry *dentry, *prev = NULL;
630 /* due to lustre specific logic, a directory
631 * can have few dentries - a bug from VFS POV */
633 spin_lock(&dcache_lock);
634 if (!list_empty(&inode->i_dentry)) {
635 dentry = list_entry(inode->i_dentry.prev,
636 struct dentry, d_alias);
637 /* in order to prevent infinite loops we
638 * break if previous dentry is busy */
639 if (dentry != prev) {
642 spin_unlock(&dcache_lock);
644 /* try to kill all child dentries */
645 shrink_dcache_parent(dentry);
648 /* now try to get rid of current dentry */
649 d_prune_aliases(inode);
653 spin_unlock(&dcache_lock);
656 static void prune_deathrow_one(struct ll_inode_info *lli)
658 struct inode *inode = ll_info2i(lli);
660 /* first, try to drop any dentries - they hold a ref on the inode */
661 if (S_ISDIR(inode->i_mode))
662 prune_dir_dentries(inode);
664 d_prune_aliases(inode);
667 /* if somebody still uses it, leave it */
668 LASSERT(atomic_read(&inode->i_count) > 0);
669 if (atomic_read(&inode->i_count) > 1)
672 CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
673 inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
675 /* seems nobody uses it anymore */
683 static void prune_deathrow(struct ll_sb_info *sbi, int try)
685 struct ll_inode_info *lli;
689 if (need_resched() && try)
693 if (!spin_trylock(&sbi->ll_deathrow_lock))
696 spin_lock(&sbi->ll_deathrow_lock);
701 if (!list_empty(&sbi->ll_deathrow)) {
702 lli = list_entry(sbi->ll_deathrow.next,
703 struct ll_inode_info,
705 list_del_init(&lli->lli_dead_list);
706 if (!list_empty(&sbi->ll_deathrow))
709 spin_unlock(&sbi->ll_deathrow_lock);
712 prune_deathrow_one(lli);
714 } while (empty == 0);
716 #else /* !HAVE_EXPORT___IGET */
717 #define prune_deathrow(sbi, try) do {} while (0)
718 #endif /* HAVE_EXPORT___IGET */
720 void client_common_put_super(struct super_block *sb)
722 struct ll_sb_info *sbi = ll_s2sbi(sb);
725 ll_close_thread_shutdown(sbi->ll_lcq);
727 lprocfs_unregister_mountpoint(sbi);
729 /* destroy inodes in deathrow */
730 prune_deathrow(sbi, 0);
732 list_del(&sbi->ll_conn_chain);
734 /* callbacks is cleared after disconnect each target */
735 obd_disconnect(sbi->ll_osc_exp);
736 sbi->ll_osc_exp = NULL;
738 obd_fid_fini(sbi->ll_mdc_exp);
739 obd_disconnect(sbi->ll_mdc_exp);
740 sbi->ll_mdc_exp = NULL;
745 void ll_kill_super(struct super_block *sb)
747 struct ll_sb_info *sbi;
752 if (!(sb->s_flags & MS_ACTIVE))
756 /* we need restore s_dev from changed for clustred NFS before put_super
757 * because new kernels have cached s_dev and change sb->s_dev in
758 * put_super not affected real removing devices */
760 sb->s_dev = sbi->ll_sdev_orig;
764 char *ll_read_opt(const char *opt, char *data)
770 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
771 if (strncmp(opt, data, strlen(opt)))
773 if ((value = strchr(data, '=')) == NULL)
777 OBD_ALLOC(retval, strlen(value) + 1);
779 CERROR("out of memory!\n");
783 memcpy(retval, value, strlen(value)+1);
784 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
788 static inline int ll_set_opt(const char *opt, char *data, int fl)
790 if (strncmp(opt, data, strlen(opt)) != 0)
796 /* non-client-specific mount options are parsed in lmd_parse */
797 static int ll_options(char *options, int *flags)
800 char *s1 = options, *s2;
806 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
809 CDEBUG(D_SUPER, "next opt=%s\n", s1);
810 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
815 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
820 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
825 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
830 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
835 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
840 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
842 /* Ignore deprecated mount option. The client will
843 * always try to mount with ACL support, whether this
844 * is used depends on whether server supports it. */
845 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
846 "mount option 'acl'.\n");
849 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
851 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
852 "mount option 'noacl'.\n");
856 tmp = ll_set_opt("checksum", s1, LL_SBI_DATA_CHECKSUM);
861 tmp = ll_set_opt("nochecksum", s1, LL_SBI_DATA_CHECKSUM);
867 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
872 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
877 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
882 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
887 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
893 s2 = strchr(s1, ',');
901 void ll_lli_init(struct ll_inode_info *lli)
903 lli->lli_inode_magic = LLI_INODE_MAGIC;
904 sema_init(&lli->lli_size_sem, 1);
905 sema_init(&lli->lli_write_sem, 1);
907 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
908 spin_lock_init(&lli->lli_lock);
909 sema_init(&lli->lli_och_sem, 1);
910 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
911 lli->lli_mds_exec_och = NULL;
912 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
913 lli->lli_open_fd_exec_count = 0;
914 INIT_LIST_HEAD(&lli->lli_dead_list);
915 #ifdef HAVE_CLOSE_THREAD
916 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
918 init_rwsem(&lli->lli_truncate_rwsem);
922 #define MDCDEV "mdc_dev"
923 static int old_lustre_process_log(struct super_block *sb, char *newprofile,
924 struct config_llog_instance *cfg)
926 struct lustre_sb_info *lsi = s2lsi(sb);
927 struct obd_device *obd;
928 struct lustre_handle mdc_conn = {0, };
929 struct obd_export *exp;
930 char *ptr, *mdt, *profile;
931 char niduuid[10] = "mdtnid0";
933 struct obd_uuid mdc_uuid;
934 struct llog_ctxt *ctxt;
935 struct obd_connect_data ocd = { 0 };
937 int i, rc = 0, recov_bk = 1, failnodes = 0;
940 ll_generate_random_uuid(uuid);
941 class_uuid_unparse(uuid, &mdc_uuid);
942 CDEBUG(D_HA, "generated uuid: %s\n", mdc_uuid.uuid);
944 /* Figure out the old mdt and profile name from new-style profile
945 ("lustre" from "mds/lustre-client") */
947 profile = strchr(mdt, '/');
948 if (profile == NULL) {
949 CDEBUG(D_CONFIG, "Can't find MDT name in %s\n", newprofile);
950 GOTO(out, rc = -EINVAL);
954 ptr = strrchr(profile, '-');
956 CDEBUG(D_CONFIG, "Can't find client name in %s\n", newprofile);
957 GOTO(out, rc = -EINVAL);
961 LCONSOLE_WARN("This looks like an old mount command; I will try to "
962 "contact MDT '%s' for profile '%s'\n", mdt, profile);
964 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
966 ptr = lsi->lsi_lmd->lmd_dev;
967 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
968 rc = do_lcfg(MDCDEV, nid, LCFG_ADD_UUID, niduuid, 0,0,0);
970 /* Stop at the first failover nid */
975 CERROR("No valid MDT nids found.\n");
976 GOTO(out, rc = -EINVAL);
980 rc = do_lcfg(MDCDEV, 0, LCFG_ATTACH, LUSTRE_MDC_NAME,mdc_uuid.uuid,0,0);
982 GOTO(out_del_uuid, rc);
984 rc = do_lcfg(MDCDEV, 0, LCFG_SETUP, mdt, niduuid, 0, 0);
986 LCONSOLE_ERROR_MSG(0x153, "I couldn't establish a connection "
987 "with the MDT. Check that the MDT host NID "
988 "is correct and the networks are up.\n");
989 GOTO(out_detach, rc);
992 obd = class_name2obd(MDCDEV);
994 GOTO(out_cleanup, rc = -EINVAL);
996 /* Add any failover nids */
997 while (*ptr == ':') {
998 /* New failover node */
999 sprintf(niduuid, "mdtnid%d", failnodes);
1001 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
1003 rc = do_lcfg(MDCDEV, nid, LCFG_ADD_UUID, niduuid,0,0,0);
1005 CERROR("Add uuid for %s failed %d\n",
1006 libcfs_nid2str(nid), rc);
1011 rc = do_lcfg(MDCDEV, 0, LCFG_ADD_CONN, niduuid, 0, 0,0);
1013 CERROR("Add conn for %s failed %d\n",
1014 libcfs_nid2str(nid), rc);
1022 /* Try all connections, but only once. */
1023 rc = obd_set_info_async(obd->obd_self_export,
1024 sizeof(KEY_INIT_RECOV_BACKUP), KEY_INIT_RECOV_BACKUP,
1025 sizeof(recov_bk), &recov_bk, NULL);
1027 GOTO(out_cleanup, rc);
1029 /* If we don't have this then an ACL MDS will refuse the connection */
1030 ocd.ocd_connect_flags = OBD_CONNECT_ACL;
1032 rc = obd_connect(&mdc_conn, obd, &mdc_uuid, &ocd, &exp);
1034 CERROR("cannot connect to %s: rc = %d\n", mdt, rc);
1035 GOTO(out_cleanup, rc);
1038 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1040 cfg->cfg_flags |= CFG_F_COMPAT146;
1043 rc = class_config_parse_llog(ctxt, profile, cfg);
1046 * For debugging, it's useful to just dump the log
1048 rc = class_config_dump_llog(ctxt, profile, cfg);
1050 llog_ctxt_put(ctxt);
1053 /* Set the caller's profile name to the old-style */
1054 memcpy(newprofile, profile, strlen(profile) + 1);
1058 LCONSOLE_ERROR_MSG(0x154, "%s: The configuration '%s' could not"
1059 " be read from the MDT '%s'. Make sure this"
1060 " client and the MDT are running compatible "
1061 "versions of Lustre.\n",
1062 obd->obd_name, profile, mdt);
1065 LCONSOLE_ERROR_MSG(0x155, "%s: The configuration '%s' could not"
1066 " be read from the MDT '%s'. This may be "
1067 "the result of communication errors between "
1068 "the client and the MDT, or if the MDT is "
1069 "not running.\n", obd->obd_name, profile,
1074 /* We don't so much care about errors in cleaning up the config llog
1075 * connection, as we have already read the config by this point. */
1076 obd_disconnect(exp);
1079 do_lcfg(MDCDEV, 0, LCFG_CLEANUP, 0, 0, 0, 0);
1082 do_lcfg(MDCDEV, 0, LCFG_DETACH, 0, 0, 0, 0);
1085 /* class_add_uuid adds a nid even if the same uuid exists; we might
1086 delete any copy here. So they all better match. */
1087 for (i = 0; i < failnodes; i++) {
1088 sprintf(niduuid, "mdtnid%d", i);
1089 do_lcfg(MDCDEV, 0, LCFG_DEL_UUID, niduuid, 0, 0, 0);
1091 /* class_import_put will get rid of the additional connections */
1095 /* end COMPAT_146 */
1097 int ll_fill_super(struct super_block *sb)
1099 struct lustre_profile *lprof = NULL;
1100 struct lustre_sb_info *lsi = s2lsi(sb);
1101 struct ll_sb_info *sbi;
1102 char *osc = NULL, *mdc = NULL;
1103 char *profilenm = get_profile_name(sb);
1104 struct config_llog_instance cfg = {0, };
1105 char ll_instance[sizeof(sb) * 2 + 1];
1108 char pseudo[32] = { 0 };
1111 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1115 /* client additional sb info */
1116 lsi->lsi_llsbi = sbi = ll_init_sbi();
1122 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1124 GOTO(out_free, err);
1126 /* Generate a string unique to this super, in case some joker tries
1127 to mount the same fs at two mount points.
1128 Use the address of the super itself.*/
1129 sprintf(ll_instance, "%p", sb);
1130 cfg.cfg_instance = ll_instance;
1131 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1134 /* set up client obds */
1135 if (strchr(profilenm, '/') != NULL) /* COMPAT_146 */
1136 err = -EINVAL; /* skip error messages, use old config code */
1138 err = lustre_process_log(sb, profilenm, &cfg);
1143 oldnamelen = strlen(profilenm) + 1;
1144 /* Temp storage for 1.4.6 profile name */
1145 OBD_ALLOC(oldname, oldnamelen);
1147 memcpy(oldname, profilenm, oldnamelen);
1148 rc = old_lustre_process_log(sb, oldname, &cfg);
1150 /* That worked - update the profile name
1153 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1154 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1155 OBD_ALLOC(lsi->lsi_lmd->lmd_profile,
1156 strlen(oldname) + 1);
1157 if (!lsi->lsi_lmd->lmd_profile) {
1158 OBD_FREE(oldname, oldnamelen);
1159 GOTO(out_free, err = -ENOMEM);
1161 memcpy(lsi->lsi_lmd->lmd_profile, oldname,
1162 strlen(oldname) + 1);
1163 profilenm = get_profile_name(sb);
1164 /* Don't ever try to recover the MGS */
1165 rc = ptlrpc_set_import_active(
1166 lsi->lsi_mgc->u.cli.cl_import, 0);
1168 OBD_FREE(oldname, oldnamelen);
1171 /* end COMPAT_146 */
1173 CERROR("Unable to process log: %d\n", err);
1174 GOTO(out_free, err);
1177 lprof = class_get_profile(profilenm);
1178 if (lprof == NULL) {
1179 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1180 " read from the MGS. Does that filesystem "
1181 "exist?\n", profilenm);
1182 GOTO(out_free, err = -EINVAL);
1186 * The configuration for 1.8 client and 2.0 client are different.
1187 * 2.0 introduces lmv, but 1.8 directly uses mdc.
1188 * Here, we will hack to use proper name for mdc if needed.
1194 save = lprof->lp_mdc;
1195 fsname_end = strrchr(save, '-');
1197 namelen = fsname_end - save;
1198 if (strcmp(fsname_end, "-clilmv") == 0) {
1199 strncpy(pseudo, save, namelen);
1200 strcat(pseudo, "-MDT0000-mdc");
1201 lprof->lp_mdc = pseudo;
1202 CDEBUG(D_INFO, "1.8.x connecting to 2.0: lmv=%s"
1203 " new mdc=%s\n", save, pseudo);
1208 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1209 lprof->lp_mdc, lprof->lp_osc);
1211 OBD_ALLOC(osc, strlen(lprof->lp_osc) +
1212 strlen(ll_instance) + 2);
1214 GOTO(out_free, err = -ENOMEM);
1215 sprintf(osc, "%s-%s", lprof->lp_osc, ll_instance);
1217 OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
1218 strlen(ll_instance) + 2);
1220 GOTO(out_free, err = -ENOMEM);
1221 sprintf(mdc, "%s-%s", lprof->lp_mdc, ll_instance);
1223 /* connections, registrations, sb setup */
1224 err = client_common_fill_super(sb, mdc, osc);
1228 lprof->lp_mdc = save;
1230 OBD_FREE(mdc, strlen(mdc) + 1);
1232 OBD_FREE(osc, strlen(osc) + 1);
1236 LCONSOLE_WARN("Client %s has started\n", profilenm);
1239 } /* ll_fill_super */
1242 void ll_put_super(struct super_block *sb)
1244 struct config_llog_instance cfg;
1245 char ll_instance[sizeof(sb) * 2 + 1];
1246 struct obd_device *obd;
1247 struct lustre_sb_info *lsi = s2lsi(sb);
1248 struct ll_sb_info *sbi = ll_s2sbi(sb);
1249 char *profilenm = get_profile_name(sb);
1250 int force = 1, next;
1253 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1255 sprintf(ll_instance, "%p", sb);
1256 cfg.cfg_instance = ll_instance;
1257 lustre_end_log(sb, NULL, &cfg);
1259 if (sbi->ll_mdc_exp) {
1260 obd = class_exp2obd(sbi->ll_mdc_exp);
1262 force = obd->obd_force;
1265 /* We need to set force before the lov_disconnect in
1266 lustre_common_put_super, since l_d cleans up osc's as well. */
1269 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1271 obd->obd_force = force;
1276 /* Only if client_common_fill_super succeeded */
1277 client_common_put_super(sb);
1281 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1282 class_manual_cleanup(obd);
1286 class_del_profile(profilenm);
1289 lsi->lsi_llsbi = NULL;
1291 lustre_common_put_super(sb);
1293 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1298 } /* client_put_super */
1300 #if defined(HAVE_REGISTER_CACHE) || defined(HAVE_SHRINKER_CACHE)
1302 #if defined(HAVE_CACHE_RETURN_INT)
1307 ll_shrink_cache(int priority, unsigned int gfp_mask)
1309 struct ll_sb_info *sbi;
1312 /* don't race with umount */
1313 spin_lock(&ll_sb_lock);
1314 list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1315 count += llap_shrink_cache(sbi, priority);
1316 spin_unlock(&ll_sb_lock);
1318 #if defined(HAVE_CACHE_RETURN_INT)
1323 struct cache_definition ll_cache_definition = {
1324 .name = "llap_cache",
1325 .shrink = ll_shrink_cache
1327 #endif /* HAVE_REGISTER_CACHE || HAVE_SHRINKER_CACHE */
1329 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1331 struct inode *inode = NULL;
1332 /* NOTE: we depend on atomic igrab() -bzzz */
1333 lock_res_and_lock(lock);
1334 if (lock->l_ast_data) {
1335 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1336 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1337 inode = igrab(lock->l_ast_data);
1339 inode = lock->l_ast_data;
1340 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1342 lock, __FILE__, __func__, __LINE__,
1343 "l_ast_data %p is bogus: magic %08x",
1344 lock->l_ast_data, lli->lli_inode_magic);
1348 unlock_res_and_lock(lock);
1352 static int null_if_equal(struct ldlm_lock *lock, void *data)
1354 if (data == lock->l_ast_data) {
1355 lock->l_ast_data = NULL;
1357 if (lock->l_req_mode != lock->l_granted_mode)
1358 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1361 return LDLM_ITER_CONTINUE;
1364 void ll_clear_inode(struct inode *inode)
1367 struct ll_inode_info *lli = ll_i2info(inode);
1368 struct ll_sb_info *sbi = ll_i2sbi(inode);
1371 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1372 inode->i_generation, inode);
1374 if (S_ISDIR(inode->i_mode)) {
1375 /* these should have been cleared in ll_file_release */
1376 LASSERT(lli->lli_sai == NULL);
1377 LASSERT(lli->lli_opendir_key == NULL);
1378 LASSERT(lli->lli_opendir_pid == 0);
1381 ll_inode2fid(&fid, inode);
1382 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1383 mdc_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
1385 LASSERT(!lli->lli_open_fd_write_count);
1386 LASSERT(!lli->lli_open_fd_read_count);
1387 LASSERT(!lli->lli_open_fd_exec_count);
1389 if (lli->lli_mds_write_och)
1390 ll_mdc_real_close(inode, FMODE_WRITE);
1391 if (lli->lli_mds_exec_och) {
1393 CERROR("No FMODE exec, bug exec och is present for "
1394 "inode %ld\n", inode->i_ino);
1395 ll_mdc_real_close(inode, FMODE_EXEC);
1397 if (lli->lli_mds_read_och)
1398 ll_mdc_real_close(inode, FMODE_READ);
1402 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
1403 null_if_equal, inode);
1405 obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
1406 lli->lli_smd = NULL;
1409 if (lli->lli_symlink_name) {
1410 OBD_FREE(lli->lli_symlink_name,
1411 strlen(lli->lli_symlink_name) + 1);
1412 lli->lli_symlink_name = NULL;
1415 #ifdef CONFIG_FS_POSIX_ACL
1416 if (lli->lli_posix_acl) {
1417 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1418 posix_acl_release(lli->lli_posix_acl);
1419 lli->lli_posix_acl = NULL;
1423 lli->lli_inode_magic = LLI_INODE_DEAD;
1425 #ifdef HAVE_EXPORT___IGET
1426 spin_lock(&sbi->ll_deathrow_lock);
1427 list_del_init(&lli->lli_dead_list);
1428 spin_unlock(&sbi->ll_deathrow_lock);
1433 static int ll_setattr_do_truncate(struct inode *inode, loff_t new_size)
1435 struct ll_sb_info *sbi = ll_i2sbi(inode);
1436 struct ll_inode_info *lli = ll_i2info(inode);
1437 struct lov_stripe_md *lsm = lli->lli_smd;
1439 ldlm_policy_data_t policy = { .l_extent = {new_size,
1441 struct lustre_handle lockh = { 0 };
1442 int local_lock = 1; /* 0 - no local lock;
1443 * 1 - lock taken by lock_extent;
1444 * 2 - by obd_match*/
1448 UNLOCK_INODE_MUTEX(inode);
1449 UP_WRITE_I_ALLOC_SEM(inode);
1451 down_write(&lli->lli_truncate_rwsem);
1452 if (sbi->ll_lockless_truncate_enable &&
1453 (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK)) {
1456 ast_flags = LDLM_FL_BLOCK_GRANTED;
1457 rc = obd_match(sbi->ll_osc_exp, lsm, LDLM_EXTENT,
1458 &policy, LCK_PW, &ast_flags, inode, &lockh,
1464 /* clear the lock handle as it not cleared
1465 * by obd_match if no matched lock found */
1467 if (rc == 0 && n_matches == 0) {
1469 rc = ll_file_punch(inode, new_size, 1);
1473 if (local_lock == 1) {
1474 /* XXX when we fix the AST intents to pass the discard-range
1475 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1477 ast_flags = (new_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
1478 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
1480 if (unlikely(rc != 0))
1484 LOCK_INODE_MUTEX(inode);
1485 DOWN_WRITE_I_ALLOC_SEM(inode);
1486 if (likely(rc == 0)) {
1487 /* Only ll_inode_size_lock is taken at this level.
1488 * lov_stripe_lock() is grabbed by ll_truncate() only over
1489 * call to obd_adjust_kms(). If vmtruncate returns 0, then
1490 * ll_truncate dropped ll_inode_size_lock() */
1491 ll_inode_size_lock(inode, 0);
1493 set_bit(LLI_F_SRVLOCK, &lli->lli_flags);
1494 rc = vmtruncate(inode, new_size);
1495 clear_bit(LLI_F_SRVLOCK, &lli->lli_flags);
1497 LASSERT(SEM_COUNT(&lli->lli_size_sem) <= 0);
1498 ll_inode_size_unlock(inode, 0);
1504 err = (local_lock == 2) ?
1505 obd_cancel(sbi->ll_osc_exp, lsm, LCK_PW, &lockh):
1506 ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1507 if (unlikely(err != 0)){
1508 CERROR("extent unlock failed: err=%d,"
1509 " unlock method =%d\n", err, local_lock);
1514 up_write(&lli->lli_truncate_rwsem);
1518 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1519 * object(s) determine the file size and mtime. Otherwise, the MDS will
1520 * keep these values until such a time that objects are allocated for it.
1521 * We do the MDS operations first, as it is checking permissions for us.
1522 * We don't to the MDS RPC if there is nothing that we want to store there,
1523 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1524 * going to do an RPC anyways.
1526 * If we are doing a truncate, we will send the mtime and ctime updates
1527 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1528 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1531 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1533 struct ll_inode_info *lli = ll_i2info(inode);
1534 struct lov_stripe_md *lsm = lli->lli_smd;
1535 struct ll_sb_info *sbi = ll_i2sbi(inode);
1536 struct ptlrpc_request *request = NULL;
1537 struct mdc_op_data *op_data;
1538 struct lustre_md md;
1539 int ia_valid = attr->ia_valid;
1543 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1545 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1547 if (ia_valid & ATTR_SIZE) {
1548 if (attr->ia_size > ll_file_maxbytes(inode)) {
1549 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1550 attr->ia_size, ll_file_maxbytes(inode));
1554 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1557 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1558 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1559 if (current->fsuid != inode->i_uid &&
1560 !cfs_capable(CFS_CAP_FOWNER))
1564 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1565 if (attr->ia_valid & ATTR_CTIME) {
1566 attr->ia_ctime = CURRENT_TIME;
1567 attr->ia_valid |= ATTR_CTIME_SET;
1569 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1570 attr->ia_atime = CURRENT_TIME;
1571 attr->ia_valid |= ATTR_ATIME_SET;
1573 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1574 attr->ia_mtime = CURRENT_TIME;
1575 attr->ia_valid |= ATTR_MTIME_SET;
1578 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1579 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1580 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1583 /* NB: ATTR_SIZE will only be set after this point if the size
1584 * resides on the MDS, ie, this file has no objects. */
1586 attr->ia_valid &= ~ATTR_SIZE;
1588 OBD_ALLOC_PTR(op_data);
1589 if (NULL == op_data)
1591 /* We always do an MDS RPC, even if we're only changing the size;
1592 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1593 ll_prepare_mdc_op_data(op_data, inode, NULL, NULL, 0, 0, NULL);
1595 rc = mdc_setattr(sbi->ll_mdc_exp, op_data,
1596 attr, NULL, 0, NULL, 0, &request);
1599 ptlrpc_req_finished(request);
1600 if (rc == -ENOENT) {
1602 /* Unlinked special device node? Or just a race?
1603 * Pretend we done everything. */
1604 if (!S_ISREG(inode->i_mode) &&
1605 !S_ISDIR(inode->i_mode))
1606 rc = inode_setattr(inode, attr);
1607 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY)
1608 CERROR("mdc_setattr fails: rc = %d\n", rc);
1609 OBD_FREE_PTR(op_data);
1612 OBD_FREE_PTR(op_data);
1614 rc = mdc_req2lustre_md(request, REPLY_REC_OFF, sbi->ll_osc_exp, &md);
1616 ptlrpc_req_finished(request);
1620 /* We call inode_setattr to adjust timestamps.
1621 * If there is at least some data in file, we cleared ATTR_SIZE above to
1622 * avoid invoking vmtruncate, otherwise it is important to call
1623 * vmtruncate in inode_setattr to update inode->i_size (bug 6196) */
1624 rc = inode_setattr(inode, attr);
1626 ll_update_inode(inode, &md);
1627 ptlrpc_req_finished(request);
1629 if (!lsm || !S_ISREG(inode->i_mode)) {
1630 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1634 /* We really need to get our PW lock before we change inode->i_size.
1635 * If we don't we can race with other i_size updaters on our node, like
1636 * ll_file_read. We can also race with i_size propogation to other
1637 * nodes through dirtying and writeback of final cached pages. This
1638 * last one is especially bad for racing o_append users on other
1640 if (ia_valid & ATTR_SIZE) {
1641 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1642 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1643 struct obd_info *oinfo;
1645 struct lustre_handle lockh = { 0 };
1648 OBD_ALLOC_PTR(oinfo);
1652 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1653 inode->i_ino, LTIME_S(attr->ia_mtime));
1657 oa->o_id = lsm->lsm_object_id;
1658 oa->o_gr = lsm->lsm_object_gr;
1659 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1661 valid = OBD_MD_FLTYPE | OBD_MD_FLFID | OBD_MD_FLGENER;
1663 if (LTIME_S(attr->ia_mtime) < LTIME_S(attr->ia_ctime)){
1664 struct ost_lvb xtimes;
1666 /* setting mtime to past is performed under PW
1667 * EOF extent lock */
1668 oinfo->oi_policy.l_extent.start = 0;
1669 oinfo->oi_policy.l_extent.end = OBD_OBJECT_EOF;
1670 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW,
1674 OBD_FREE_PTR(oinfo);
1678 /* setattr under locks
1680 * 1. restore inode's timestamps which
1681 * are about to be set as long as
1682 * concurrent stat (via
1683 * ll_glimpse_size) might bring
1686 * 2. update lsm so that next stat
1687 * (via ll_glimpse_size) could get
1688 * correct values in lsm */
1689 lov_stripe_lock(lli->lli_smd);
1690 if (ia_valid & ATTR_ATIME) {
1691 LTIME_S(inode->i_atime) =
1693 LTIME_S(attr->ia_atime);
1694 valid |= OBD_MD_FLATIME;
1696 if (ia_valid & ATTR_MTIME) {
1697 LTIME_S(inode->i_mtime) =
1699 LTIME_S(attr->ia_mtime);
1700 valid |= OBD_MD_FLMTIME;
1702 if (ia_valid & ATTR_CTIME) {
1703 LTIME_S(inode->i_ctime) =
1705 LTIME_S(attr->ia_ctime);
1706 valid |= OBD_MD_FLCTIME;
1709 obd_update_lvb(ll_i2obdexp(inode), lli->lli_smd,
1711 lov_stripe_unlock(lli->lli_smd);
1715 * 1. do not use inode's timestamps
1716 * because concurrent stat might fill
1717 * the inode with out-of-date times,
1718 * send values from attr instead
1720 * 2.do no update lsm, as long as stat
1721 * (via ll_glimpse_size) will bring
1722 * attributes from osts anyway */
1723 if (ia_valid & ATTR_ATIME) {
1724 oa->o_atime = LTIME_S(attr->ia_atime);
1725 oa->o_valid |= OBD_MD_FLATIME;
1727 if (ia_valid & ATTR_MTIME) {
1728 oa->o_mtime = LTIME_S(attr->ia_mtime);
1729 oa->o_valid |= OBD_MD_FLMTIME;
1731 if (ia_valid & ATTR_CTIME) {
1732 oa->o_ctime = LTIME_S(attr->ia_ctime);
1733 oa->o_valid |= OBD_MD_FLCTIME;
1737 obdo_from_inode(oa, inode, valid);
1742 rc = obd_setattr_rqset(sbi->ll_osc_exp, oinfo, NULL);
1744 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1746 if (LTIME_S(attr->ia_mtime) < LTIME_S(attr->ia_ctime)){
1749 err = ll_extent_unlock(NULL, inode, lsm,
1751 if (unlikely(err != 0)) {
1752 CERROR("extent unlock failed: "
1762 OBD_FREE_PTR(oinfo);
1767 int ll_setattr(struct dentry *de, struct iattr *attr)
1769 int mode = de->d_inode->i_mode;
1771 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1772 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1773 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1774 if ((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1775 (ATTR_SIZE|ATTR_MODE)) {
1776 if (((mode & S_ISUID) && (!(attr->ia_mode & S_ISUID))) ||
1777 ((mode & S_ISGID) && (mode & S_IXGRP) &&
1778 (!(attr->ia_mode & S_ISGID))))
1779 attr->ia_valid |= ATTR_FORCE;
1782 if ((mode & S_ISUID) && S_ISREG(mode) &&
1783 !(attr->ia_mode & S_ISUID))
1784 attr->ia_valid |= ATTR_KILL_SUID;
1786 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1787 S_ISREG(mode) && !(attr->ia_mode & S_ISGID))
1788 attr->ia_valid |= ATTR_KILL_SGID;
1790 return ll_setattr_raw(de->d_inode, attr);
1793 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1794 __u64 max_age, __u32 flags)
1796 struct ll_sb_info *sbi = ll_s2sbi(sb);
1797 struct obd_statfs obd_osfs;
1801 rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age, flags);
1803 CERROR("mdc_statfs fails: rc = %d\n", rc);
1807 osfs->os_type = sb->s_magic;
1809 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1810 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1812 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1813 flags |= OBD_STATFS_NODELAY;
1815 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_osc_exp),
1816 &obd_osfs, max_age, flags);
1818 CERROR("obd_statfs fails: rc = %d\n", rc);
1822 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1823 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1826 osfs->os_bsize = obd_osfs.os_bsize;
1827 osfs->os_blocks = obd_osfs.os_blocks;
1828 osfs->os_bfree = obd_osfs.os_bfree;
1829 osfs->os_bavail = obd_osfs.os_bavail;
1831 /* If we don't have as many objects free on the OST as inodes
1832 * on the MDS, we reduce the total number of inodes to
1833 * compensate, so that the "inodes in use" number is correct.
1835 if (obd_osfs.os_ffree < osfs->os_ffree) {
1836 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1838 osfs->os_ffree = obd_osfs.os_ffree;
1843 #ifndef HAVE_STATFS_DENTRY_PARAM
1844 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1847 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1849 struct super_block *sb = de->d_sb;
1851 struct obd_statfs osfs;
1854 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1855 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1857 /* For now we will always get up-to-date statfs values, but in the
1858 * future we may allow some amount of caching on the client (e.g.
1859 * from QOS or lprocfs updates). */
1860 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1864 statfs_unpack(sfs, &osfs);
1866 /* We need to downshift for all 32-bit kernels, because we can't
1867 * tell if the kernel is being called via sys_statfs64() or not.
1868 * Stop before overflowing f_bsize - in which case it is better
1869 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1870 if (sizeof(long) < 8) {
1871 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1874 osfs.os_blocks >>= 1;
1875 osfs.os_bfree >>= 1;
1876 osfs.os_bavail >>= 1;
1880 sfs->f_blocks = osfs.os_blocks;
1881 sfs->f_bfree = osfs.os_bfree;
1882 sfs->f_bavail = osfs.os_bavail;
1887 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1889 struct ll_inode_info *lli;
1890 struct lov_stripe_md *lsm;
1892 lli = ll_i2info(inode);
1893 LASSERT(lli->lli_size_sem_owner != current);
1894 down(&lli->lli_size_sem);
1895 LASSERT(lli->lli_size_sem_owner == NULL);
1896 lli->lli_size_sem_owner = current;
1898 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1901 lov_stripe_lock(lsm);
1904 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1906 struct ll_inode_info *lli;
1907 struct lov_stripe_md *lsm;
1909 lli = ll_i2info(inode);
1911 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1914 lov_stripe_unlock(lsm);
1915 LASSERT(lli->lli_size_sem_owner == current);
1916 lli->lli_size_sem_owner = NULL;
1917 up(&lli->lli_size_sem);
1920 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1922 struct ll_inode_info *lli = ll_i2info(inode);
1924 dump_lsm(D_INODE, lsm);
1925 dump_lsm(D_INODE, lli->lli_smd);
1926 LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1927 "lsm must be joined lsm %p\n", lsm);
1928 obd_free_memmd(ll_i2obdexp(inode), &lli->lli_smd);
1929 CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1930 lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1932 lli->lli_maxbytes = lsm->lsm_maxbytes;
1933 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1934 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1937 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1939 struct ll_inode_info *lli = ll_i2info(inode);
1940 struct mds_body *body = md->body;
1941 struct lov_stripe_md *lsm = md->lsm;
1942 struct ll_sb_info *sbi = ll_i2sbi(inode);
1945 CDEBUG(D_INODE, "body->valid = "LPX64"\n", body->valid);
1947 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1949 if (lli->lli_smd == NULL) {
1950 if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1951 lsm->lsm_magic != LOV_MAGIC_V3 &&
1952 lsm->lsm_magic != LOV_MAGIC_JOIN) {
1953 dump_lsm(D_ERROR, lsm);
1956 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1957 lsm, inode->i_ino, inode->i_generation, inode);
1958 /* ll_inode_size_lock() requires it is only called
1959 * with lli_smd != NULL or lock_lsm == 0 or we can
1960 * race between lock/unlock. bug 9547 */
1962 lli->lli_maxbytes = lsm->lsm_maxbytes;
1963 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1964 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1966 if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1967 lli->lli_smd->lsm_stripe_count ==
1968 lsm->lsm_stripe_count) {
1969 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1970 CERROR("lsm mismatch for inode %ld\n",
1972 CERROR("lli_smd:\n");
1973 dump_lsm(D_ERROR, lli->lli_smd);
1975 dump_lsm(D_ERROR, lsm);
1979 ll_replace_lsm(inode, lsm);
1981 if (lli->lli_smd != lsm)
1982 obd_free_memmd(ll_i2obdexp(inode), &lsm);
1985 #ifdef CONFIG_FS_POSIX_ACL
1986 LASSERT(!md->posix_acl || (body->valid & OBD_MD_FLACL));
1987 if (body->valid & OBD_MD_FLACL) {
1988 spin_lock(&lli->lli_lock);
1989 if (lli->lli_posix_acl)
1990 posix_acl_release(lli->lli_posix_acl);
1991 lli->lli_posix_acl = md->posix_acl;
1992 spin_unlock(&lli->lli_lock);
1996 inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1997 inode->i_generation = ll_fid_build_gen(sbi, &body->fid1);
1998 *ll_inode_lu_fid(inode) = *((struct lu_fid*)&md->body->fid1);
2000 if (body->valid & OBD_MD_FLATIME) {
2001 if (body->atime > LTIME_S(inode->i_atime))
2002 LTIME_S(inode->i_atime) = body->atime;
2003 lli->lli_lvb.lvb_atime = body->atime;
2005 if (body->valid & OBD_MD_FLMTIME) {
2006 if (body->mtime > LTIME_S(inode->i_mtime)) {
2007 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
2008 "to "LPU64"\n", inode->i_ino,
2009 LTIME_S(inode->i_mtime), body->mtime);
2010 LTIME_S(inode->i_mtime) = body->mtime;
2012 lli->lli_lvb.lvb_mtime = body->mtime;
2014 if (body->valid & OBD_MD_FLCTIME) {
2015 if (body->ctime > LTIME_S(inode->i_ctime))
2016 LTIME_S(inode->i_ctime) = body->ctime;
2017 lli->lli_lvb.lvb_ctime = body->ctime;
2019 if (body->valid & OBD_MD_FLMODE)
2020 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
2021 if (body->valid & OBD_MD_FLTYPE)
2022 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
2023 if (S_ISREG(inode->i_mode)) {
2024 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS+1, LL_MAX_BLKSIZE_BITS);
2026 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2028 #ifdef HAVE_INODE_BLKSIZE
2029 inode->i_blksize = 1<<inode->i_blkbits;
2031 if (body->valid & OBD_MD_FLUID)
2032 inode->i_uid = body->uid;
2033 if (body->valid & OBD_MD_FLGID)
2034 inode->i_gid = body->gid;
2035 if (body->valid & OBD_MD_FLFLAGS)
2036 inode->i_flags = ll_ext_to_inode_flags(body->flags |
2037 MDS_BFLAG_EXT_FLAGS);
2038 if (body->valid & OBD_MD_FLNLINK)
2039 inode->i_nlink = body->nlink;
2041 if (body->valid & OBD_MD_FLRDEV)
2042 inode->i_rdev = old_decode_dev(body->rdev);
2043 if (body->valid & OBD_MD_FLSIZE) {
2044 #if 0 /* Can't block ll_test_inode->ll_update_inode, b=14326*/
2045 ll_inode_size_lock(inode, 0);
2046 i_size_write(inode, body->size);
2047 ll_inode_size_unlock(inode, 0);
2049 inode->i_size = body->size;
2052 if (body->valid & OBD_MD_FLBLOCKS)
2053 inode->i_blocks = body->blocks;
2055 if (body->valid & OBD_MD_FLSIZE)
2056 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
2060 static struct backing_dev_info ll_backing_dev_info = {
2061 .ra_pages = 0, /* No readahead */
2062 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
2063 .capabilities = 0, /* Does contribute to dirty memory */
2065 .memory_backed = 0, /* Does contribute to dirty memory */
2069 void ll_read_inode2(struct inode *inode, void *opaque)
2071 struct lustre_md *md = opaque;
2072 struct ll_inode_info *lli = ll_i2info(inode);
2075 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
2076 inode->i_generation, inode);
2080 LASSERT(!lli->lli_smd);
2082 /* Core attributes from the MDS first. This is a new inode, and
2083 * the VFS doesn't zero times in the core inode so we have to do
2084 * it ourselves. They will be overwritten by either MDS or OST
2085 * attributes - we just need to make sure they aren't newer. */
2086 LTIME_S(inode->i_mtime) = 0;
2087 LTIME_S(inode->i_atime) = 0;
2088 LTIME_S(inode->i_ctime) = 0;
2090 ll_update_inode(inode, md);
2092 /* OIDEBUG(inode); */
2094 if (S_ISREG(inode->i_mode)) {
2095 struct ll_sb_info *sbi = ll_i2sbi(inode);
2096 inode->i_op = &ll_file_inode_operations;
2097 inode->i_fop = sbi->ll_fop;
2098 inode->i_mapping->a_ops = &ll_aops;
2100 } else if (S_ISDIR(inode->i_mode)) {
2101 inode->i_op = &ll_dir_inode_operations;
2102 inode->i_fop = &ll_dir_operations;
2103 inode->i_mapping->a_ops = &ll_dir_aops;
2105 } else if (S_ISLNK(inode->i_mode)) {
2106 inode->i_op = &ll_fast_symlink_inode_operations;
2109 inode->i_op = &ll_special_inode_operations;
2110 init_special_inode(inode, inode->i_mode,
2111 kdev_t_to_nr(inode->i_rdev));
2112 /* initializing backing dev info. */
2113 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2118 int ll_iocontrol(struct inode *inode, struct file *file,
2119 unsigned int cmd, unsigned long arg)
2121 struct ll_sb_info *sbi = ll_i2sbi(inode);
2122 struct ptlrpc_request *req = NULL;
2127 case FSFILT_IOC_GETFLAGS: {
2129 struct mds_body *body;
2131 ll_inode2fid(&fid, inode);
2132 rc = mdc_getattr(sbi->ll_mdc_exp, &fid, OBD_MD_FLFLAGS,0,&req);
2134 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2138 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
2141 /* We want to return EXT3_*_FL flags to the caller via this
2142 * ioctl. An older MDS may be sending S_* flags, fix it up. */
2143 flags = ll_inode_to_ext_flags(body->flags,
2144 MDS_BFLAG_EXT_FLAGS);
2145 ptlrpc_req_finished (req);
2147 RETURN(put_user(flags, (int *)arg));
2149 case FSFILT_IOC_SETFLAGS: {
2150 struct mdc_op_data op_data = { { 0 } };
2151 struct ll_iattr_struct attr;
2152 struct obd_info oinfo = { { { 0 } } };
2153 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2155 if (get_user(flags, (int *)arg))
2159 OBDO_ALLOC(oinfo.oi_oa);
2163 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0, NULL);
2165 memset(&attr, 0, sizeof(attr));
2166 attr.ia_attr_flags = flags;
2167 ((struct iattr *)&attr)->ia_valid |= ATTR_ATTR_FLAG;
2169 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
2170 (struct iattr *)&attr, NULL, 0, NULL, 0, &req);
2171 ptlrpc_req_finished(req);
2172 if (rc || lsm == NULL) {
2173 OBDO_FREE(oinfo.oi_oa);
2177 oinfo.oi_oa->o_id = lsm->lsm_object_id;
2178 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
2179 oinfo.oi_oa->o_flags = flags;
2180 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP |
2183 obdo_from_inode(oinfo.oi_oa, inode,
2184 OBD_MD_FLFID | OBD_MD_FLGENER);
2185 rc = obd_setattr_rqset(sbi->ll_osc_exp, &oinfo, NULL);
2186 OBDO_FREE(oinfo.oi_oa);
2188 if (rc != -EPERM && rc != -EACCES)
2189 CERROR("mdc_setattr_async fails: rc = %d\n", rc);
2193 inode->i_flags = ll_ext_to_inode_flags(flags |
2194 MDS_BFLAG_EXT_FLAGS);
2204 /* umount -f client means force down, don't save state */
2205 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2206 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2208 struct super_block *sb = vfsmnt->mnt_sb;
2210 void ll_umount_begin(struct super_block *sb)
2213 struct lustre_sb_info *lsi = s2lsi(sb);
2214 struct ll_sb_info *sbi = ll_s2sbi(sb);
2215 struct obd_device *obd;
2216 struct obd_ioctl_data ioc_data = { 0 };
2219 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2220 if (!(flags & MNT_FORCE)) {
2226 /* Tell the MGC we got umount -f */
2227 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2229 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2230 sb->s_count, atomic_read(&sb->s_active));
2232 obd = class_exp2obd(sbi->ll_mdc_exp);
2234 CERROR("Invalid MDC connection handle "LPX64"\n",
2235 sbi->ll_mdc_exp->exp_handle.h_cookie);
2240 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
2243 obd = class_exp2obd(sbi->ll_osc_exp);
2245 CERROR("Invalid LOV connection handle "LPX64"\n",
2246 sbi->ll_osc_exp->exp_handle.h_cookie);
2252 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
2255 /* Really, we'd like to wait until there are no requests outstanding,
2256 * and then continue. For now, we just invalidate the requests,
2257 * schedule() and sleep one second if needed, and hope.
2260 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2261 if (atomic_read(&vfsmnt->mnt_count) > 2) {
2262 cfs_schedule_timeout(CFS_TASK_INTERRUPTIBLE,
2263 cfs_time_seconds(1));
2264 if (atomic_read(&vfsmnt->mnt_count) > 2)
2265 LCONSOLE_WARN("Mount still busy with %d refs! You "
2266 "may try to umount it a bit later\n",
2267 atomic_read(&vfsmnt->mnt_count));
2274 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2276 struct ll_sb_info *sbi = ll_s2sbi(sb);
2280 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2281 read_only = *flags & MS_RDONLY;
2282 err = obd_set_info_async(sbi->ll_mdc_exp, sizeof(KEY_READONLY),
2283 KEY_READONLY, sizeof(read_only),
2286 /* MDS might have expected a different ro key value, b=17493 */
2287 if (err == -EINVAL) {
2288 CDEBUG(D_CONFIG, "Retrying remount with 1.6.6 ro key\n");
2289 err = obd_set_info_async(sbi->ll_mdc_exp,
2290 sizeof(KEY_READONLY_166COMPAT),
2291 KEY_READONLY_166COMPAT,
2297 CERROR("Failed to change the read-only flag during "
2298 "remount: %d\n", err);
2303 sb->s_flags |= MS_RDONLY;
2305 sb->s_flags &= ~MS_RDONLY;
2310 int ll_prep_inode(struct obd_export *exp, struct inode **inode,
2311 struct ptlrpc_request *req, int offset,struct super_block *sb)
2313 struct lustre_md md;
2314 struct ll_sb_info *sbi = NULL;
2318 LASSERT(*inode || sb);
2319 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2320 prune_deathrow(sbi, 1);
2322 rc = mdc_req2lustre_md(req, offset, exp, &md);
2327 ll_update_inode(*inode, &md);
2330 /** hashing VFS inode by FIDs.
2331 * IGIF will be used for for compatibility if needed.
2333 *inode =ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2334 if (*inode == NULL || is_bad_inode(*inode)) {
2335 mdc_free_lustre_md(exp, &md);
2337 CERROR("new_inode -fatal: rc %d\n", rc);
2342 rc = obd_checkmd(exp, ll_i2mdcexp(*inode),
2343 ll_i2info(*inode)->lli_smd);
2348 char *llap_origins[] = {
2349 [LLAP_ORIGIN_UNKNOWN] = "--",
2350 [LLAP_ORIGIN_READPAGE] = "rp",
2351 [LLAP_ORIGIN_READAHEAD] = "ra",
2352 [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2353 [LLAP_ORIGIN_WRITEPAGE] = "wp",
2354 [LLAP_ORIGIN_LOCKLESS_IO] = "ls"
2357 struct ll_async_page *llite_pglist_next_llap(struct list_head *head,
2358 struct list_head *list)
2360 struct ll_async_page *llap;
2361 struct list_head *pos;
2363 list_for_each(pos, list) {
2366 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2367 if (llap->llap_page == NULL)
2375 int ll_obd_statfs(struct inode *inode, void *arg)
2377 struct ll_sb_info *sbi = NULL;
2378 struct obd_device *client_obd = NULL, *lov_obd = NULL;
2379 struct lov_obd *lov = NULL;
2380 struct obd_statfs stat_buf = {0};
2382 struct obd_ioctl_data *data = NULL;
2386 if (!inode || !(sbi = ll_i2sbi(inode)))
2387 GOTO(out_statfs, rc = -EINVAL);
2389 rc = obd_ioctl_getdata(&buf, &len, arg);
2391 GOTO(out_statfs, rc);
2394 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2395 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2396 GOTO(out_statfs, rc = -EINVAL);
2398 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2399 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2401 if (type == LL_STATFS_MDC) {
2403 GOTO(out_statfs, rc = -ENODEV);
2404 client_obd = class_exp2obd(sbi->ll_mdc_exp);
2405 } else if (type == LL_STATFS_LOV) {
2406 lov_obd = class_exp2obd(sbi->ll_osc_exp);
2407 lov = &lov_obd->u.lov;
2409 if (index >= lov->desc.ld_tgt_count)
2410 GOTO(out_statfs, rc = -ENODEV);
2412 if (!lov->lov_tgts[index])
2413 /* Try again with the next index */
2414 GOTO(out_statfs, rc = -EAGAIN);
2416 client_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2417 if (!lov->lov_tgts[index]->ltd_active)
2418 GOTO(out_uuid, rc = -ENODATA);
2422 GOTO(out_statfs, rc = -EINVAL);
2424 rc = obd_statfs(client_obd, &stat_buf, cfs_time_current_64() - HZ, 1);
2426 GOTO(out_statfs, rc);
2428 if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2429 GOTO(out_statfs, rc = -EFAULT);
2433 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(client_obd),
2437 if (copy_to_user(data->ioc_pbuf2, "Unknown UUID",
2438 sizeof("Unknown UUID") + 1))
2444 obd_ioctl_freedata(buf, len);
2448 int ll_process_config(struct lustre_cfg *lcfg)
2452 struct lprocfs_static_vars lvars;
2456 lprocfs_llite_init_vars(&lvars);
2458 /* The instance name contains the sb: lustre-client-aacfe000 */
2459 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2460 if (!ptr || !*(++ptr))
2462 if (sscanf(ptr, "%lx", &x) != 1)
2465 /* This better be a real Lustre superblock! */
2466 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2468 /* Note we have not called client_common_fill_super yet, so
2469 proc fns must be able to handle that! */
2470 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2475 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2477 struct ll_sb_info *sbi;
2479 LASSERT((seq != NULL) && (vfs != NULL));
2480 sbi = ll_s2sbi(vfs->mnt_sb);
2482 if (sbi->ll_flags & LL_SBI_NOLCK)
2483 seq_puts(seq, ",nolock");
2485 if (sbi->ll_flags & LL_SBI_FLOCK)
2486 seq_puts(seq, ",flock");
2488 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2489 seq_puts(seq, ",localflock");
2491 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2492 seq_puts(seq, ",user_xattr");
2494 if (sbi->ll_flags & LL_SBI_ACL)
2495 seq_puts(seq, ",acl");
2497 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2498 seq_puts(seq, ",lazystatfs");