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 struct rw_semaphore ll_sb_sem;
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 sbi->ll_direct_io_default = SBI_DEFAULT_DIRECT_IO_DEFAULT;
180 INIT_LIST_HEAD(&sbi->ll_conn_chain);
181 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
183 ll_generate_random_uuid(uuid);
184 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
185 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
187 down_write(&ll_sb_sem);
188 list_add_tail(&sbi->ll_list, &ll_super_blocks);
189 up_write(&ll_sb_sem);
191 #ifdef ENABLE_CHECKSUM
192 sbi->ll_flags |= LL_SBI_DATA_CHECKSUM;
194 #ifdef ENABLE_LLITE_CHECKSUM
195 sbi->ll_flags |= LL_SBI_LLITE_CHECKSUM;
198 #ifdef HAVE_LRU_RESIZE_SUPPORT
199 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
202 #ifdef HAVE_EXPORT___IGET
203 INIT_LIST_HEAD(&sbi->ll_deathrow);
204 spin_lock_init(&sbi->ll_deathrow_lock);
206 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
207 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
208 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
211 /* metadata statahead is enabled by default */
212 sbi->ll_sa_max = LL_SA_RPC_DEF;
217 if (sbi->ll_async_page_sample)
218 OBD_FREE(sbi->ll_async_page_sample,
219 sizeof(long) * num_possible_cpus());
221 OBD_FREE(sbi, sizeof(*sbi));
225 void ll_free_sbi(struct super_block *sb)
227 struct ll_sb_info *sbi = ll_s2sbi(sb);
231 down_write(&ll_sb_sem);
232 list_del(&sbi->ll_list);
233 up_write(&ll_sb_sem);
234 /* dont allow find cache via sb list first */
236 lcounter_destroy(&sbi->ll_async_page_count);
237 OBD_FREE(sbi->ll_async_page_sample,
238 sizeof(long) * num_possible_cpus());
239 OBD_FREE(sbi, sizeof(*sbi));
244 static struct dentry_operations ll_d_root_ops = {
245 .d_compare = ll_dcompare,
246 .d_revalidate = ll_revalidate_nd,
249 static int client_common_fill_super(struct super_block *sb,
250 char *mdc, char *osc)
252 struct inode *root = 0;
253 struct ll_sb_info *sbi = ll_s2sbi(sb);
254 struct obd_device *obd;
255 struct ll_fid rootfid;
256 struct obd_statfs osfs;
257 struct ptlrpc_request *request = NULL;
258 struct lustre_handle osc_conn = {0, };
259 struct lustre_handle mdc_conn = {0, };
261 struct obd_connect_data *data = NULL;
265 obd = class_name2obd(mdc);
267 CERROR("MDC %s: not setup or attached\n", mdc);
271 OBD_ALLOC(data, sizeof(*data));
275 if (proc_lustre_fs_root) {
276 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
279 CERROR("could not register mount in /proc/fs/lustre\n");
282 /* indicate the features supported by this client */
283 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_IBITS |
284 OBD_CONNECT_JOIN | OBD_CONNECT_ATTRFID |
285 OBD_CONNECT_NODEVOH | OBD_CONNECT_CANCELSET |
286 OBD_CONNECT_AT | OBD_CONNECT_FID |
287 OBD_CONNECT_VBR | OBD_CONNECT_LOV_V3;
288 #ifdef HAVE_LRU_RESIZE_SUPPORT
289 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
290 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
292 #ifdef CONFIG_FS_POSIX_ACL
293 data->ocd_connect_flags |= OBD_CONNECT_ACL;
295 data->ocd_ibits_known = MDS_INODELOCK_FULL;
296 data->ocd_version = LUSTRE_VERSION_CODE;
298 if (sb->s_flags & MS_RDONLY)
299 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
300 if (sbi->ll_flags & LL_SBI_USER_XATTR)
301 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
303 #ifdef HAVE_MS_FLOCK_LOCK
304 /* force vfs to use lustre handler for flock() calls - bug 10743 */
305 sb->s_flags |= MS_FLOCK_LOCK;
308 if (sbi->ll_flags & LL_SBI_FLOCK)
309 sbi->ll_fop = &ll_file_operations_flock;
310 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
311 sbi->ll_fop = &ll_file_operations;
313 sbi->ll_fop = &ll_file_operations_noflock;
316 err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid, data, &sbi->ll_mdc_exp);
318 LCONSOLE_ERROR_MSG(0x14f, "An MDT (mdc %s) is performing "
319 "recovery, of which this client is not a "
320 "part. Please wait for recovery to complete,"
321 " abort, or time out.\n", mdc);
324 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
328 err = obd_fid_init(sbi->ll_mdc_exp);
332 err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
336 /* MDC connect is surely finished by now because we actually sent
337 * a statfs RPC, otherwise obd_connect() is asynchronous. */
338 *data = class_exp2cliimp(sbi->ll_mdc_exp)->imp_connect_data;
340 LASSERT(osfs.os_bsize);
341 sb->s_blocksize = osfs.os_bsize;
342 sb->s_blocksize_bits = log2(osfs.os_bsize);
343 sb->s_magic = LL_SUPER_MAGIC;
345 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
346 * retval = in_file->f_op->sendfile(...);
348 * retval = -EOVERFLOW;
350 * it will check if *ppos is greater than max. However, max equals to
351 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
352 * has been defined as a signed long long ineger in linux kernel. */
353 #if BITS_PER_LONG == 64
354 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
356 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
358 sbi->ll_namelen = osfs.os_namelen;
359 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
361 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
362 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
363 LCONSOLE_INFO("Disabling user_xattr feature because "
364 "it is not supported on the server\n");
365 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
368 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
370 sb->s_flags |= MS_POSIXACL;
372 sbi->ll_flags |= LL_SBI_ACL;
374 sbi->ll_flags &= ~LL_SBI_ACL;
376 if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
377 sbi->ll_flags |= LL_SBI_JOIN;
379 obd = class_name2obd(osc);
381 CERROR("OSC %s: not setup or attached\n", osc);
382 GOTO(out_mdc, err = -ENODEV);
385 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_GRANT |
386 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
387 OBD_CONNECT_SRVLOCK | OBD_CONNECT_CANCELSET|
388 OBD_CONNECT_AT | OBD_CONNECT_FID |
389 OBD_CONNECT_VBR | OBD_CONNECT_TRUNCLOCK;
391 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
392 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
393 * disabled by default, because it can still be enabled on the
394 * fly via /proc. As a consequence, we still need to come to an
395 * agreement on the supported algorithms at connect time */
396 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
398 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
399 data->ocd_cksum_types = OBD_CKSUM_ADLER;
401 /* send the list of supported checksum types */
402 data->ocd_cksum_types = OBD_CKSUM_ALL;
405 #ifdef HAVE_LRU_RESIZE_SUPPORT
406 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
407 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
410 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
411 "ocd_grant: %d\n", data->ocd_connect_flags,
412 data->ocd_version, data->ocd_grant);
414 obd->obd_upcall.onu_owner = &sbi->ll_lco;
415 obd->obd_upcall.onu_upcall = ll_ocd_update;
416 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
418 /* ask lov to generate OBD_NOTIFY_CREATE events for already registered
420 obd_notify(obd, NULL, OBD_NOTIFY_CREATE, NULL);
422 obd_register_lock_cancel_cb(obd, ll_extent_lock_cancel_cb);
423 obd_register_page_removal_cb(obd, ll_page_removal_cb, ll_pin_extent_cb);
425 err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid, data, &sbi->ll_osc_exp);
427 LCONSOLE_ERROR_MSG(0x150, "An OST (osc %s) is performing "
428 "recovery, of which this client is not a "
429 "part. Please wait for recovery to "
430 "complete, abort, or time out.\n", osc);
433 CERROR("cannot connect to %s: rc = %d\n", osc, err);
437 mutex_down(&sbi->ll_lco.lco_lock);
438 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
439 sbi->ll_lco.lco_mdc_exp = sbi->ll_mdc_exp;
440 sbi->ll_lco.lco_osc_exp = sbi->ll_osc_exp;
441 mutex_up(&sbi->ll_lco.lco_lock);
443 err = mdc_init_ea_size(sbi->ll_mdc_exp, sbi->ll_osc_exp);
445 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
449 err = obd_prep_async_page(sbi->ll_osc_exp, NULL, NULL, NULL,
450 0, NULL, NULL, NULL, 0, NULL);
452 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
453 "filesystem. There must be at least one "
454 "active OST for a client to start.\n");
458 if (!ll_async_page_slab) {
459 ll_async_page_slab_size =
460 size_round(sizeof(struct ll_async_page)) + err;
461 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
462 ll_async_page_slab_size,
464 if (!ll_async_page_slab)
465 GOTO(out_osc, err = -ENOMEM);
468 err = mdc_getstatus(sbi->ll_mdc_exp, &rootfid);
470 CERROR("cannot mds_connect: rc = %d\n", err);
473 CDEBUG(D_SUPER, "rootfid "LPU64":"DFID"\n", rootfid.id,
474 PFID((struct lu_fid*)&rootfid));
475 sbi->ll_rootino = rootfid.id;
477 sb->s_op = &lustre_super_operations;
478 #if THREAD_SIZE >= 8192
479 /* Disable the NFS export because of stack overflow
480 * when THREAD_SIZE < 8192. Please refer to 17630. */
481 sb->s_export_op = &lustre_export_operations;
485 * XXX: move this to after cbd setup? */
486 err = mdc_getattr(sbi->ll_mdc_exp, &rootfid,
487 OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS |
488 (sbi->ll_flags & LL_SBI_ACL ? OBD_MD_FLACL : 0),
491 CERROR("mdc_getattr failed for root: rc = %d\n", err);
495 err = mdc_req2lustre_md(request, REPLY_REC_OFF, sbi->ll_osc_exp, &md);
497 CERROR("failed to understand root inode md: rc = %d\n",err);
498 ptlrpc_req_finished (request);
502 LASSERT(sbi->ll_rootino != 0);
503 root = ll_iget(sb, ll_fid_build_ino(sbi, &rootfid), &md);
505 ptlrpc_req_finished(request);
507 if (root == NULL || is_bad_inode(root)) {
508 mdc_free_lustre_md(sbi->ll_osc_exp, &md);
509 CERROR("lustre_lite: bad iget4 for root\n");
510 GOTO(out_root, err = -EBADF);
513 err = ll_close_thread_start(&sbi->ll_lcq);
515 CERROR("cannot start close thread: rc %d\n", err);
519 checksum = sbi->ll_flags & LL_SBI_DATA_CHECKSUM;
520 err = obd_set_info_async(sbi->ll_osc_exp, sizeof(KEY_CHECKSUM),
521 KEY_CHECKSUM, sizeof(checksum),
524 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
525 backing dev info assigned to inode mapping is used for
526 determining maximal readahead. */
527 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
528 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
529 /* bug 2805 - set VM readahead to zero */
530 vm_max_readahead = vm_min_readahead = 0;
533 sb->s_root = d_alloc_root(root);
535 OBD_FREE(data, sizeof(*data));
536 sb->s_root->d_op = &ll_d_root_ops;
538 sbi->ll_sdev_orig = sb->s_dev;
539 /* We set sb->s_dev equal on all lustre clients in order to support
540 * NFS export clustering. NFSD requires that the FSID be the same
542 /* s_dev is also used in lt_compare() to compare two fs, but that is
543 * only a node-local comparison. */
544 sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
545 strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
553 obd_disconnect(sbi->ll_osc_exp);
554 sbi->ll_osc_exp = NULL;
556 obd = class_name2obd(osc);
557 obd_unregister_lock_cancel_cb(obd, ll_extent_lock_cancel_cb);
558 obd_unregister_page_removal_cb(obd, ll_page_removal_cb);
560 obd_fid_fini(sbi->ll_mdc_exp);
561 obd_disconnect(sbi->ll_mdc_exp);
562 sbi->ll_mdc_exp = NULL;
565 OBD_FREE(data, sizeof(*data));
566 lprocfs_unregister_mountpoint(sbi);
570 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
574 *lmmsize = obd_size_diskmd(sbi->ll_osc_exp, NULL);
576 rc = obd_get_info(sbi->ll_mdc_exp, sizeof(KEY_MAX_EASIZE),
577 KEY_MAX_EASIZE, &size, lmmsize, NULL);
579 CERROR("Get max mdsize error rc %d \n", rc);
584 void ll_dump_inode(struct inode *inode)
586 struct list_head *tmp;
587 int dentry_count = 0;
589 LASSERT(inode != NULL);
591 list_for_each(tmp, &inode->i_dentry)
594 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
595 inode, ll_i2mdcexp(inode)->exp_obd->obd_name, inode->i_ino,
596 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
599 void lustre_dump_dentry(struct dentry *dentry, int recur)
601 struct list_head *tmp;
604 LASSERT(dentry != NULL);
606 list_for_each(tmp, &dentry->d_subdirs)
609 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
610 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
611 dentry->d_name.len, dentry->d_name.name,
612 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
613 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
614 dentry->d_flags, dentry->d_fsdata, subdirs);
615 if (dentry->d_inode != NULL)
616 ll_dump_inode(dentry->d_inode);
621 list_for_each(tmp, &dentry->d_subdirs) {
622 struct dentry *d = list_entry(tmp, struct dentry, d_child);
623 lustre_dump_dentry(d, recur - 1);
627 #ifdef HAVE_EXPORT___IGET
628 static void prune_dir_dentries(struct inode *inode)
630 struct dentry *dentry, *prev = NULL;
632 /* due to lustre specific logic, a directory
633 * can have few dentries - a bug from VFS POV */
635 spin_lock(&dcache_lock);
636 if (!list_empty(&inode->i_dentry)) {
637 dentry = list_entry(inode->i_dentry.prev,
638 struct dentry, d_alias);
639 /* in order to prevent infinite loops we
640 * break if previous dentry is busy */
641 if (dentry != prev) {
644 spin_unlock(&dcache_lock);
646 /* try to kill all child dentries */
647 shrink_dcache_parent(dentry);
650 /* now try to get rid of current dentry */
651 d_prune_aliases(inode);
655 spin_unlock(&dcache_lock);
658 static void prune_deathrow_one(struct ll_inode_info *lli)
660 struct inode *inode = ll_info2i(lli);
662 /* first, try to drop any dentries - they hold a ref on the inode */
663 if (S_ISDIR(inode->i_mode))
664 prune_dir_dentries(inode);
666 d_prune_aliases(inode);
669 /* if somebody still uses it, leave it */
670 LASSERT(atomic_read(&inode->i_count) > 0);
671 if (atomic_read(&inode->i_count) > 1)
674 CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
675 inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
677 /* seems nobody uses it anymore */
685 static void prune_deathrow(struct ll_sb_info *sbi, int try)
687 struct ll_inode_info *lli;
691 if (need_resched() && try)
695 if (!spin_trylock(&sbi->ll_deathrow_lock))
698 spin_lock(&sbi->ll_deathrow_lock);
703 if (!list_empty(&sbi->ll_deathrow)) {
704 lli = list_entry(sbi->ll_deathrow.next,
705 struct ll_inode_info,
707 list_del_init(&lli->lli_dead_list);
708 if (!list_empty(&sbi->ll_deathrow))
711 spin_unlock(&sbi->ll_deathrow_lock);
714 prune_deathrow_one(lli);
716 } while (empty == 0);
718 #else /* !HAVE_EXPORT___IGET */
719 #define prune_deathrow(sbi, try) do {} while (0)
720 #endif /* HAVE_EXPORT___IGET */
722 void client_common_put_super(struct super_block *sb)
724 struct ll_sb_info *sbi = ll_s2sbi(sb);
727 ll_close_thread_shutdown(sbi->ll_lcq);
729 lprocfs_unregister_mountpoint(sbi);
731 /* destroy inodes in deathrow */
732 prune_deathrow(sbi, 0);
734 list_del(&sbi->ll_conn_chain);
736 /* callbacks is cleared after disconnect each target */
737 obd_disconnect(sbi->ll_osc_exp);
738 sbi->ll_osc_exp = NULL;
740 obd_fid_fini(sbi->ll_mdc_exp);
741 obd_disconnect(sbi->ll_mdc_exp);
742 sbi->ll_mdc_exp = NULL;
747 void ll_kill_super(struct super_block *sb)
749 struct ll_sb_info *sbi;
754 if (!(sb->s_flags & MS_ACTIVE))
758 /* we need restore s_dev from changed for clustred NFS before put_super
759 * because new kernels have cached s_dev and change sb->s_dev in
760 * put_super not affected real removing devices */
762 sb->s_dev = sbi->ll_sdev_orig;
766 char *ll_read_opt(const char *opt, char *data)
772 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
773 if (strncmp(opt, data, strlen(opt)))
775 if ((value = strchr(data, '=')) == NULL)
779 OBD_ALLOC(retval, strlen(value) + 1);
781 CERROR("out of memory!\n");
785 memcpy(retval, value, strlen(value)+1);
786 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
790 static inline int ll_set_opt(const char *opt, char *data, int fl)
792 if (strncmp(opt, data, strlen(opt)) != 0)
798 /* non-client-specific mount options are parsed in lmd_parse */
799 static int ll_options(char *options, int *flags)
802 char *s1 = options, *s2;
808 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
811 CDEBUG(D_SUPER, "next opt=%s\n", s1);
812 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
817 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
822 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
827 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
832 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
837 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
842 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
844 /* Ignore deprecated mount option. The client will
845 * always try to mount with ACL support, whether this
846 * is used depends on whether server supports it. */
847 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
848 "mount option 'acl'.\n");
851 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
853 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
854 "mount option 'noacl'.\n");
858 tmp = ll_set_opt("checksum", s1, LL_SBI_DATA_CHECKSUM);
863 tmp = ll_set_opt("nochecksum", s1, LL_SBI_DATA_CHECKSUM);
869 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
874 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
879 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
884 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
889 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
895 s2 = strchr(s1, ',');
903 void ll_lli_init(struct ll_inode_info *lli)
905 lli->lli_inode_magic = LLI_INODE_MAGIC;
906 sema_init(&lli->lli_size_sem, 1);
907 sema_init(&lli->lli_write_sem, 1);
909 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
910 spin_lock_init(&lli->lli_lock);
911 sema_init(&lli->lli_och_sem, 1);
912 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
913 lli->lli_mds_exec_och = NULL;
914 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
915 lli->lli_open_fd_exec_count = 0;
916 INIT_LIST_HEAD(&lli->lli_dead_list);
917 #ifdef HAVE_CLOSE_THREAD
918 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
920 init_rwsem(&lli->lli_truncate_rwsem);
924 #define MDCDEV "mdc_dev"
925 static int old_lustre_process_log(struct super_block *sb, char *newprofile,
926 struct config_llog_instance *cfg)
928 struct lustre_sb_info *lsi = s2lsi(sb);
929 struct obd_device *obd;
930 struct lustre_handle mdc_conn = {0, };
931 struct obd_export *exp;
932 char *ptr, *mdt, *profile;
933 char niduuid[10] = "mdtnid0";
935 struct obd_uuid mdc_uuid;
936 struct llog_ctxt *ctxt;
937 struct obd_connect_data ocd = { 0 };
939 int i, rc = 0, recov_bk = 1, failnodes = 0;
942 ll_generate_random_uuid(uuid);
943 class_uuid_unparse(uuid, &mdc_uuid);
944 CDEBUG(D_HA, "generated uuid: %s\n", mdc_uuid.uuid);
946 /* Figure out the old mdt and profile name from new-style profile
947 ("lustre" from "mds/lustre-client") */
949 profile = strchr(mdt, '/');
950 if (profile == NULL) {
951 CDEBUG(D_CONFIG, "Can't find MDT name in %s\n", newprofile);
952 GOTO(out, rc = -EINVAL);
956 ptr = strrchr(profile, '-');
958 CDEBUG(D_CONFIG, "Can't find client name in %s\n", newprofile);
959 GOTO(out, rc = -EINVAL);
963 LCONSOLE_WARN("This looks like an old mount command; I will try to "
964 "contact MDT '%s' for profile '%s'\n", mdt, profile);
966 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
968 ptr = lsi->lsi_lmd->lmd_dev;
969 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
970 rc = do_lcfg(MDCDEV, nid, LCFG_ADD_UUID, niduuid, 0,0,0);
972 /* Stop at the first failover nid */
977 CERROR("No valid MDT nids found.\n");
978 GOTO(out, rc = -EINVAL);
982 rc = do_lcfg(MDCDEV, 0, LCFG_ATTACH, LUSTRE_MDC_NAME,mdc_uuid.uuid,0,0);
984 GOTO(out_del_uuid, rc);
986 rc = do_lcfg(MDCDEV, 0, LCFG_SETUP, mdt, niduuid, 0, 0);
988 LCONSOLE_ERROR_MSG(0x153, "I couldn't establish a connection "
989 "with the MDT. Check that the MDT host NID "
990 "is correct and the networks are up.\n");
991 GOTO(out_detach, rc);
994 obd = class_name2obd(MDCDEV);
996 GOTO(out_cleanup, rc = -EINVAL);
998 /* Add any failover nids */
999 while (*ptr == ':') {
1000 /* New failover node */
1001 sprintf(niduuid, "mdtnid%d", failnodes);
1003 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
1005 rc = do_lcfg(MDCDEV, nid, LCFG_ADD_UUID, niduuid,0,0,0);
1007 CERROR("Add uuid for %s failed %d\n",
1008 libcfs_nid2str(nid), rc);
1013 rc = do_lcfg(MDCDEV, 0, LCFG_ADD_CONN, niduuid, 0, 0,0);
1015 CERROR("Add conn for %s failed %d\n",
1016 libcfs_nid2str(nid), rc);
1024 /* Try all connections, but only once. */
1025 rc = obd_set_info_async(obd->obd_self_export,
1026 sizeof(KEY_INIT_RECOV_BACKUP), KEY_INIT_RECOV_BACKUP,
1027 sizeof(recov_bk), &recov_bk, NULL);
1029 GOTO(out_cleanup, rc);
1031 /* If we don't have this then an ACL MDS will refuse the connection */
1032 ocd.ocd_connect_flags = OBD_CONNECT_ACL;
1034 rc = obd_connect(&mdc_conn, obd, &mdc_uuid, &ocd, &exp);
1036 CERROR("cannot connect to %s: rc = %d\n", mdt, rc);
1037 GOTO(out_cleanup, rc);
1040 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1042 cfg->cfg_flags |= CFG_F_COMPAT146;
1045 rc = class_config_parse_llog(ctxt, profile, cfg);
1048 * For debugging, it's useful to just dump the log
1050 rc = class_config_dump_llog(ctxt, profile, cfg);
1052 llog_ctxt_put(ctxt);
1055 /* Set the caller's profile name to the old-style */
1056 memcpy(newprofile, profile, strlen(profile) + 1);
1060 LCONSOLE_ERROR_MSG(0x154, "%s: The configuration '%s' could not"
1061 " be read from the MDT '%s'. Make sure this"
1062 " client and the MDT are running compatible "
1063 "versions of Lustre.\n",
1064 obd->obd_name, profile, mdt);
1067 LCONSOLE_ERROR_MSG(0x155, "%s: The configuration '%s' could not"
1068 " be read from the MDT '%s'. This may be "
1069 "the result of communication errors between "
1070 "the client and the MDT, or if the MDT is "
1071 "not running.\n", obd->obd_name, profile,
1076 /* We don't so much care about errors in cleaning up the config llog
1077 * connection, as we have already read the config by this point. */
1078 obd_disconnect(exp);
1081 do_lcfg(MDCDEV, 0, LCFG_CLEANUP, 0, 0, 0, 0);
1084 do_lcfg(MDCDEV, 0, LCFG_DETACH, 0, 0, 0, 0);
1087 /* class_add_uuid adds a nid even if the same uuid exists; we might
1088 delete any copy here. So they all better match. */
1089 for (i = 0; i < failnodes; i++) {
1090 sprintf(niduuid, "mdtnid%d", i);
1091 do_lcfg(MDCDEV, 0, LCFG_DEL_UUID, niduuid, 0, 0, 0);
1093 /* class_import_put will get rid of the additional connections */
1097 /* end COMPAT_146 */
1099 int ll_fill_super(struct super_block *sb)
1101 struct lustre_profile *lprof = NULL;
1102 struct lustre_sb_info *lsi = s2lsi(sb);
1103 struct ll_sb_info *sbi;
1104 char *osc = NULL, *mdc = NULL;
1105 char *profilenm = get_profile_name(sb);
1106 struct config_llog_instance cfg = {0, };
1107 char ll_instance[sizeof(sb) * 2 + 1];
1110 char pseudo[32] = { 0 };
1113 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1117 /* client additional sb info */
1118 lsi->lsi_llsbi = sbi = ll_init_sbi();
1124 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1126 GOTO(out_free, err);
1128 /* Generate a string unique to this super, in case some joker tries
1129 to mount the same fs at two mount points.
1130 Use the address of the super itself.*/
1131 sprintf(ll_instance, "%p", sb);
1132 cfg.cfg_instance = ll_instance;
1133 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1136 /* set up client obds */
1137 if (strchr(profilenm, '/') != NULL) /* COMPAT_146 */
1138 err = -EINVAL; /* skip error messages, use old config code */
1140 err = lustre_process_log(sb, profilenm, &cfg);
1145 oldnamelen = strlen(profilenm) + 1;
1146 /* Temp storage for 1.4.6 profile name */
1147 OBD_ALLOC(oldname, oldnamelen);
1149 memcpy(oldname, profilenm, oldnamelen);
1150 rc = old_lustre_process_log(sb, oldname, &cfg);
1152 /* That worked - update the profile name
1155 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1156 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1157 OBD_ALLOC(lsi->lsi_lmd->lmd_profile,
1158 strlen(oldname) + 1);
1159 if (!lsi->lsi_lmd->lmd_profile) {
1160 OBD_FREE(oldname, oldnamelen);
1161 GOTO(out_free, err = -ENOMEM);
1163 memcpy(lsi->lsi_lmd->lmd_profile, oldname,
1164 strlen(oldname) + 1);
1165 profilenm = get_profile_name(sb);
1166 /* Don't ever try to recover the MGS */
1167 rc = ptlrpc_set_import_active(
1168 lsi->lsi_mgc->u.cli.cl_import, 0);
1170 OBD_FREE(oldname, oldnamelen);
1173 /* end COMPAT_146 */
1175 CERROR("Unable to process log: %d\n", err);
1176 GOTO(out_free, err);
1179 lprof = class_get_profile(profilenm);
1180 if (lprof == NULL) {
1181 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1182 " read from the MGS. Does that filesystem "
1183 "exist?\n", profilenm);
1184 GOTO(out_free, err = -EINVAL);
1188 * The configuration for 1.8 client and 2.0 client are different.
1189 * 2.0 introduces lmv, but 1.8 directly uses mdc.
1190 * Here, we will hack to use proper name for mdc if needed.
1196 save = lprof->lp_mdc;
1197 fsname_end = strrchr(save, '-');
1199 namelen = fsname_end - save;
1200 if (strcmp(fsname_end, "-clilmv") == 0) {
1201 strncpy(pseudo, save, namelen);
1202 strcat(pseudo, "-MDT0000-mdc");
1203 lprof->lp_mdc = pseudo;
1204 CDEBUG(D_INFO, "1.8.x connecting to 2.0: lmv=%s"
1205 " new mdc=%s\n", save, pseudo);
1210 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1211 lprof->lp_mdc, lprof->lp_osc);
1213 OBD_ALLOC(osc, strlen(lprof->lp_osc) +
1214 strlen(ll_instance) + 2);
1216 GOTO(out_free, err = -ENOMEM);
1217 sprintf(osc, "%s-%s", lprof->lp_osc, ll_instance);
1219 OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
1220 strlen(ll_instance) + 2);
1222 GOTO(out_free, err = -ENOMEM);
1223 sprintf(mdc, "%s-%s", lprof->lp_mdc, ll_instance);
1225 /* connections, registrations, sb setup */
1226 err = client_common_fill_super(sb, mdc, osc);
1230 lprof->lp_mdc = save;
1232 OBD_FREE(mdc, strlen(mdc) + 1);
1234 OBD_FREE(osc, strlen(osc) + 1);
1238 LCONSOLE_WARN("Client %s has started\n", profilenm);
1241 } /* ll_fill_super */
1244 void ll_put_super(struct super_block *sb)
1246 struct config_llog_instance cfg;
1247 char ll_instance[sizeof(sb) * 2 + 1];
1248 struct obd_device *obd;
1249 struct lustre_sb_info *lsi = s2lsi(sb);
1250 struct ll_sb_info *sbi = ll_s2sbi(sb);
1251 char *profilenm = get_profile_name(sb);
1252 int force = 1, next;
1255 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1257 sprintf(ll_instance, "%p", sb);
1258 cfg.cfg_instance = ll_instance;
1259 lustre_end_log(sb, NULL, &cfg);
1261 if (sbi->ll_mdc_exp) {
1262 obd = class_exp2obd(sbi->ll_mdc_exp);
1264 force = obd->obd_force;
1267 /* We need to set force before the lov_disconnect in
1268 lustre_common_put_super, since l_d cleans up osc's as well. */
1271 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1273 obd->obd_force = force;
1278 /* Only if client_common_fill_super succeeded */
1279 client_common_put_super(sb);
1283 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1284 class_manual_cleanup(obd);
1288 class_del_profile(profilenm);
1291 lsi->lsi_llsbi = NULL;
1293 lustre_common_put_super(sb);
1295 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1300 } /* client_put_super */
1302 int ll_shrink_cache(int nr_to_scan, SHRINKER_MASK_T gfp_mask)
1304 struct ll_sb_info *sbi;
1307 if (gfp_mask & __GFP_FS)
1310 /* don't race with umount */
1311 down_read(&ll_sb_sem);
1312 list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1313 count += llap_shrink_cache(sbi, nr_to_scan);
1314 up_read(&ll_sb_sem);
1319 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1321 struct inode *inode = NULL;
1322 /* NOTE: we depend on atomic igrab() -bzzz */
1323 lock_res_and_lock(lock);
1324 if (lock->l_ast_data) {
1325 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1326 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1327 inode = igrab(lock->l_ast_data);
1329 inode = lock->l_ast_data;
1330 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1332 lock, __FILE__, __func__, __LINE__,
1333 "l_ast_data %p is bogus: magic %08x",
1334 lock->l_ast_data, lli->lli_inode_magic);
1338 unlock_res_and_lock(lock);
1342 static int null_if_equal(struct ldlm_lock *lock, void *data)
1344 if (data == lock->l_ast_data) {
1345 lock->l_ast_data = NULL;
1347 if (lock->l_req_mode != lock->l_granted_mode)
1348 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1351 return LDLM_ITER_CONTINUE;
1354 void ll_clear_inode(struct inode *inode)
1357 struct ll_inode_info *lli = ll_i2info(inode);
1358 struct ll_sb_info *sbi = ll_i2sbi(inode);
1361 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1362 inode->i_generation, inode);
1364 if (S_ISDIR(inode->i_mode)) {
1365 /* these should have been cleared in ll_file_release */
1366 LASSERT(lli->lli_sai == NULL);
1367 LASSERT(lli->lli_opendir_key == NULL);
1368 LASSERT(lli->lli_opendir_pid == 0);
1371 ll_inode2fid(&fid, inode);
1372 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1373 mdc_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
1375 LASSERT(!lli->lli_open_fd_write_count);
1376 LASSERT(!lli->lli_open_fd_read_count);
1377 LASSERT(!lli->lli_open_fd_exec_count);
1379 if (lli->lli_mds_write_och)
1380 ll_mdc_real_close(inode, FMODE_WRITE);
1381 if (lli->lli_mds_exec_och) {
1383 CERROR("No FMODE exec, bug exec och is present for "
1384 "inode %ld\n", inode->i_ino);
1385 ll_mdc_real_close(inode, FMODE_EXEC);
1387 if (lli->lli_mds_read_och)
1388 ll_mdc_real_close(inode, FMODE_READ);
1392 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
1393 null_if_equal, inode);
1395 obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
1396 lli->lli_smd = NULL;
1399 if (lli->lli_symlink_name) {
1400 OBD_FREE(lli->lli_symlink_name,
1401 strlen(lli->lli_symlink_name) + 1);
1402 lli->lli_symlink_name = NULL;
1405 #ifdef CONFIG_FS_POSIX_ACL
1406 if (lli->lli_posix_acl) {
1407 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1408 posix_acl_release(lli->lli_posix_acl);
1409 lli->lli_posix_acl = NULL;
1413 lli->lli_inode_magic = LLI_INODE_DEAD;
1415 #ifdef HAVE_EXPORT___IGET
1416 spin_lock(&sbi->ll_deathrow_lock);
1417 list_del_init(&lli->lli_dead_list);
1418 spin_unlock(&sbi->ll_deathrow_lock);
1423 static int ll_setattr_do_truncate(struct inode *inode, loff_t new_size)
1425 struct ll_sb_info *sbi = ll_i2sbi(inode);
1426 struct ll_inode_info *lli = ll_i2info(inode);
1427 struct lov_stripe_md *lsm = lli->lli_smd;
1429 ldlm_policy_data_t policy = { .l_extent = {new_size,
1431 struct lustre_handle lockh = { 0 };
1432 int local_lock = 1; /* 0 - no local lock;
1433 * 1 - lock taken by lock_extent;
1434 * 2 - by obd_match*/
1438 UNLOCK_INODE_MUTEX(inode);
1439 UP_WRITE_I_ALLOC_SEM(inode);
1441 down_write(&lli->lli_truncate_rwsem);
1442 if (sbi->ll_lockless_truncate_enable &&
1443 (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK)) {
1446 ast_flags = LDLM_FL_BLOCK_GRANTED;
1447 rc = obd_match(sbi->ll_osc_exp, lsm, LDLM_EXTENT,
1448 &policy, LCK_PW, &ast_flags, inode, &lockh,
1454 /* clear the lock handle as it not cleared
1455 * by obd_match if no matched lock found */
1457 if (rc == 0 && n_matches == 0) {
1459 rc = ll_file_punch(inode, new_size, 1);
1463 if (local_lock == 1) {
1464 /* XXX when we fix the AST intents to pass the discard-range
1465 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1467 ast_flags = (new_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
1468 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
1470 if (unlikely(rc != 0))
1474 LOCK_INODE_MUTEX(inode);
1475 DOWN_WRITE_I_ALLOC_SEM(inode);
1476 if (likely(rc == 0)) {
1477 /* Only ll_inode_size_lock is taken at this level.
1478 * lov_stripe_lock() is grabbed by ll_truncate() only over
1479 * call to obd_adjust_kms(). If vmtruncate returns 0, then
1480 * ll_truncate dropped ll_inode_size_lock() */
1481 ll_inode_size_lock(inode, 0);
1483 set_bit(LLI_F_SRVLOCK, &lli->lli_flags);
1484 rc = vmtruncate(inode, new_size);
1485 clear_bit(LLI_F_SRVLOCK, &lli->lli_flags);
1487 LASSERT(SEM_COUNT(&lli->lli_size_sem) <= 0);
1488 ll_inode_size_unlock(inode, 0);
1494 err = (local_lock == 2) ?
1495 obd_cancel(sbi->ll_osc_exp, lsm, LCK_PW, &lockh, 0, 0):
1496 ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1497 if (unlikely(err != 0)){
1498 CERROR("extent unlock failed: err=%d,"
1499 " unlock method =%d\n", err, local_lock);
1504 up_write(&lli->lli_truncate_rwsem);
1508 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1509 * object(s) determine the file size and mtime. Otherwise, the MDS will
1510 * keep these values until such a time that objects are allocated for it.
1511 * We do the MDS operations first, as it is checking permissions for us.
1512 * We don't to the MDS RPC if there is nothing that we want to store there,
1513 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1514 * going to do an RPC anyways.
1516 * If we are doing a truncate, we will send the mtime and ctime updates
1517 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1518 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1521 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1523 struct ll_inode_info *lli = ll_i2info(inode);
1524 struct lov_stripe_md *lsm = lli->lli_smd;
1525 struct ll_sb_info *sbi = ll_i2sbi(inode);
1526 struct ptlrpc_request *request = NULL;
1527 struct mdc_op_data *op_data;
1528 struct lustre_md md;
1529 int ia_valid = attr->ia_valid;
1533 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1535 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1537 if (ia_valid & ATTR_SIZE) {
1538 if (attr->ia_size > ll_file_maxbytes(inode)) {
1539 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1540 attr->ia_size, ll_file_maxbytes(inode));
1544 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1547 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1548 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1549 if (cfs_curproc_fsuid() != inode->i_uid &&
1550 !cfs_capable(CFS_CAP_FOWNER))
1554 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1555 if (attr->ia_valid & ATTR_CTIME) {
1556 attr->ia_ctime = CURRENT_TIME;
1557 attr->ia_valid |= ATTR_CTIME_SET;
1559 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1560 attr->ia_atime = CURRENT_TIME;
1561 attr->ia_valid |= ATTR_ATIME_SET;
1563 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1564 attr->ia_mtime = CURRENT_TIME;
1565 attr->ia_valid |= ATTR_MTIME_SET;
1568 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1569 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1570 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1573 /* NB: ATTR_SIZE will only be set after this point if the size
1574 * resides on the MDS, ie, this file has no objects. */
1576 attr->ia_valid &= ~ATTR_SIZE;
1578 OBD_ALLOC_PTR(op_data);
1579 if (NULL == op_data)
1581 /* We always do an MDS RPC, even if we're only changing the size;
1582 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1583 ll_prepare_mdc_op_data(op_data, inode, NULL, NULL, 0, 0, NULL);
1585 rc = mdc_setattr(sbi->ll_mdc_exp, op_data,
1586 attr, NULL, 0, NULL, 0, &request);
1589 ptlrpc_req_finished(request);
1590 if (rc == -ENOENT) {
1592 /* Unlinked special device node? Or just a race?
1593 * Pretend we done everything. */
1594 if (!S_ISREG(inode->i_mode) &&
1595 !S_ISDIR(inode->i_mode))
1596 rc = inode_setattr(inode, attr);
1597 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY)
1598 CERROR("mdc_setattr fails: rc = %d\n", rc);
1599 OBD_FREE_PTR(op_data);
1602 OBD_FREE_PTR(op_data);
1604 rc = mdc_req2lustre_md(request, REPLY_REC_OFF, sbi->ll_osc_exp, &md);
1606 ptlrpc_req_finished(request);
1610 /* We call inode_setattr to adjust timestamps.
1611 * If there is at least some data in file, we cleared ATTR_SIZE above to
1612 * avoid invoking vmtruncate, otherwise it is important to call
1613 * vmtruncate in inode_setattr to update inode->i_size (bug 6196) */
1614 rc = inode_setattr(inode, attr);
1616 ll_update_inode(inode, &md);
1617 ptlrpc_req_finished(request);
1619 if (!lsm || !S_ISREG(inode->i_mode)) {
1620 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1624 /* We really need to get our PW lock before we change inode->i_size.
1625 * If we don't we can race with other i_size updaters on our node, like
1626 * ll_file_read. We can also race with i_size propogation to other
1627 * nodes through dirtying and writeback of final cached pages. This
1628 * last one is especially bad for racing o_append users on other
1630 if (ia_valid & ATTR_SIZE) {
1631 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1632 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1633 struct obd_info *oinfo;
1635 struct lustre_handle lockh = { 0 };
1638 OBD_ALLOC_PTR(oinfo);
1642 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1643 inode->i_ino, LTIME_S(attr->ia_mtime));
1647 oa->o_id = lsm->lsm_object_id;
1648 oa->o_gr = lsm->lsm_object_gr;
1649 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1651 valid = OBD_MD_FLTYPE | OBD_MD_FLFID | OBD_MD_FLGENER;
1653 if (LTIME_S(attr->ia_mtime) < LTIME_S(attr->ia_ctime)){
1654 struct ost_lvb xtimes;
1656 /* setting mtime to past is performed under PW
1657 * EOF extent lock */
1658 oinfo->oi_policy.l_extent.start = 0;
1659 oinfo->oi_policy.l_extent.end = OBD_OBJECT_EOF;
1660 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW,
1664 OBD_FREE_PTR(oinfo);
1668 /* setattr under locks
1670 * 1. restore inode's timestamps which
1671 * are about to be set as long as
1672 * concurrent stat (via
1673 * ll_glimpse_size) might bring
1676 * 2. update lsm so that next stat
1677 * (via ll_glimpse_size) could get
1678 * correct values in lsm */
1679 lov_stripe_lock(lli->lli_smd);
1680 if (ia_valid & ATTR_ATIME) {
1681 LTIME_S(inode->i_atime) =
1683 LTIME_S(attr->ia_atime);
1684 valid |= OBD_MD_FLATIME;
1686 if (ia_valid & ATTR_MTIME) {
1687 LTIME_S(inode->i_mtime) =
1689 LTIME_S(attr->ia_mtime);
1690 valid |= OBD_MD_FLMTIME;
1692 if (ia_valid & ATTR_CTIME) {
1693 LTIME_S(inode->i_ctime) =
1695 LTIME_S(attr->ia_ctime);
1696 valid |= OBD_MD_FLCTIME;
1699 obd_update_lvb(ll_i2obdexp(inode), lli->lli_smd,
1701 lov_stripe_unlock(lli->lli_smd);
1705 * 1. do not use inode's timestamps
1706 * because concurrent stat might fill
1707 * the inode with out-of-date times,
1708 * send values from attr instead
1710 * 2.do no update lsm, as long as stat
1711 * (via ll_glimpse_size) will bring
1712 * attributes from osts anyway */
1713 if (ia_valid & ATTR_ATIME) {
1714 oa->o_atime = LTIME_S(attr->ia_atime);
1715 oa->o_valid |= OBD_MD_FLATIME;
1717 if (ia_valid & ATTR_MTIME) {
1718 oa->o_mtime = LTIME_S(attr->ia_mtime);
1719 oa->o_valid |= OBD_MD_FLMTIME;
1721 if (ia_valid & ATTR_CTIME) {
1722 oa->o_ctime = LTIME_S(attr->ia_ctime);
1723 oa->o_valid |= OBD_MD_FLCTIME;
1727 obdo_from_inode(oa, inode, valid);
1732 rc = obd_setattr_rqset(sbi->ll_osc_exp, oinfo, NULL);
1734 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1736 if (LTIME_S(attr->ia_mtime) < LTIME_S(attr->ia_ctime)){
1739 err = ll_extent_unlock(NULL, inode, lsm,
1741 if (unlikely(err != 0)) {
1742 CERROR("extent unlock failed: "
1752 OBD_FREE_PTR(oinfo);
1757 int ll_setattr(struct dentry *de, struct iattr *attr)
1759 int mode = de->d_inode->i_mode;
1761 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1762 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1763 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1764 if ((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1765 (ATTR_SIZE|ATTR_MODE)) {
1766 if (((mode & S_ISUID) && (!(attr->ia_mode & S_ISUID))) ||
1767 ((mode & S_ISGID) && (mode & S_IXGRP) &&
1768 (!(attr->ia_mode & S_ISGID))))
1769 attr->ia_valid |= ATTR_FORCE;
1772 if ((mode & S_ISUID) && S_ISREG(mode) &&
1773 !(attr->ia_mode & S_ISUID))
1774 attr->ia_valid |= ATTR_KILL_SUID;
1776 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1777 S_ISREG(mode) && !(attr->ia_mode & S_ISGID))
1778 attr->ia_valid |= ATTR_KILL_SGID;
1780 return ll_setattr_raw(de->d_inode, attr);
1783 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1784 __u64 max_age, __u32 flags)
1786 struct ll_sb_info *sbi = ll_s2sbi(sb);
1787 struct obd_statfs obd_osfs;
1791 rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age, flags);
1793 CERROR("mdc_statfs fails: rc = %d\n", rc);
1797 osfs->os_type = sb->s_magic;
1799 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1800 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1802 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1803 flags |= OBD_STATFS_NODELAY;
1805 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_osc_exp),
1806 &obd_osfs, max_age, flags);
1808 CERROR("obd_statfs fails: rc = %d\n", rc);
1812 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1813 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1816 osfs->os_bsize = obd_osfs.os_bsize;
1817 osfs->os_blocks = obd_osfs.os_blocks;
1818 osfs->os_bfree = obd_osfs.os_bfree;
1819 osfs->os_bavail = obd_osfs.os_bavail;
1821 /* If we don't have as many objects free on the OST as inodes
1822 * on the MDS, we reduce the total number of inodes to
1823 * compensate, so that the "inodes in use" number is correct.
1825 if (obd_osfs.os_ffree < osfs->os_ffree) {
1826 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1828 osfs->os_ffree = obd_osfs.os_ffree;
1833 #ifndef HAVE_STATFS_DENTRY_PARAM
1834 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1837 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1839 struct super_block *sb = de->d_sb;
1841 struct obd_statfs osfs;
1844 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1845 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1847 /* For now we will always get up-to-date statfs values, but in the
1848 * future we may allow some amount of caching on the client (e.g.
1849 * from QOS or lprocfs updates). */
1850 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1854 statfs_unpack(sfs, &osfs);
1856 /* We need to downshift for all 32-bit kernels, because we can't
1857 * tell if the kernel is being called via sys_statfs64() or not.
1858 * Stop before overflowing f_bsize - in which case it is better
1859 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1860 if (sizeof(long) < 8) {
1861 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1864 osfs.os_blocks >>= 1;
1865 osfs.os_bfree >>= 1;
1866 osfs.os_bavail >>= 1;
1870 sfs->f_blocks = osfs.os_blocks;
1871 sfs->f_bfree = osfs.os_bfree;
1872 sfs->f_bavail = osfs.os_bavail;
1877 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1879 struct ll_inode_info *lli;
1880 struct lov_stripe_md *lsm;
1882 lli = ll_i2info(inode);
1883 LASSERT(lli->lli_size_sem_owner != current);
1884 down(&lli->lli_size_sem);
1885 LASSERT(lli->lli_size_sem_owner == NULL);
1886 lli->lli_size_sem_owner = current;
1888 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1891 lov_stripe_lock(lsm);
1894 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1896 struct ll_inode_info *lli;
1897 struct lov_stripe_md *lsm;
1899 lli = ll_i2info(inode);
1901 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1904 lov_stripe_unlock(lsm);
1905 LASSERT(lli->lli_size_sem_owner == current);
1906 lli->lli_size_sem_owner = NULL;
1907 up(&lli->lli_size_sem);
1910 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1912 struct ll_inode_info *lli = ll_i2info(inode);
1914 dump_lsm(D_INODE, lsm);
1915 dump_lsm(D_INODE, lli->lli_smd);
1916 LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1917 "lsm must be joined lsm %p\n", lsm);
1918 obd_free_memmd(ll_i2obdexp(inode), &lli->lli_smd);
1919 CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1920 lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1922 lli->lli_maxbytes = lsm->lsm_maxbytes;
1923 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1924 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1927 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1929 struct ll_inode_info *lli = ll_i2info(inode);
1930 struct mds_body *body = md->body;
1931 struct lov_stripe_md *lsm = md->lsm;
1932 struct ll_sb_info *sbi = ll_i2sbi(inode);
1935 CDEBUG(D_INODE, "body->valid = "LPX64"\n", body->valid);
1937 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1939 if (lli->lli_smd == NULL) {
1940 if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1941 lsm->lsm_magic != LOV_MAGIC_V3 &&
1942 lsm->lsm_magic != LOV_MAGIC_JOIN) {
1943 dump_lsm(D_ERROR, lsm);
1946 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1947 lsm, inode->i_ino, inode->i_generation, inode);
1948 /* ll_inode_size_lock() requires it is only called
1949 * with lli_smd != NULL or lock_lsm == 0 or we can
1950 * race between lock/unlock. bug 9547 */
1952 lli->lli_maxbytes = lsm->lsm_maxbytes;
1953 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1954 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1956 if ((lli->lli_smd->lsm_magic == lsm->lsm_magic ||
1957 (lli->lli_smd->lsm_magic == LOV_MAGIC_V3 &&
1958 lsm->lsm_magic == LOV_MAGIC_V1) ||
1959 (lli->lli_smd->lsm_magic == LOV_MAGIC_V1 &&
1960 lsm->lsm_magic == LOV_MAGIC_V3)) &&
1961 lli->lli_smd->lsm_stripe_count ==
1962 lsm->lsm_stripe_count) {
1963 /* The MDS can suddenly change the magic to
1964 * v1/v3 if it has been upgraded/downgraded to a
1965 * version that does/doesn't support OST pools.
1966 * In this case, we consider that the cached lsm
1967 * is equivalent to the new one and keep it in
1968 * place until the inode is evicted from the
1970 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1971 CERROR("lsm mismatch for inode %ld\n",
1973 CERROR("lli_smd:\n");
1974 dump_lsm(D_ERROR, lli->lli_smd);
1976 dump_lsm(D_ERROR, lsm);
1980 ll_replace_lsm(inode, lsm);
1982 if (lli->lli_smd != lsm)
1983 obd_free_memmd(ll_i2obdexp(inode), &lsm);
1986 #ifdef CONFIG_FS_POSIX_ACL
1987 LASSERT(!md->posix_acl || (body->valid & OBD_MD_FLACL));
1988 if (body->valid & OBD_MD_FLACL) {
1989 spin_lock(&lli->lli_lock);
1990 if (lli->lli_posix_acl)
1991 posix_acl_release(lli->lli_posix_acl);
1992 lli->lli_posix_acl = md->posix_acl;
1993 spin_unlock(&lli->lli_lock);
1997 inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1998 inode->i_generation = ll_fid_build_gen(sbi, &body->fid1);
1999 *ll_inode_lu_fid(inode) = *((struct lu_fid*)&md->body->fid1);
2001 if (body->valid & OBD_MD_FLATIME) {
2002 if (body->atime > LTIME_S(inode->i_atime))
2003 LTIME_S(inode->i_atime) = body->atime;
2004 lli->lli_lvb.lvb_atime = body->atime;
2006 if (body->valid & OBD_MD_FLMTIME) {
2007 if (body->mtime > LTIME_S(inode->i_mtime)) {
2008 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
2009 "to "LPU64"\n", inode->i_ino,
2010 LTIME_S(inode->i_mtime), body->mtime);
2011 LTIME_S(inode->i_mtime) = body->mtime;
2013 lli->lli_lvb.lvb_mtime = body->mtime;
2015 if (body->valid & OBD_MD_FLCTIME) {
2016 if (body->ctime > LTIME_S(inode->i_ctime))
2017 LTIME_S(inode->i_ctime) = body->ctime;
2018 lli->lli_lvb.lvb_ctime = body->ctime;
2020 if (body->valid & OBD_MD_FLMODE)
2021 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
2022 if (body->valid & OBD_MD_FLTYPE)
2023 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
2024 if (S_ISREG(inode->i_mode)) {
2025 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS+1, LL_MAX_BLKSIZE_BITS);
2027 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2029 #ifdef HAVE_INODE_BLKSIZE
2030 inode->i_blksize = 1<<inode->i_blkbits;
2032 if (body->valid & OBD_MD_FLUID)
2033 inode->i_uid = body->uid;
2034 if (body->valid & OBD_MD_FLGID)
2035 inode->i_gid = body->gid;
2036 if (body->valid & OBD_MD_FLFLAGS)
2037 inode->i_flags = ll_ext_to_inode_flags(body->flags |
2038 MDS_BFLAG_EXT_FLAGS);
2039 if (body->valid & OBD_MD_FLNLINK)
2040 inode->i_nlink = body->nlink;
2042 if (body->valid & OBD_MD_FLRDEV)
2043 inode->i_rdev = old_decode_dev(body->rdev);
2044 if (body->valid & OBD_MD_FLSIZE) {
2045 #if 0 /* Can't block ll_test_inode->ll_update_inode, b=14326*/
2046 ll_inode_size_lock(inode, 0);
2047 i_size_write(inode, body->size);
2048 ll_inode_size_unlock(inode, 0);
2050 inode->i_size = body->size;
2053 if (body->valid & OBD_MD_FLBLOCKS)
2054 inode->i_blocks = body->blocks;
2056 if (body->valid & OBD_MD_FLSIZE)
2057 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
2061 static struct backing_dev_info ll_backing_dev_info = {
2062 .ra_pages = 0, /* No readahead */
2063 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
2064 .capabilities = 0, /* Does contribute to dirty memory */
2066 .memory_backed = 0, /* Does contribute to dirty memory */
2070 void ll_read_inode2(struct inode *inode, void *opaque)
2072 struct lustre_md *md = opaque;
2073 struct ll_inode_info *lli = ll_i2info(inode);
2076 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
2077 inode->i_generation, inode);
2081 LASSERT(!lli->lli_smd);
2083 /* Core attributes from the MDS first. This is a new inode, and
2084 * the VFS doesn't zero times in the core inode so we have to do
2085 * it ourselves. They will be overwritten by either MDS or OST
2086 * attributes - we just need to make sure they aren't newer. */
2087 LTIME_S(inode->i_mtime) = 0;
2088 LTIME_S(inode->i_atime) = 0;
2089 LTIME_S(inode->i_ctime) = 0;
2091 ll_update_inode(inode, md);
2093 /* OIDEBUG(inode); */
2095 if (S_ISREG(inode->i_mode)) {
2096 struct ll_sb_info *sbi = ll_i2sbi(inode);
2097 inode->i_op = &ll_file_inode_operations;
2098 inode->i_fop = sbi->ll_fop;
2099 inode->i_mapping->a_ops = &ll_aops;
2101 } else if (S_ISDIR(inode->i_mode)) {
2102 inode->i_op = &ll_dir_inode_operations;
2103 inode->i_fop = &ll_dir_operations;
2104 inode->i_mapping->a_ops = &ll_dir_aops;
2106 } else if (S_ISLNK(inode->i_mode)) {
2107 inode->i_op = &ll_fast_symlink_inode_operations;
2110 inode->i_op = &ll_special_inode_operations;
2111 init_special_inode(inode, inode->i_mode,
2112 kdev_t_to_nr(inode->i_rdev));
2113 /* initializing backing dev info. */
2114 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2119 int ll_iocontrol(struct inode *inode, struct file *file,
2120 unsigned int cmd, unsigned long arg)
2122 struct ll_sb_info *sbi = ll_i2sbi(inode);
2123 struct ptlrpc_request *req = NULL;
2128 case FSFILT_IOC_GETFLAGS: {
2130 struct mds_body *body;
2132 ll_inode2fid(&fid, inode);
2133 rc = mdc_getattr(sbi->ll_mdc_exp, &fid, OBD_MD_FLFLAGS,0,&req);
2135 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2139 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
2142 /* We want to return EXT3_*_FL flags to the caller via this
2143 * ioctl. An older MDS may be sending S_* flags, fix it up. */
2144 flags = ll_inode_to_ext_flags(body->flags,
2145 MDS_BFLAG_EXT_FLAGS);
2146 ptlrpc_req_finished (req);
2148 RETURN(put_user(flags, (int *)arg));
2150 case FSFILT_IOC_SETFLAGS: {
2151 struct mdc_op_data op_data = { { 0 } };
2152 struct ll_iattr_struct attr;
2153 struct obd_info oinfo = { { { 0 } } };
2154 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2156 if (get_user(flags, (int *)arg))
2160 OBDO_ALLOC(oinfo.oi_oa);
2164 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0, NULL);
2166 memset(&attr, 0, sizeof(attr));
2167 attr.ia_attr_flags = flags;
2168 ((struct iattr *)&attr)->ia_valid |= ATTR_ATTR_FLAG;
2170 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
2171 (struct iattr *)&attr, NULL, 0, NULL, 0, &req);
2172 ptlrpc_req_finished(req);
2174 OBDO_FREE(oinfo.oi_oa);
2179 OBDO_FREE(oinfo.oi_oa);
2180 GOTO(update_cache, rc);
2183 oinfo.oi_oa->o_id = lsm->lsm_object_id;
2184 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
2185 oinfo.oi_oa->o_flags = flags;
2186 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP |
2189 obdo_from_inode(oinfo.oi_oa, inode,
2190 OBD_MD_FLFID | OBD_MD_FLGENER);
2191 rc = obd_setattr_rqset(sbi->ll_osc_exp, &oinfo, NULL);
2192 OBDO_FREE(oinfo.oi_oa);
2194 if (rc != -EPERM && rc != -EACCES)
2195 CERROR("osc_setattr_async fails: rc = %d\n",rc);
2201 inode->i_flags = ll_ext_to_inode_flags(flags |
2202 MDS_BFLAG_EXT_FLAGS);
2212 /* umount -f client means force down, don't save state */
2213 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2214 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2216 struct super_block *sb = vfsmnt->mnt_sb;
2218 void ll_umount_begin(struct super_block *sb)
2221 struct lustre_sb_info *lsi = s2lsi(sb);
2222 struct ll_sb_info *sbi = ll_s2sbi(sb);
2223 struct obd_device *obd;
2224 struct obd_ioctl_data ioc_data = { 0 };
2227 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2228 if (!(flags & MNT_FORCE)) {
2234 /* Tell the MGC we got umount -f */
2235 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2237 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2238 sb->s_count, atomic_read(&sb->s_active));
2240 obd = class_exp2obd(sbi->ll_mdc_exp);
2242 CERROR("Invalid MDC connection handle "LPX64"\n",
2243 sbi->ll_mdc_exp->exp_handle.h_cookie);
2248 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
2251 obd = class_exp2obd(sbi->ll_osc_exp);
2253 CERROR("Invalid LOV connection handle "LPX64"\n",
2254 sbi->ll_osc_exp->exp_handle.h_cookie);
2260 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
2263 /* Really, we'd like to wait until there are no requests outstanding,
2264 * and then continue. For now, we just invalidate the requests,
2265 * schedule() and sleep one second if needed, and hope.
2268 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2269 if (atomic_read(&vfsmnt->mnt_count) > 2) {
2270 cfs_schedule_timeout(CFS_TASK_INTERRUPTIBLE,
2271 cfs_time_seconds(1));
2272 if (atomic_read(&vfsmnt->mnt_count) > 2)
2273 LCONSOLE_WARN("Mount still busy with %d refs! You "
2274 "may try to umount it a bit later\n",
2275 atomic_read(&vfsmnt->mnt_count));
2282 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2284 struct ll_sb_info *sbi = ll_s2sbi(sb);
2288 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2289 read_only = *flags & MS_RDONLY;
2290 err = obd_set_info_async(sbi->ll_mdc_exp, sizeof(KEY_READONLY),
2291 KEY_READONLY, sizeof(read_only),
2294 /* MDS might have expected a different ro key value, b=17493 */
2295 if (err == -EINVAL) {
2296 CDEBUG(D_CONFIG, "Retrying remount with 1.6.6 ro key\n");
2297 err = obd_set_info_async(sbi->ll_mdc_exp,
2298 sizeof(KEY_READONLY_166COMPAT),
2299 KEY_READONLY_166COMPAT,
2305 CERROR("Failed to change the read-only flag during "
2306 "remount: %d\n", err);
2311 sb->s_flags |= MS_RDONLY;
2313 sb->s_flags &= ~MS_RDONLY;
2318 int ll_prep_inode(struct obd_export *exp, struct inode **inode,
2319 struct ptlrpc_request *req, int offset,struct super_block *sb)
2321 struct lustre_md md;
2322 struct ll_sb_info *sbi = NULL;
2326 LASSERT(*inode || sb);
2327 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2328 prune_deathrow(sbi, 1);
2330 rc = mdc_req2lustre_md(req, offset, exp, &md);
2335 ll_update_inode(*inode, &md);
2338 /** hashing VFS inode by FIDs.
2339 * IGIF will be used for for compatibility if needed.
2341 *inode =ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2342 if (*inode == NULL || is_bad_inode(*inode)) {
2343 mdc_free_lustre_md(exp, &md);
2345 CERROR("new_inode -fatal: rc %d\n", rc);
2350 rc = obd_checkmd(exp, ll_i2mdcexp(*inode),
2351 ll_i2info(*inode)->lli_smd);
2356 char *llap_origins[] = {
2357 [LLAP_ORIGIN_UNKNOWN] = "--",
2358 [LLAP_ORIGIN_READPAGE] = "rp",
2359 [LLAP_ORIGIN_READAHEAD] = "ra",
2360 [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2361 [LLAP_ORIGIN_WRITEPAGE] = "wp"
2364 struct ll_async_page *llite_pglist_next_llap(struct list_head *head,
2365 struct list_head *list)
2367 struct ll_async_page *llap;
2368 struct list_head *pos;
2370 list_for_each(pos, list) {
2373 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2374 if (llap->llap_page == NULL)
2382 int ll_obd_statfs(struct inode *inode, void *arg)
2384 struct ll_sb_info *sbi = NULL;
2385 struct obd_device *client_obd = NULL, *lov_obd = NULL;
2386 struct lov_obd *lov = NULL;
2387 struct obd_statfs stat_buf = {0};
2389 struct obd_ioctl_data *data = NULL;
2393 if (!inode || !(sbi = ll_i2sbi(inode)))
2394 GOTO(out_statfs, rc = -EINVAL);
2396 rc = obd_ioctl_getdata(&buf, &len, arg);
2398 GOTO(out_statfs, rc);
2401 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2402 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2403 GOTO(out_statfs, rc = -EINVAL);
2405 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2406 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2408 if (type == LL_STATFS_MDC) {
2410 GOTO(out_statfs, rc = -ENODEV);
2411 client_obd = class_exp2obd(sbi->ll_mdc_exp);
2412 } else if (type == LL_STATFS_LOV) {
2413 lov_obd = class_exp2obd(sbi->ll_osc_exp);
2414 lov = &lov_obd->u.lov;
2416 if (index >= lov->desc.ld_tgt_count)
2417 GOTO(out_statfs, rc = -ENODEV);
2419 if (!lov->lov_tgts[index])
2420 /* Try again with the next index */
2421 GOTO(out_statfs, rc = -EAGAIN);
2423 client_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2424 if (!lov->lov_tgts[index]->ltd_active)
2425 GOTO(out_uuid, rc = -ENODATA);
2429 GOTO(out_statfs, rc = -EINVAL);
2431 rc = obd_statfs(client_obd, &stat_buf, cfs_time_current_64() - HZ, 1);
2433 GOTO(out_statfs, rc);
2435 if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2436 GOTO(out_statfs, rc = -EFAULT);
2440 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(client_obd),
2444 if (copy_to_user(data->ioc_pbuf2, "Unknown UUID",
2445 sizeof("Unknown UUID") + 1))
2451 obd_ioctl_freedata(buf, len);
2455 int ll_process_config(struct lustre_cfg *lcfg)
2459 struct lprocfs_static_vars lvars;
2463 lprocfs_llite_init_vars(&lvars);
2465 /* The instance name contains the sb: lustre-client-aacfe000 */
2466 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2467 if (!ptr || !*(++ptr))
2469 if (sscanf(ptr, "%lx", &x) != 1)
2472 /* This better be a real Lustre superblock! */
2473 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2475 /* Note we have not called client_common_fill_super yet, so
2476 proc fns must be able to handle that! */
2477 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2482 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2484 struct ll_sb_info *sbi;
2486 LASSERT((seq != NULL) && (vfs != NULL));
2487 sbi = ll_s2sbi(vfs->mnt_sb);
2489 if (sbi->ll_flags & LL_SBI_NOLCK)
2490 seq_puts(seq, ",nolock");
2492 if (sbi->ll_flags & LL_SBI_FLOCK)
2493 seq_puts(seq, ",flock");
2495 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2496 seq_puts(seq, ",localflock");
2498 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2499 seq_puts(seq, ",user_xattr");
2501 if (sbi->ll_flags & LL_SBI_ACL)
2502 seq_puts(seq, ",acl");
2504 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2505 seq_puts(seq, ",lazystatfs");