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 (c) 2003, 2010, Oracle and/or its affiliates. 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.
37 #ifndef LLITE_INTERNAL_H
38 #define LLITE_INTERNAL_H
40 #ifdef CONFIG_FS_POSIX_ACL
41 # include <linux/fs.h>
43 # include <linux/xattr_acl.h>
45 #ifdef HAVE_LINUX_POSIX_ACL_XATTR_H
46 # include <linux/posix_acl_xattr.h>
50 #include <lustre_debug.h>
51 #include <lustre_ver.h>
52 #include <linux/lustre_version.h>
53 #include <lustre_disk.h> /* for s2sbi */
61 struct lustre_intent_data {
62 __u64 it_lock_handle[2];
68 /* If there is no FMODE_EXEC defined, make it to match nothing */
73 #ifndef DCACHE_LUSTRE_INVALID
74 #define DCACHE_LUSTRE_INVALID 0x100
77 #define LL_IT2STR(it) ((it) ? ldlm_it2str((it)->it_op) : "0")
78 #define LUSTRE_FPRIVATE(file) ((file)->private_data)
80 #ifdef HAVE_VFS_INTENT_PATCHES
81 static inline struct lookup_intent *ll_nd2it(struct nameidata *nd)
88 * Directory entries are currently in the same format as ext2/ext3, but will
89 * be changed in the future to accomodate FIDs
91 #define LL_DIR_NAME_LEN (255)
92 #define LL_DIR_PAD (4)
95 /* number of inode, referenced by this entry */
97 /* total record length, multiple of LL_DIR_PAD */
101 /* file type: regular, directory, device, etc. */
103 /* name. NOT NUL-terminated */
104 char lde_name[LL_DIR_NAME_LEN];
107 struct ll_dentry_data {
110 struct obd_client_handle lld_cwd_och;
111 struct obd_client_handle lld_mnt_och;
112 #ifndef HAVE_VFS_INTENT_PATCHES
113 struct lookup_intent *lld_it;
115 unsigned int lld_sa_generation;
118 #define ll_d2d(de) ((struct ll_dentry_data*)((de)->d_fsdata))
120 extern struct file_operations ll_pgcache_seq_fops;
122 #define LLI_INODE_MAGIC 0x111d0de5
123 #define LLI_INODE_DEAD 0xdeadd00d
124 #define LLI_F_HAVE_OST_SIZE_LOCK 0
125 #define LLI_F_HAVE_MDS_SIZE_LOCK 1
126 #define LLI_F_CONTENDED 2
127 #define LLI_F_SRVLOCK 3
129 struct ll_inode_info {
131 struct semaphore lli_size_sem; /* protect open and change size */
132 void *lli_size_sem_owner;
133 struct semaphore lli_write_sem;
134 struct lov_stripe_md *lli_smd;
135 char *lli_symlink_name;
138 unsigned long lli_flags;
139 cfs_time_t lli_contention_time;
141 /* this lock protects s_d_w and p_w_ll and mmap_cnt;
142 * atomic check-update of lli_smd */
144 #ifdef HAVE_CLOSE_THREAD
145 struct list_head lli_pending_write_llaps;
146 struct list_head lli_close_item;
147 int lli_send_done_writing;
149 atomic_t lli_mmap_cnt;
151 /* for writepage() only to communicate to fsync */
154 struct posix_acl *lli_posix_acl;
156 struct list_head lli_dead_list;
158 struct semaphore lli_och_sem; /* Protects access to och pointers
159 and their usage counters */
160 /* We need all three because every inode may be opened in different
162 struct obd_client_handle *lli_mds_read_och;
163 __u64 lli_open_fd_read_count;
164 struct obd_client_handle *lli_mds_write_och;
165 __u64 lli_open_fd_write_count;
166 struct obd_client_handle *lli_mds_exec_och;
167 __u64 lli_open_fd_exec_count;
169 /** fid of this object. */
175 /* metadata stat-ahead */
177 * "opendir_pid" is the token when lookup/revalid -- I am the owner of
180 pid_t lli_opendir_pid;
182 * since parent-child threads can share the same @file struct,
183 * "opendir_key" is the token when dir close for case of parent exit
184 * before child -- it is me should cleanup the dir readahead. */
185 void *lli_opendir_key;
186 struct ll_statahead_info *lli_sai;
187 struct rw_semaphore lli_truncate_rwsem;
188 /* the most recent attributes from mds, it is used for timestamps
190 struct ost_lvb lli_lvb;
191 struct inode lli_vfs_inode;
195 * Locking to guarantee consistency of non-atomic updates to long long i_size,
196 * consistency between file size and KMS, and consistency within
197 * ->lli_smd->lsm_oinfo[]'s.
199 * Implemented by ->lli_size_sem and ->lsm_sem, nested in that order.
202 void ll_inode_size_lock(struct inode *inode, int lock_lsm);
203 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm);
205 // FIXME: replace the name of this with LL_I to conform to kernel stuff
206 // static inline struct ll_inode_info *LL_I(struct inode *inode)
207 static inline struct ll_inode_info *ll_i2info(struct inode *inode)
209 return container_of(inode, struct ll_inode_info, lli_vfs_inode);
212 /* default to about 40meg of readahead on a given system. That much tied
213 * up in 512k readahead requests serviced at 40ms each is about 1GB/s. */
214 #define SBI_DEFAULT_READAHEAD_MAX (40UL << (20 - CFS_PAGE_SHIFT))
216 /* default to read-ahead full files smaller than 2MB on the second read */
217 #define SBI_DEFAULT_READAHEAD_WHOLE_MAX (2UL << (20 - CFS_PAGE_SHIFT))
222 RA_STAT_DISTANT_READPAGE,
223 RA_STAT_MISS_IN_WINDOW,
224 RA_STAT_FAILED_GRAB_PAGE,
225 RA_STAT_FAILED_MATCH,
230 RA_STAT_MAX_IN_FLIGHT,
231 RA_STAT_WRONG_GRAB_PAGE,
235 #define LL_RA_STAT _NR_RA_STAT
236 #define LL_RA_STAT_STRINGS { \
237 [RA_STAT_HIT] = "hits", \
238 [RA_STAT_MISS] = "misses", \
239 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive", \
240 [RA_STAT_MISS_IN_WINDOW] = "miss inside window", \
241 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page", \
242 [RA_STAT_FAILED_MATCH] = "failed lock match", \
243 [RA_STAT_DISCARDED] = "read but discarded", \
244 [RA_STAT_ZERO_LEN] = "zero length file", \
245 [RA_STAT_ZERO_WINDOW] = "zero size window", \
246 [RA_STAT_EOF] = "read-ahead to EOF", \
247 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue", \
248 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",\
252 atomic_t ra_cur_pages;
253 unsigned long ra_max_pages;
254 unsigned long ra_max_pages_per_file;
255 unsigned long ra_max_read_ahead_whole_pages;
258 /* LL_HIST_MAX=32 causes an overflow */
259 #define LL_HIST_MAX 28
260 #define LL_HIST_START 12 /* buckets start at 2^12 = 4k */
261 #define LL_PROCESS_HIST_MAX 10
262 struct per_process_info {
264 struct obd_histogram pp_r_hist;
265 struct obd_histogram pp_w_hist;
268 /* pp_extents[LL_PROCESS_HIST_MAX] will hold the combined process info */
269 struct ll_rw_extents_info {
270 struct per_process_info pp_extents[LL_PROCESS_HIST_MAX + 1];
273 #define LL_OFFSET_HIST_MAX 100
274 struct ll_rw_process_info {
277 loff_t rw_range_start;
279 loff_t rw_last_file_pos;
281 size_t rw_smallest_extent;
282 size_t rw_largest_extent;
283 struct file *rw_last_file;
287 enum stats_track_type {
288 STATS_TRACK_ALL = 0, /* track all processes */
289 STATS_TRACK_PID, /* track process with this pid */
290 STATS_TRACK_PPID, /* track processes with this ppid */
291 STATS_TRACK_GID, /* track processes with this gid */
295 /* flags for sbi->ll_flags */
296 #define LL_SBI_NOLCK 0x01 /* DLM locking disabled (directio-only) */
297 #define LL_SBI_DATA_CHECKSUM 0x02 /* checksum each page on the wire */
298 #define LL_SBI_FLOCK 0x04
299 #define LL_SBI_USER_XATTR 0x08 /* support user xattr */
300 #define LL_SBI_ACL 0x10 /* support ACL */
301 #define LL_SBI_JOIN 0x20 /* support JOIN */
302 #define LL_SBI_LOCALFLOCK 0x40 /* Local flocks support by kernel */
303 #define LL_SBI_LRU_RESIZE 0x80 /* support lru resize */
304 #define LL_SBI_LLITE_CHECKSUM 0x100 /* checksum each page in memory */
305 #define LL_SBI_LAZYSTATFS 0x200 /* lazystatfs mount option */
307 /* default value for ll_sb_info->contention_time */
308 #define SBI_DEFAULT_CONTENTION_SECONDS 60
309 /* default value for lockless_truncate_enable */
310 #define SBI_DEFAULT_LOCKLESS_TRUNCATE_ENABLE 1
311 /* default value for ll_direct_io_default */
312 #define SBI_DEFAULT_DIRECT_IO_DEFAULT 0
314 /* percpu data structure for lustre lru page list */
315 struct ll_pglist_data {
316 spinlock_t llpd_lock; /* lock to protect llpg_list */
317 struct list_head llpd_list; /* all pages (llap_pglist_item) */
318 unsigned long llpd_gen; /* generation # of this list */
319 unsigned long llpd_count; /* How many pages in this list */
320 atomic_t llpd_sample_count;
321 unsigned long llpd_reblnc_count;
322 /* the pages in this list shouldn't be over this number */
323 unsigned long llpd_budget;
325 /* which page the pglist data is in */
326 struct page *llpd_page;
329 unsigned long llpd_hit;
330 unsigned long llpd_miss;
331 unsigned long llpd_cross;
335 struct list_head ll_list;
336 /* this protects pglist(only ll_async_page_max) and ra_info.
337 * It isn't safe to grab from interrupt contexts. */
339 spinlock_t ll_pp_extent_lock; /* Lock for pp_extent entries */
340 spinlock_t ll_process_lock; /* Lock for ll_rw_process_info */
341 struct obd_uuid ll_sb_uuid;
342 struct obd_export *ll_mdc_exp;
343 struct obd_export *ll_osc_exp;
344 struct proc_dir_entry *ll_proc_root;
345 obd_id ll_rootino; /* number of root inode */
348 struct list_head ll_conn_chain; /* per-conn chain of SBs */
349 struct lustre_client_ocd ll_lco;
351 struct list_head ll_orphan_dentry_list; /*please don't ask -p*/
352 struct ll_close_queue *ll_lcq;
354 struct lprocfs_stats *ll_stats; /* lprocfs stats counter */
356 /* reblnc lock protects llpd_budget */
357 spinlock_t ll_async_page_reblnc_lock;
358 unsigned long ll_async_page_reblnc_count;
359 unsigned long ll_async_page_sample_max;
360 /* I defined this array here rather than in ll_pglist_data
361 * because it is always accessed by only one cpu. -jay */
362 unsigned long *ll_async_page_sample;
363 unsigned long ll_async_page_max;
364 unsigned long ll_async_page_clock_hand;
365 lcounter_t ll_async_page_count;
366 struct ll_pglist_data **ll_pglist;
368 struct lprocfs_stats *ll_ra_stats;
370 unsigned ll_contention_time; /* seconds */
371 unsigned ll_lockless_truncate_enable; /* true/false */
372 unsigned ll_direct_io_default; /* true/false */
374 struct ll_ra_info ll_ra_info;
375 unsigned int ll_namelen;
376 struct file_operations *ll_fop;
378 /* =0 - hold lock over whole read/write
379 * >0 - max. chunk to be read/written w/o lock re-acquiring */
380 unsigned long ll_max_rw_chunk;
383 struct ll_rw_extents_info ll_rw_extents_info;
384 int ll_extent_process_count;
385 struct ll_rw_process_info ll_rw_process_info[LL_PROCESS_HIST_MAX];
386 unsigned int ll_offset_process_count;
387 struct ll_rw_process_info ll_rw_offset_info[LL_OFFSET_HIST_MAX];
388 unsigned int ll_rw_offset_entry_count;
389 enum stats_track_type ll_stats_track_type;
390 int ll_stats_track_id;
392 dev_t ll_sdev_orig; /* save s_dev before assign for
395 /* metadata stat-ahead */
396 unsigned int ll_sa_max; /* max statahead RPCs */
397 unsigned int ll_sa_wrong; /* statahead thread stopped for
399 unsigned int ll_sa_total; /* statahead thread started
401 unsigned long long ll_sa_blocked; /* ls count waiting for
403 unsigned long long ll_sa_cached; /* ls count got in cache */
404 unsigned long long ll_sa_hit; /* hit count */
405 unsigned long long ll_sa_miss; /* miss count */
408 #define LL_DEFAULT_MAX_RW_CHUNK (32 * 1024 * 1024)
410 #define LL_PGLIST_DATA_CPU(sbi, cpu) ((sbi)->ll_pglist[cpu])
411 #define LL_PGLIST_DATA(sbi) LL_PGLIST_DATA_CPU(sbi, smp_processor_id())
413 static inline struct ll_pglist_data *ll_pglist_cpu_lock(
414 struct ll_sb_info *sbi,
417 spin_lock(&sbi->ll_pglist[cpu]->llpd_lock);
418 return LL_PGLIST_DATA_CPU(sbi, cpu);
421 static inline void ll_pglist_cpu_unlock(struct ll_sb_info *sbi, int cpu)
423 spin_unlock(&sbi->ll_pglist[cpu]->llpd_lock);
426 static inline struct ll_pglist_data *ll_pglist_double_lock(
427 struct ll_sb_info *sbi,
428 int cpu, struct ll_pglist_data **pd_cpu)
430 int current_cpu = cfs_get_cpu();
432 if (cpu == current_cpu) {
433 ll_pglist_cpu_lock(sbi, cpu);
434 } else if (current_cpu < cpu) {
435 ll_pglist_cpu_lock(sbi, current_cpu);
436 ll_pglist_cpu_lock(sbi, cpu);
438 ll_pglist_cpu_lock(sbi, cpu);
439 ll_pglist_cpu_lock(sbi, current_cpu);
443 *pd_cpu = LL_PGLIST_DATA_CPU(sbi, cpu);
445 return LL_PGLIST_DATA(sbi);
448 static inline void ll_pglist_double_unlock(struct ll_sb_info *sbi, int cpu)
450 int current_cpu = smp_processor_id();
451 if (cpu == current_cpu) {
452 ll_pglist_cpu_unlock(sbi, cpu);
454 ll_pglist_cpu_unlock(sbi, cpu);
455 ll_pglist_cpu_unlock(sbi, current_cpu);
460 static inline struct ll_pglist_data *ll_pglist_lock(struct ll_sb_info *sbi)
462 ll_pglist_cpu_lock(sbi, cfs_get_cpu());
463 return LL_PGLIST_DATA(sbi);
466 static inline void ll_pglist_unlock(struct ll_sb_info *sbi)
468 ll_pglist_cpu_unlock(sbi, smp_processor_id());
475 struct task_struct *lrr_reader;
476 struct list_head lrr_linkage;
480 * per file-descriptor read-ahead data.
482 struct ll_readahead_state {
485 * index of the last page that read(2) needed and that wasn't in the
486 * cache. Used by ras_update() to detect seeks.
488 * XXX nikita: if access seeks into cached region, Lustre doesn't see
491 unsigned long ras_last_readpage;
493 * number of pages read after last read-ahead window reset. As window
494 * is reset on each seek, this is effectively a number of consecutive
495 * accesses. Maybe ->ras_accessed_in_window is better name.
497 * XXX nikita: window is also reset (by ras_update()) when Lustre
498 * believes that memory pressure evicts read-ahead pages. In that
499 * case, it probably doesn't make sense to expand window to
500 * PTLRPC_MAX_BRW_PAGES on the third access.
502 unsigned long ras_consecutive_pages;
504 * number of read requests after the last read-ahead window reset
505 * As window is reset on each seek, this is effectively the number
506 * on consecutive read request and is used to trigger read-ahead.
508 unsigned long ras_consecutive_requests;
510 * Parameters of current read-ahead window. Handled by
511 * ras_update(). On the initial access to the file or after a seek,
512 * window is reset to 0. After 3 consecutive accesses, window is
513 * expanded to PTLRPC_MAX_BRW_PAGES. Afterwards, window is enlarged by
514 * PTLRPC_MAX_BRW_PAGES chunks up to ->ra_max_pages.
516 unsigned long ras_window_start, ras_window_len;
518 * Where next read-ahead should start at. This lies within read-ahead
519 * window. Read-ahead window is read in pieces rather than at once
520 * because: 1. lustre limits total number of pages under read-ahead by
521 * ->ra_max_pages (see ll_ra_count_get()), 2. client cannot read pages
522 * not covered by DLM lock.
524 unsigned long ras_next_readahead;
526 * Total number of ll_file_read requests issued, reads originating
527 * due to mmap are not counted in this total. This value is used to
528 * trigger full file read-ahead after multiple reads to a small file.
530 unsigned long ras_requests;
532 * Page index with respect to the current request, these value
533 * will not be accurate when dealing with reads issued via mmap.
535 unsigned long ras_request_index;
537 * list of struct ll_ra_read's one per read(2) call current in
538 * progress against this file descriptor. Used by read-ahead code,
539 * protected by ->ras_lock.
541 struct list_head ras_read_beads;
543 * The following 3 items are used for detecting the stride I/O
545 * In stride I/O mode,
546 * ...............|-----data-----|****gap*****|--------|******|....
547 * offset |-stride_pages-|-stride_gap-|
548 * ras_stride_offset = offset;
549 * ras_stride_length = stride_pages + stride_gap;
550 * ras_stride_pages = stride_pages;
551 * Note: all these three items are counted by pages.
553 unsigned long ras_stride_length;
554 unsigned long ras_stride_pages;
555 pgoff_t ras_stride_offset;
557 * number of consecutive stride request count, and it is similar as
558 * ras_consecutive_requests, but used for stride I/O mode.
559 * Note: only more than 2 consecutive stride request are detected,
560 * stride read-ahead will be enable
562 unsigned long ras_consecutive_stride_requests;
565 extern cfs_mem_cache_t *ll_file_data_slab;
566 extern struct rw_semaphore ll_sb_sem;
567 struct lustre_handle;
568 struct ll_file_data {
569 struct ll_readahead_state fd_ras;
571 struct lustre_handle fd_cwlockh;
572 unsigned long fd_gid;
576 struct lov_stripe_md;
578 extern spinlock_t inode_lock;
580 extern struct proc_dir_entry *proc_lustre_fs_root;
582 static inline struct inode *ll_info2i(struct ll_inode_info *lli)
584 return &lli->lli_vfs_inode;
588 struct inode *icbd_parent;
589 struct dentry **icbd_childp;
593 void ll_i2gids(__u32 *suppgids, struct inode *i1,struct inode *i2);
595 #define LLAP_MAGIC 98764321
597 extern cfs_mem_cache_t *ll_async_page_slab;
598 extern size_t ll_async_page_slab_size;
599 struct ll_async_page {
601 /* only trust these if the page lock is providing exclusion */
602 unsigned int llap_write_queued:1,
603 llap_defer_uptodate:1,
608 unsigned int llap_pglist_cpu:16;
610 struct page *llap_page;
611 struct list_head llap_pending_write;
612 struct list_head llap_pglist_item;
613 /* checksum for paranoid I/O debugging */
615 struct lustre_handle llap_lockh_granted;
619 * enumeration of llap_from_page() call-sites. Used to export statistics in
620 * /proc/fs/lustre/llite/fsN/dump_page_cache.
623 LLAP_ORIGIN_UNKNOWN = 0,
624 LLAP_ORIGIN_READPAGE,
625 LLAP_ORIGIN_READAHEAD,
626 LLAP_ORIGIN_COMMIT_WRITE,
627 LLAP_ORIGIN_WRITEPAGE,
628 LLAP_ORIGIN_REMOVEPAGE,
631 extern char *llap_origins[];
633 void ll_ra_read_init(struct file *f, struct ll_ra_read *rar,
634 loff_t offset, size_t count);
635 void ll_ra_read_ex(struct file *f, struct ll_ra_read *rar);
636 struct ll_ra_read *ll_ra_read_get(struct file *f);
638 /* llite/lproc_llite.c */
640 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
641 struct super_block *sb, char *osc, char *mdc);
642 void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi);
643 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count);
644 void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars);
646 static inline int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
647 struct super_block *sb, char *osc, char *mdc){return 0;}
648 static inline void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi) {}
649 static void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count) {}
650 static void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars)
652 memset(lvars, 0, sizeof(*lvars));
658 extern struct file_operations ll_dir_operations;
659 extern struct inode_operations ll_dir_inode_operations;
661 struct page *ll_get_dir_page(struct inode *dir, unsigned long n);
663 static inline unsigned ll_dir_rec_len(unsigned name_len)
665 return (name_len + 8 + LL_DIR_PAD - 1) & ~(LL_DIR_PAD - 1);
668 static inline struct ll_dir_entry *ll_entry_at(void *base, unsigned offset)
670 return (struct ll_dir_entry *)((char *)base + offset);
674 * p is at least 6 bytes before the end of page
676 static inline struct ll_dir_entry *ll_dir_next_entry(struct ll_dir_entry *p)
678 return ll_entry_at(p, le16_to_cpu(p->lde_rec_len));
681 static inline void ll_put_page(struct page *page)
684 page_cache_release(page);
687 static inline unsigned long dir_pages(struct inode *inode)
689 return (inode->i_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
692 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir);
693 struct inode *ll_iget(struct super_block *sb, ino_t hash,
694 struct lustre_md *lic);
695 int ll_mdc_cancel_unused(struct lustre_handle *, struct inode *, int flags,
697 int ll_mdc_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
698 void *data, int flag);
699 int ll_prepare_mdc_op_data(struct mdc_op_data *,
700 struct inode *i1, struct inode *i2,
701 const char *name, int namelen, int mode, void *data);
702 #ifndef HAVE_VFS_INTENT_PATCHES
703 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
706 void ll_pin_extent_cb(void *data);
707 int ll_page_removal_cb(void *data, int discard);
708 int ll_extent_lock_cancel_cb(struct ldlm_lock *lock, struct ldlm_lock_desc *new,
709 void *data, int flag);
710 int lookup_it_finish(struct ptlrpc_request *request, int offset,
711 struct lookup_intent *it, void *data);
712 void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry);
715 int ll_prepare_write(struct file *, struct page *, unsigned from, unsigned to);
716 int ll_commit_write(struct file *, struct page *, unsigned from, unsigned to);
717 int ll_writepage(struct page *page);
718 void ll_inode_fill_obdo(struct inode *inode, int cmd, struct obdo *oa);
719 int ll_ap_completion(void *data, int cmd, struct obdo *oa, int rc);
720 int llap_shrink_cache(struct ll_sb_info *sbi, int shrink_fraction);
721 extern struct cache_definition ll_cache_definition;
722 void ll_removepage(struct page *page);
723 int ll_readpage(struct file *file, struct page *page);
724 struct ll_async_page *llap_cast_private(struct page *page);
725 void ll_readahead_init(struct inode *inode, struct ll_readahead_state *ras);
726 void ll_ra_accounting(struct ll_async_page *llap,struct address_space *mapping);
727 void ll_truncate(struct inode *inode);
728 int ll_file_punch(struct inode *, loff_t, int);
729 ssize_t ll_file_lockless_io(struct file *, const struct iovec *,
730 unsigned long, loff_t *, int, ssize_t);
731 ssize_t ll_direct_IO(int rw, struct file *file,const struct iovec *iov,
732 loff_t file_offset, unsigned long nr_segs, int locked);
733 void ll_clear_file_contended(struct inode*);
734 int ll_sync_page_range(struct inode *, struct address_space *, loff_t, size_t);
737 extern struct file_operations ll_file_operations;
738 extern struct file_operations ll_file_operations_flock;
739 extern struct file_operations ll_file_operations_noflock;
740 extern struct inode_operations ll_file_inode_operations;
741 extern int ll_inode_revalidate_it(struct dentry *, struct lookup_intent *);
742 extern int ll_have_md_lock(struct inode *inode, __u64 bits);
743 int ll_region_mapped(unsigned long addr, size_t count);
744 int ll_extent_lock(struct ll_file_data *, struct inode *,
745 struct lov_stripe_md *, int mode, ldlm_policy_data_t *,
746 struct lustre_handle *, int ast_flags);
747 int ll_extent_unlock(struct ll_file_data *, struct inode *,
748 struct lov_stripe_md *, int mode, struct lustre_handle *);
749 int __ll_inode_revalidate_it(struct dentry *, struct lookup_intent *, __u64 bits);
750 int ll_revalidate_nd(struct dentry *dentry, struct nameidata *nd);
751 int ll_file_open(struct inode *inode, struct file *file);
752 int ll_file_release(struct inode *inode, struct file *file);
753 int ll_lsm_getattr(struct obd_export *, struct lov_stripe_md *, struct obdo *);
754 int ll_glimpse_ioctl(struct ll_sb_info *sbi,
755 struct lov_stripe_md *lsm, lstat_t *st);
756 int ll_glimpse_size(struct inode *inode, int ast_flags);
757 int ll_local_open(struct file *file,
758 struct lookup_intent *it, struct ll_file_data *fd,
759 struct obd_client_handle *och);
760 int ll_release_openhandle(struct dentry *, struct lookup_intent *);
761 int ll_mdc_close(struct obd_export *mdc_exp, struct inode *inode,
763 int ll_mdc_real_close(struct inode *inode, int flags);
764 extern void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid, struct file
765 *file, size_t count, int rw);
766 int ll_getattr_it(struct vfsmount *mnt, struct dentry *de,
767 struct lookup_intent *it, struct kstat *stat);
768 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat);
769 struct ll_file_data *ll_file_data_get(void);
770 #ifndef HAVE_INODE_PERMISION_2ARGS
771 int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd);
773 int ll_inode_permission(struct inode *inode, int mask);
775 int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
776 int flags, struct lov_user_md *lum,
778 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
779 struct lov_mds_md **lmm, int *lmm_size,
780 struct ptlrpc_request **request);
781 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
783 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmm,
784 int *lmm_size, struct ptlrpc_request **request);
785 int ll_fsync(struct file *file, struct dentry *dentry, int data);
786 int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
792 * protect race ll_find_aliases vs ll_revalidate_it vs ll_unhash_aliases
794 extern spinlock_t ll_lookup_lock;
795 extern struct dentry_operations ll_d_ops;
796 void ll_intent_drop_lock(struct lookup_intent *);
797 void ll_intent_release(struct lookup_intent *);
798 extern void ll_set_dd(struct dentry *de);
799 int ll_drop_dentry(struct dentry *dentry);
800 void ll_unhash_aliases(struct inode *);
801 void ll_frob_intent(struct lookup_intent **itp, struct lookup_intent *deft);
802 void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry);
803 int ll_dcompare(struct dentry *parent, struct qstr *d_name, struct qstr *name);
804 int revalidate_it_finish(struct ptlrpc_request *request, int offset,
805 struct lookup_intent *it, struct dentry *de);
807 /* llite/llite_lib.c */
808 extern struct super_operations lustre_super_operations;
810 char *ll_read_opt(const char *opt, char *data);
811 void ll_lli_init(struct ll_inode_info *lli);
812 int ll_fill_super(struct super_block *sb);
813 void ll_put_super(struct super_block *sb);
814 void ll_kill_super(struct super_block *sb);
815 int ll_shrink_cache(int nr_to_scan, SHRINKER_MASK_T gfp_mask);
816 struct inode *ll_inode_from_lock(struct ldlm_lock *lock);
817 void ll_clear_inode(struct inode *inode);
818 int ll_setattr_raw(struct inode *inode, struct iattr *attr);
819 int ll_setattr(struct dentry *de, struct iattr *attr);
820 #ifndef HAVE_STATFS_DENTRY_PARAM
821 int ll_statfs(struct super_block *sb, struct kstatfs *sfs);
823 int ll_statfs(struct dentry *de, struct kstatfs *sfs);
825 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
826 __u64 max_age, __u32 flags);
827 void ll_update_inode(struct inode *inode, struct lustre_md *md);
828 void ll_read_inode2(struct inode *inode, void *opaque);
829 int ll_iocontrol(struct inode *inode, struct file *file,
830 unsigned int cmd, unsigned long arg);
831 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
832 void ll_umount_begin(struct vfsmount *vfsmnt, int flags);
834 void ll_umount_begin(struct super_block *sb);
836 int ll_remount_fs(struct super_block *sb, int *flags, char *data);
837 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs);
838 int ll_prep_inode(struct obd_export *exp, struct inode **inode,
839 struct ptlrpc_request *req, int offset, struct super_block *);
840 void lustre_dump_dentry(struct dentry *, int recur);
841 void lustre_dump_inode(struct inode *);
842 struct ll_async_page *llite_pglist_next_llap(struct list_head *head,
843 struct list_head *list);
844 int ll_obd_statfs(struct inode *inode, void *arg);
845 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *max_mdsize);
846 int ll_process_config(struct lustre_cfg *lcfg);
848 /* llite/llite_nfs.c */
849 extern struct export_operations lustre_export_operations;
850 __u32 get_uuid2int(const char *name, int len);
852 /* llite/special.c */
853 extern struct inode_operations ll_special_inode_operations;
854 extern struct file_operations ll_special_chr_inode_fops;
855 extern struct file_operations ll_special_chr_file_fops;
856 extern struct file_operations ll_special_blk_inode_fops;
857 extern struct file_operations ll_special_fifo_inode_fops;
858 extern struct file_operations ll_special_fifo_file_fops;
859 extern struct file_operations ll_special_sock_inode_fops;
861 /* llite/symlink.c */
862 extern struct inode_operations ll_fast_symlink_inode_operations;
864 /* llite/llite_close.c */
865 struct ll_close_queue {
867 struct list_head lcq_list;
868 wait_queue_head_t lcq_waitq;
869 struct completion lcq_comp;
872 #ifdef HAVE_CLOSE_THREAD
873 void llap_write_pending(struct inode *inode, struct ll_async_page *llap);
874 void llap_write_complete(struct inode *inode, struct ll_async_page *llap);
875 void ll_open_complete(struct inode *inode);
876 int ll_is_inode_dirty(struct inode *inode);
877 void ll_try_done_writing(struct inode *inode);
878 void ll_queue_done_writing(struct inode *inode);
880 static inline void llap_write_pending(struct inode *inode,
881 struct ll_async_page *llap) { return; };
882 static inline void llap_write_complete(struct inode *inode,
883 struct ll_async_page *llap) { return; };
884 static inline void ll_open_complete(struct inode *inode) { return; };
885 static inline int ll_is_inode_dirty(struct inode *inode) { return 0; };
886 static inline void ll_try_done_writing(struct inode *inode) { return; };
887 static inline void ll_queue_done_writing(struct inode *inode) { return; };
888 //static inline void ll_close_thread_shutdown(struct ll_close_queue *lcq) { return; };
889 //static inline int ll_close_thread_start(struct ll_close_queue **lcq_ret) { return 0; };
891 void ll_close_thread_shutdown(struct ll_close_queue *lcq);
892 int ll_close_thread_start(struct ll_close_queue **lcq_ret);
894 /* llite/llite_mmap.c */
895 typedef struct rb_root rb_root_t;
896 typedef struct rb_node rb_node_t;
898 struct ll_lock_tree_node;
899 struct ll_lock_tree {
901 struct list_head lt_locked_list;
902 struct ll_file_data *lt_fd;
905 int ll_teardown_mmaps(struct address_space *mapping, __u64 first, __u64 last);
906 int ll_file_mmap(struct file * file, struct vm_area_struct * vma);
907 struct ll_lock_tree_node * ll_node_from_inode(struct inode *inode, __u64 start,
908 __u64 end, ldlm_mode_t mode);
909 int ll_tree_lock(struct ll_lock_tree *tree,
910 struct ll_lock_tree_node *first_node,
911 const char *buf, size_t count, int ast_flags);
912 int ll_tree_lock_iov(struct ll_lock_tree *tree,
913 struct ll_lock_tree_node *first_node,
914 const struct iovec *iov, unsigned long nr_segs,
916 int ll_tree_unlock(struct ll_lock_tree *tree);
919 LL_LOCK_STYLE_NOLOCK = 0,
920 LL_LOCK_STYLE_FASTLOCK = 1,
921 LL_LOCK_STYLE_TREELOCK = 2
924 struct ll_thread_data {
927 struct list_head *tree_list;
929 struct ll_lock_tree tree;
930 struct lustre_handle lockh;
933 struct ll_thread_data *ll_td_get(void);
934 void ll_td_set(struct ll_thread_data *ltd);
935 struct lustre_handle *ltd2lockh(struct ll_thread_data *ltd, __u64 start,
938 #define ll_s2sbi(sb) (s2lsi(sb)->lsi_llsbi)
940 static inline __u64 ll_ts2u64(struct timespec *time)
942 __u64 t = time->tv_sec;
946 /* don't need an addref as the sb_info should be holding one */
947 static inline struct obd_export *ll_s2obdexp(struct super_block *sb)
949 return ll_s2sbi(sb)->ll_osc_exp;
952 /* don't need an addref as the sb_info should be holding one */
953 static inline struct obd_export *ll_s2mdcexp(struct super_block *sb)
955 return ll_s2sbi(sb)->ll_mdc_exp;
958 static inline struct client_obd *sbi2mdc(struct ll_sb_info *sbi)
960 struct obd_device *obd = sbi->ll_mdc_exp->exp_obd;
966 // FIXME: replace the name of this with LL_SB to conform to kernel stuff
967 static inline struct ll_sb_info *ll_i2sbi(struct inode *inode)
969 return ll_s2sbi(inode->i_sb);
972 static inline struct obd_export *ll_i2obdexp(struct inode *inode)
974 return ll_s2obdexp(inode->i_sb);
977 static inline struct obd_export *ll_i2mdcexp(struct inode *inode)
979 return ll_s2mdcexp(inode->i_sb);
982 /** get lu_fid from inode. */
983 static inline struct lu_fid *ll_inode_lu_fid(struct inode *inode)
985 return &ll_i2info(inode)->lli_fid.f20;
988 /** get ll_fid from inode. */
989 static inline struct ll_fid *ll_inode_ll_fid(struct inode *inode)
991 return &ll_i2info(inode)->lli_fid.f16;
994 static inline void ll_inode2fid(struct ll_fid *fid, struct inode *inode)
996 *fid = *ll_inode_ll_fid(inode);
999 static inline int ll_mds_max_easize(struct super_block *sb)
1001 return sbi2mdc(ll_s2sbi(sb))->cl_max_mds_easize;
1004 static inline __u64 ll_file_maxbytes(struct inode *inode)
1006 return ll_i2info(inode)->lli_maxbytes;
1010 int ll_setxattr(struct dentry *dentry, const char *name,
1011 const void *value, size_t size, int flags);
1012 ssize_t ll_getxattr(struct dentry *dentry, const char *name,
1013 void *buffer, size_t size);
1014 ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size);
1015 int ll_removexattr(struct dentry *dentry, const char *name);
1019 #define LL_SA_RPC_MIN 2
1020 #define LL_SA_RPC_DEF 32
1021 #define LL_SA_RPC_MAX 8192
1023 /* per inode struct, for dir only */
1024 struct ll_statahead_info {
1025 struct inode *sai_inode;
1026 unsigned int sai_generation; /* generation for statahead */
1027 atomic_t sai_refcount; /* when access this struct, hold
1029 unsigned int sai_sent; /* stat requests sent count */
1030 unsigned int sai_replied; /* stat requests which received
1032 unsigned int sai_max; /* max ahead of lookup */
1033 unsigned int sai_index; /* index of statahead entry */
1034 unsigned int sai_index_next; /* index for the next statahead
1035 * entry to be stated */
1036 unsigned int sai_hit; /* hit count */
1037 unsigned int sai_miss; /* miss count:
1038 * for "ls -al" case, it includes
1039 * hidden dentry miss;
1040 * for "ls -l" case, it does not
1041 * include hidden dentry miss.
1042 * "sai_miss_hidden" is used for
1045 unsigned int sai_consecutive_miss; /* consecutive miss */
1046 unsigned int sai_miss_hidden;/* "ls -al", but first dentry
1047 * is not a hidden one */
1048 unsigned int sai_skip_hidden;/* skipped hidden dentry count */
1049 unsigned int sai_ls_all:1; /* "ls -al", do stat-ahead for
1051 cfs_waitq_t sai_waitq; /* stat-ahead wait queue */
1052 struct ptlrpc_thread sai_thread; /* stat-ahead thread */
1053 struct list_head sai_entries_sent; /* entries sent out */
1054 struct list_head sai_entries_received; /* entries returned */
1055 struct list_head sai_entries_stated; /* entries stated */
1058 int do_statahead_enter(struct inode *dir, struct dentry **dentry, int lookup);
1059 void ll_statahead_exit(struct inode *dir, struct dentry *dentry, int result);
1060 void ll_stop_statahead(struct inode *inode, void *key);
1063 void ll_statahead_mark(struct inode *dir, struct dentry *dentry)
1065 struct ll_inode_info *lli;
1066 struct ll_dentry_data *ldd = ll_d2d(dentry);
1068 /* dentry has been move to other directory, no need mark */
1069 if (unlikely(dir != dentry->d_parent->d_inode))
1072 lli = ll_i2info(dir);
1073 /* not the same process, don't mark */
1074 if (lli->lli_opendir_pid != cfs_curproc_pid())
1077 spin_lock(&lli->lli_lock);
1078 if (likely(lli->lli_sai != NULL && ldd != NULL))
1079 ldd->lld_sa_generation = lli->lli_sai->sai_generation;
1080 spin_unlock(&lli->lli_lock);
1084 int ll_statahead_enter(struct inode *dir, struct dentry **dentryp, int lookup)
1086 struct ll_inode_info *lli;
1087 struct ll_sb_info *sbi;
1088 struct ll_dentry_data *ldd = ll_d2d(*dentryp);
1090 if (unlikely(dir == NULL))
1093 sbi = ll_i2sbi(dir);
1094 /* temporarily disable dir stat ahead in interoperability mode */
1095 if (sbi->ll_mdc_exp->exp_connect_flags & OBD_CONNECT_FID)
1098 if (sbi->ll_sa_max == 0)
1101 lli = ll_i2info(dir);
1102 /* not the same process, don't statahead */
1103 if (lli->lli_opendir_pid != cfs_curproc_pid())
1107 * When "ls" a dentry, the system trigger more than once "revalidate" or
1108 * "lookup", for "getattr", for "getxattr", and maybe for others.
1109 * Under patchless client mode, the operation intent is not accurate,
1110 * it maybe misguide the statahead thread. For example:
1111 * The "revalidate" call for "getattr" and "getxattr" of a dentry maybe
1112 * have the same operation intent -- "IT_GETATTR".
1113 * In fact, one dentry should has only one chance to interact with the
1114 * statahead thread, otherwise the statahead windows will be confused.
1115 * The solution is as following:
1116 * Assign "lld_sa_generation" with "sai_generation" when a dentry
1117 * "IT_GETATTR" for the first time, and the subsequent "IT_GETATTR"
1118 * will bypass interacting with statahead thread for checking:
1119 * "lld_sa_generation == lli_sai->sai_generation"
1121 if (ldd && lli->lli_sai &&
1122 ldd->lld_sa_generation == lli->lli_sai->sai_generation)
1125 return do_statahead_enter(dir, dentryp, lookup);
1128 static void inline ll_dops_init(struct dentry *de, int block)
1130 struct ll_dentry_data *lld = ll_d2d(de);
1132 if (lld == NULL && block != 0) {
1138 lld->lld_sa_generation = 0;
1140 de->d_op = &ll_d_ops;
1143 /* llite ioctl register support rountine */
1150 #define LLIOC_MAX_CMD 256
1153 * Rules to write a callback function:
1156 * @magic: Dynamic ioctl call routine will feed this vaule with the pointer
1157 * returned to ll_iocontrol_register. Callback functions should use this
1158 * data to check the potential collasion of ioctl cmd. If collasion is
1159 * found, callback function should return LLIOC_CONT.
1160 * @rcp: The result of ioctl command.
1163 * If @magic matches the pointer returned by ll_iocontrol_data, the
1164 * callback should return LLIOC_STOP; return LLIOC_STOP otherwise.
1166 typedef enum llioc_iter (*llioc_callback_t)(struct inode *inode,
1167 struct file *file, unsigned int cmd, unsigned long arg,
1168 void *magic, int *rcp);
1170 enum llioc_iter ll_iocontrol_call(struct inode *inode, struct file *file,
1171 unsigned int cmd, unsigned long arg, int *rcp);
1173 /* export functions */
1174 /* Register ioctl block dynamatically for a regular file.
1176 * @cmd: the array of ioctl command set
1177 * @count: number of commands in the @cmd
1178 * @cb: callback function, it will be called if an ioctl command is found to
1179 * belong to the command list @cmd.
1182 * A magic pointer will be returned if success;
1183 * otherwise, NULL will be returned.
1185 void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd);
1186 void ll_iocontrol_unregister(void *magic);
1188 ino_t ll_fid_build_ino(struct ll_sb_info *sbi,
1189 struct ll_fid *fid);
1190 __u32 ll_fid_build_gen(struct ll_sb_info *sbi,
1191 struct ll_fid *fid);
1195 #endif /* LLITE_INTERNAL_H */