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.
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 #define LL_IT2STR(it) ((it) ? ldlm_it2str((it)->it_op) : "0")
74 #define LUSTRE_FPRIVATE(file) ((file)->private_data)
76 #ifdef HAVE_VFS_INTENT_PATCHES
77 static inline struct lookup_intent *ll_nd2it(struct nameidata *nd)
84 * Directory entries are currently in the same format as ext2/ext3, but will
85 * be changed in the future to accomodate FIDs
87 #define LL_DIR_NAME_LEN (255)
88 #define LL_DIR_PAD (4)
91 /* number of inode, referenced by this entry */
93 /* total record length, multiple of LL_DIR_PAD */
97 /* file type: regular, directory, device, etc. */
99 /* name. NOT NUL-terminated */
100 char lde_name[LL_DIR_NAME_LEN];
103 struct ll_dentry_data {
106 struct obd_client_handle lld_cwd_och;
107 struct obd_client_handle lld_mnt_och;
108 #ifndef HAVE_VFS_INTENT_PATCHES
109 struct lookup_intent *lld_it;
111 unsigned int lld_sa_generation;
114 #define ll_d2d(de) ((struct ll_dentry_data*)((de)->d_fsdata))
116 extern struct file_operations ll_pgcache_seq_fops;
118 #define LLI_INODE_MAGIC 0x111d0de5
119 #define LLI_INODE_DEAD 0xdeadd00d
120 #define LLI_F_HAVE_OST_SIZE_LOCK 0
121 #define LLI_F_HAVE_MDS_SIZE_LOCK 1
122 #define LLI_F_CONTENDED 2
123 #define LLI_F_SRVLOCK 3
125 struct ll_inode_info {
127 struct semaphore lli_size_sem; /* protect open and change size */
128 void *lli_size_sem_owner;
129 struct semaphore lli_write_sem;
130 struct lov_stripe_md *lli_smd;
131 char *lli_symlink_name;
134 unsigned long lli_flags;
135 cfs_time_t lli_contention_time;
137 /* this lock protects s_d_w and p_w_ll and mmap_cnt */
139 #ifdef HAVE_CLOSE_THREAD
140 struct list_head lli_pending_write_llaps;
141 struct list_head lli_close_item;
142 int lli_send_done_writing;
144 atomic_t lli_mmap_cnt;
146 /* for writepage() only to communicate to fsync */
149 struct posix_acl *lli_posix_acl;
151 struct list_head lli_dead_list;
153 struct semaphore lli_och_sem; /* Protects access to och pointers
154 and their usage counters */
155 /* We need all three because every inode may be opened in different
157 struct obd_client_handle *lli_mds_read_och;
158 __u64 lli_open_fd_read_count;
159 struct obd_client_handle *lli_mds_write_och;
160 __u64 lli_open_fd_write_count;
161 struct obd_client_handle *lli_mds_exec_och;
162 __u64 lli_open_fd_exec_count;
164 /** fid of this object. */
170 /* metadata stat-ahead */
172 * "opendir_pid" is the token when lookup/revalid -- I am the owner of
175 pid_t lli_opendir_pid;
177 * since parent-child threads can share the same @file struct,
178 * "opendir_key" is the token when dir close for case of parent exit
179 * before child -- it is me should cleanup the dir readahead. */
180 void *lli_opendir_key;
181 struct ll_statahead_info *lli_sai;
182 /* the most recent attributes from mds, it is used for timestampts
184 struct ost_lvb lli_lvb;
185 struct inode lli_vfs_inode;
189 * Locking to guarantee consistency of non-atomic updates to long long i_size,
190 * consistency between file size and KMS, and consistency within
191 * ->lli_smd->lsm_oinfo[]'s.
193 * Implemented by ->lli_size_sem and ->lsm_sem, nested in that order.
196 void ll_inode_size_lock(struct inode *inode, int lock_lsm);
197 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm);
199 // FIXME: replace the name of this with LL_I to conform to kernel stuff
200 // static inline struct ll_inode_info *LL_I(struct inode *inode)
201 static inline struct ll_inode_info *ll_i2info(struct inode *inode)
203 return container_of(inode, struct ll_inode_info, lli_vfs_inode);
206 /* default to about 40meg of readahead on a given system. That much tied
207 * up in 512k readahead requests serviced at 40ms each is about 1GB/s. */
208 #define SBI_DEFAULT_READAHEAD_MAX (40UL << (20 - CFS_PAGE_SHIFT))
210 /* default to read-ahead full files smaller than 2MB on the second read */
211 #define SBI_DEFAULT_READAHEAD_WHOLE_MAX (2UL << (20 - CFS_PAGE_SHIFT))
216 RA_STAT_DISTANT_READPAGE,
217 RA_STAT_MISS_IN_WINDOW,
218 RA_STAT_FAILED_GRAB_PAGE,
219 RA_STAT_FAILED_MATCH,
224 RA_STAT_MAX_IN_FLIGHT,
225 RA_STAT_WRONG_GRAB_PAGE,
229 #define LL_RA_STAT _NR_RA_STAT
230 #define LL_RA_STAT_STRINGS { \
231 [RA_STAT_HIT] = "hits", \
232 [RA_STAT_MISS] = "misses", \
233 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive", \
234 [RA_STAT_MISS_IN_WINDOW] = "miss inside window", \
235 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page", \
236 [RA_STAT_FAILED_MATCH] = "failed lock match", \
237 [RA_STAT_DISCARDED] = "read but discarded", \
238 [RA_STAT_ZERO_LEN] = "zero length file", \
239 [RA_STAT_ZERO_WINDOW] = "zero size window", \
240 [RA_STAT_EOF] = "read-ahead to EOF", \
241 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue", \
242 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",\
246 atomic_t ra_cur_pages;
247 unsigned long ra_max_pages;
248 unsigned long ra_max_read_ahead_whole_pages;
251 /* LL_HIST_MAX=32 causes an overflow */
252 #define LL_HIST_MAX 28
253 #define LL_HIST_START 12 /* buckets start at 2^12 = 4k */
254 #define LL_PROCESS_HIST_MAX 10
255 struct per_process_info {
257 struct obd_histogram pp_r_hist;
258 struct obd_histogram pp_w_hist;
261 /* pp_extents[LL_PROCESS_HIST_MAX] will hold the combined process info */
262 struct ll_rw_extents_info {
263 struct per_process_info pp_extents[LL_PROCESS_HIST_MAX + 1];
266 #define LL_OFFSET_HIST_MAX 100
267 struct ll_rw_process_info {
270 loff_t rw_range_start;
272 loff_t rw_last_file_pos;
274 size_t rw_smallest_extent;
275 size_t rw_largest_extent;
276 struct file *rw_last_file;
280 enum stats_track_type {
281 STATS_TRACK_ALL = 0, /* track all processes */
282 STATS_TRACK_PID, /* track process with this pid */
283 STATS_TRACK_PPID, /* track processes with this ppid */
284 STATS_TRACK_GID, /* track processes with this gid */
288 /* flags for sbi->ll_flags */
289 #define LL_SBI_NOLCK 0x01 /* DLM locking disabled (directio-only) */
290 #define LL_SBI_DATA_CHECKSUM 0x02 /* checksum each page on the wire */
291 #define LL_SBI_FLOCK 0x04
292 #define LL_SBI_USER_XATTR 0x08 /* support user xattr */
293 #define LL_SBI_ACL 0x10 /* support ACL */
294 #define LL_SBI_JOIN 0x20 /* support JOIN */
295 #define LL_SBI_LOCALFLOCK 0x40 /* Local flocks support by kernel */
296 #define LL_SBI_LRU_RESIZE 0x80 /* support lru resize */
297 #define LL_SBI_LLITE_CHECKSUM 0x100 /* checksum each page in memory */
299 /* default value for ll_sb_info->contention_time */
300 #define SBI_DEFAULT_CONTENTION_SECONDS 60
301 /* default value for lockless_truncate_enable */
302 #define SBI_DEFAULT_LOCKLESS_TRUNCATE_ENABLE 1
304 /* percpu data structure for lustre lru page list */
305 struct ll_pglist_data {
306 spinlock_t llpd_lock; /* lock to protect llpg_list */
307 struct list_head llpd_list; /* all pages (llap_pglist_item) */
308 unsigned long llpd_gen; /* generation # of this list */
309 unsigned long llpd_count; /* How many pages in this list */
310 atomic_t llpd_sample_count;
311 unsigned long llpd_reblnc_count;
312 /* the pages in this list shouldn't be over this number */
313 unsigned long llpd_budget;
315 /* which page the pglist data is in */
316 struct page *llpd_page;
319 unsigned long llpd_hit;
320 unsigned long llpd_miss;
321 unsigned long llpd_cross;
325 struct list_head ll_list;
326 /* this protects pglist(only ll_async_page_max) and ra_info.
327 * It isn't safe to grab from interrupt contexts. */
329 spinlock_t ll_pp_extent_lock; /* Lock for pp_extent entries */
330 spinlock_t ll_process_lock; /* Lock for ll_rw_process_info */
331 struct obd_uuid ll_sb_uuid;
332 struct obd_export *ll_mdc_exp;
333 struct obd_export *ll_osc_exp;
334 struct proc_dir_entry *ll_proc_root;
335 obd_id ll_rootino; /* number of root inode */
338 struct list_head ll_conn_chain; /* per-conn chain of SBs */
339 struct lustre_client_ocd ll_lco;
341 struct list_head ll_orphan_dentry_list; /*please don't ask -p*/
342 struct ll_close_queue *ll_lcq;
344 struct lprocfs_stats *ll_stats; /* lprocfs stats counter */
346 /* reblnc lock protects llpd_budget */
347 spinlock_t ll_async_page_reblnc_lock;
348 unsigned long ll_async_page_reblnc_count;
349 unsigned long ll_async_page_sample_max;
350 /* I defined this array here rather than in ll_pglist_data
351 * because it is always accessed by only one cpu. -jay */
352 unsigned long *ll_async_page_sample;
353 unsigned long ll_async_page_max;
354 unsigned long ll_async_page_clock_hand;
355 lcounter_t ll_async_page_count;
356 struct ll_pglist_data **ll_pglist;
358 struct lprocfs_stats *ll_ra_stats;
360 unsigned ll_contention_time; /* seconds */
361 unsigned ll_lockless_truncate_enable; /* true/false */
363 struct ll_ra_info ll_ra_info;
364 unsigned int ll_namelen;
365 struct file_operations *ll_fop;
367 #ifdef HAVE_EXPORT___IGET
368 struct list_head ll_deathrow; /* inodes to be destroyed (b1443) */
369 spinlock_t ll_deathrow_lock;
371 /* =0 - hold lock over whole read/write
372 * >0 - max. chunk to be read/written w/o lock re-acquiring */
373 unsigned long ll_max_rw_chunk;
376 struct ll_rw_extents_info ll_rw_extents_info;
377 int ll_extent_process_count;
378 struct ll_rw_process_info ll_rw_process_info[LL_PROCESS_HIST_MAX];
379 unsigned int ll_offset_process_count;
380 struct ll_rw_process_info ll_rw_offset_info[LL_OFFSET_HIST_MAX];
381 unsigned int ll_rw_offset_entry_count;
382 enum stats_track_type ll_stats_track_type;
383 int ll_stats_track_id;
385 dev_t ll_sdev_orig; /* save s_dev before assign for
388 /* metadata stat-ahead */
389 unsigned int ll_sa_max; /* max statahead RPCs */
390 unsigned int ll_sa_wrong; /* statahead thread stopped for
392 unsigned int ll_sa_total; /* statahead thread started
394 unsigned long long ll_sa_blocked; /* ls count waiting for
396 unsigned long long ll_sa_cached; /* ls count got in cache */
397 unsigned long long ll_sa_hit; /* hit count */
398 unsigned long long ll_sa_miss; /* miss count */
401 #define LL_DEFAULT_MAX_RW_CHUNK (32 * 1024 * 1024)
403 #define LL_PGLIST_DATA_CPU(sbi, cpu) ((sbi)->ll_pglist[cpu])
404 #define LL_PGLIST_DATA(sbi) LL_PGLIST_DATA_CPU(sbi, smp_processor_id())
406 static inline struct ll_pglist_data *ll_pglist_cpu_lock(
407 struct ll_sb_info *sbi,
410 spin_lock(&sbi->ll_pglist[cpu]->llpd_lock);
411 return LL_PGLIST_DATA_CPU(sbi, cpu);
414 static inline void ll_pglist_cpu_unlock(struct ll_sb_info *sbi, int cpu)
416 spin_unlock(&sbi->ll_pglist[cpu]->llpd_lock);
419 static inline struct ll_pglist_data *ll_pglist_double_lock(
420 struct ll_sb_info *sbi,
421 int cpu, struct ll_pglist_data **pd_cpu)
423 int current_cpu = get_cpu();
425 if (cpu == current_cpu) {
426 ll_pglist_cpu_lock(sbi, cpu);
427 } else if (current_cpu < cpu) {
428 ll_pglist_cpu_lock(sbi, current_cpu);
429 ll_pglist_cpu_lock(sbi, cpu);
431 ll_pglist_cpu_lock(sbi, cpu);
432 ll_pglist_cpu_lock(sbi, current_cpu);
436 *pd_cpu = LL_PGLIST_DATA_CPU(sbi, cpu);
438 return LL_PGLIST_DATA(sbi);
441 static inline void ll_pglist_double_unlock(struct ll_sb_info *sbi, int cpu)
443 int current_cpu = smp_processor_id();
444 if (cpu == current_cpu) {
445 ll_pglist_cpu_unlock(sbi, cpu);
447 ll_pglist_cpu_unlock(sbi, cpu);
448 ll_pglist_cpu_unlock(sbi, current_cpu);
453 static inline struct ll_pglist_data *ll_pglist_lock(struct ll_sb_info *sbi)
455 ll_pglist_cpu_lock(sbi, get_cpu());
456 return LL_PGLIST_DATA(sbi);
459 static inline void ll_pglist_unlock(struct ll_sb_info *sbi)
461 ll_pglist_cpu_unlock(sbi, smp_processor_id());
468 struct task_struct *lrr_reader;
469 struct list_head lrr_linkage;
473 * per file-descriptor read-ahead data.
475 struct ll_readahead_state {
478 * index of the last page that read(2) needed and that wasn't in the
479 * cache. Used by ras_update() to detect seeks.
481 * XXX nikita: if access seeks into cached region, Lustre doesn't see
484 unsigned long ras_last_readpage;
486 * number of pages read after last read-ahead window reset. As window
487 * is reset on each seek, this is effectively a number of consecutive
488 * accesses. Maybe ->ras_accessed_in_window is better name.
490 * XXX nikita: window is also reset (by ras_update()) when Lustre
491 * believes that memory pressure evicts read-ahead pages. In that
492 * case, it probably doesn't make sense to expand window to
493 * PTLRPC_MAX_BRW_PAGES on the third access.
495 unsigned long ras_consecutive_pages;
497 * number of read requests after the last read-ahead window reset
498 * As window is reset on each seek, this is effectively the number
499 * on consecutive read request and is used to trigger read-ahead.
501 unsigned long ras_consecutive_requests;
503 * Parameters of current read-ahead window. Handled by
504 * ras_update(). On the initial access to the file or after a seek,
505 * window is reset to 0. After 3 consecutive accesses, window is
506 * expanded to PTLRPC_MAX_BRW_PAGES. Afterwards, window is enlarged by
507 * PTLRPC_MAX_BRW_PAGES chunks up to ->ra_max_pages.
509 unsigned long ras_window_start, ras_window_len;
511 * Where next read-ahead should start at. This lies within read-ahead
512 * window. Read-ahead window is read in pieces rather than at once
513 * because: 1. lustre limits total number of pages under read-ahead by
514 * ->ra_max_pages (see ll_ra_count_get()), 2. client cannot read pages
515 * not covered by DLM lock.
517 unsigned long ras_next_readahead;
519 * Total number of ll_file_read requests issued, reads originating
520 * due to mmap are not counted in this total. This value is used to
521 * trigger full file read-ahead after multiple reads to a small file.
523 unsigned long ras_requests;
525 * Page index with respect to the current request, these value
526 * will not be accurate when dealing with reads issued via mmap.
528 unsigned long ras_request_index;
530 * list of struct ll_ra_read's one per read(2) call current in
531 * progress against this file descriptor. Used by read-ahead code,
532 * protected by ->ras_lock.
534 struct list_head ras_read_beads;
536 * The following 3 items are used for detecting the stride I/O
538 * In stride I/O mode,
539 * ...............|-----data-----|****gap*****|--------|******|....
540 * offset |-stride_pages-|-stride_gap-|
541 * ras_stride_offset = offset;
542 * ras_stride_length = stride_pages + stride_gap;
543 * ras_stride_pages = stride_pages;
544 * Note: all these three items are counted by pages.
546 unsigned long ras_stride_length;
547 unsigned long ras_stride_pages;
548 pgoff_t ras_stride_offset;
550 * number of consecutive stride request count, and it is similar as
551 * ras_consecutive_requests, but used for stride I/O mode.
552 * Note: only more than 2 consecutive stride request are detected,
553 * stride read-ahead will be enable
555 unsigned long ras_consecutive_stride_requests;
558 extern cfs_mem_cache_t *ll_file_data_slab;
559 struct lustre_handle;
560 struct ll_file_data {
561 struct ll_readahead_state fd_ras;
563 struct lustre_handle fd_cwlockh;
564 unsigned long fd_gid;
568 struct lov_stripe_md;
570 extern spinlock_t inode_lock;
572 extern struct proc_dir_entry *proc_lustre_fs_root;
574 static inline struct inode *ll_info2i(struct ll_inode_info *lli)
576 return &lli->lli_vfs_inode;
580 struct inode *icbd_parent;
581 struct dentry **icbd_childp;
585 void ll_i2gids(__u32 *suppgids, struct inode *i1,struct inode *i2);
587 #define LLAP_MAGIC 98764321
589 extern cfs_mem_cache_t *ll_async_page_slab;
590 extern size_t ll_async_page_slab_size;
591 struct ll_async_page {
593 /* only trust these if the page lock is providing exclusion */
594 unsigned int llap_write_queued:1,
595 llap_defer_uptodate:1,
600 llap_lockless_io_page:1,
602 unsigned int llap_pglist_cpu:16;
604 struct page *llap_page;
605 struct list_head llap_pending_write;
606 struct list_head llap_pglist_item;
607 /* checksum for paranoid I/O debugging */
612 * enumeration of llap_from_page() call-sites. Used to export statistics in
613 * /proc/fs/lustre/llite/fsN/dump_page_cache.
616 LLAP_ORIGIN_UNKNOWN = 0,
617 LLAP_ORIGIN_READPAGE,
618 LLAP_ORIGIN_READAHEAD,
619 LLAP_ORIGIN_COMMIT_WRITE,
620 LLAP_ORIGIN_WRITEPAGE,
621 LLAP_ORIGIN_REMOVEPAGE,
622 LLAP_ORIGIN_LOCKLESS_IO,
625 extern char *llap_origins[];
627 #ifdef HAVE_REGISTER_CACHE
628 #include <linux/cache_def.h>
629 #define ll_register_cache(cache) register_cache(cache)
630 #define ll_unregister_cache(cache) unregister_cache(cache)
631 #elif defined(HAVE_SHRINKER_CACHE)
632 struct cache_definition {
635 struct shrinker *shrinker;
638 #define ll_register_cache(cache) do { \
639 struct cache_definition *c = (cache); \
640 c->shrinker = set_shrinker(DEFAULT_SEEKS, c->shrink); \
643 #define ll_unregister_cache(cache) do { \
644 remove_shrinker((cache)->shrinker); \
645 (cache)->shrinker = NULL; \
648 #define ll_register_cache(cache) do {} while (0)
649 #define ll_unregister_cache(cache) do {} while (0)
652 void ll_ra_read_in(struct file *f, struct ll_ra_read *rar);
653 void ll_ra_read_ex(struct file *f, struct ll_ra_read *rar);
654 struct ll_ra_read *ll_ra_read_get(struct file *f);
656 /* llite/lproc_llite.c */
658 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
659 struct super_block *sb, char *osc, char *mdc);
660 void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi);
661 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count);
662 void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars);
664 static inline int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
665 struct super_block *sb, char *osc, char *mdc){return 0;}
666 static inline void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi) {}
667 static void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count) {}
668 static void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars)
670 memset(lvars, 0, sizeof(*lvars));
676 extern struct file_operations ll_dir_operations;
677 extern struct inode_operations ll_dir_inode_operations;
679 struct page *ll_get_dir_page(struct inode *dir, unsigned long n);
681 static inline unsigned ll_dir_rec_len(unsigned name_len)
683 return (name_len + 8 + LL_DIR_PAD - 1) & ~(LL_DIR_PAD - 1);
686 static inline struct ll_dir_entry *ll_entry_at(void *base, unsigned offset)
688 return (struct ll_dir_entry *)((char *)base + offset);
692 * p is at least 6 bytes before the end of page
694 static inline struct ll_dir_entry *ll_dir_next_entry(struct ll_dir_entry *p)
696 return ll_entry_at(p, le16_to_cpu(p->lde_rec_len));
699 static inline void ll_put_page(struct page *page)
702 page_cache_release(page);
705 static inline unsigned long dir_pages(struct inode *inode)
707 return (inode->i_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
710 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir);
711 struct inode *ll_iget(struct super_block *sb, ino_t hash,
712 struct lustre_md *lic);
713 int ll_mdc_cancel_unused(struct lustre_handle *, struct inode *, int flags,
715 int ll_mdc_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
716 void *data, int flag);
717 int ll_prepare_mdc_op_data(struct mdc_op_data *,
718 struct inode *i1, struct inode *i2,
719 const char *name, int namelen, int mode, void *data);
720 #ifndef HAVE_VFS_INTENT_PATCHES
721 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
724 void ll_pin_extent_cb(void *data);
725 int ll_page_removal_cb(void *data, int discard);
726 int ll_extent_lock_cancel_cb(struct ldlm_lock *lock, struct ldlm_lock_desc *new,
727 void *data, int flag);
728 int lookup_it_finish(struct ptlrpc_request *request, int offset,
729 struct lookup_intent *it, void *data);
730 void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry);
733 int ll_prepare_write(struct file *, struct page *, unsigned from, unsigned to);
734 int ll_commit_write(struct file *, struct page *, unsigned from, unsigned to);
735 int ll_writepage(struct page *page);
736 void ll_inode_fill_obdo(struct inode *inode, int cmd, struct obdo *oa);
737 int ll_ap_completion(void *data, int cmd, struct obdo *oa, int rc);
738 int llap_shrink_cache(struct ll_sb_info *sbi, int shrink_fraction);
739 extern struct cache_definition ll_cache_definition;
740 void ll_removepage(struct page *page);
741 int ll_readpage(struct file *file, struct page *page);
742 struct ll_async_page *llap_cast_private(struct page *page);
743 void ll_readahead_init(struct inode *inode, struct ll_readahead_state *ras);
744 void ll_ra_accounting(struct ll_async_page *llap,struct address_space *mapping);
745 void ll_truncate(struct inode *inode);
746 int ll_file_punch(struct inode *, loff_t, int);
747 ssize_t ll_file_lockless_io(struct file *, const struct iovec *,
748 unsigned long, loff_t *, int, ssize_t);
749 void ll_clear_file_contended(struct inode*);
750 int ll_sync_page_range(struct inode *, struct address_space *, loff_t, size_t);
753 extern struct file_operations ll_file_operations;
754 extern struct file_operations ll_file_operations_flock;
755 extern struct file_operations ll_file_operations_noflock;
756 extern struct inode_operations ll_file_inode_operations;
757 extern int ll_inode_revalidate_it(struct dentry *, struct lookup_intent *);
758 extern int ll_have_md_lock(struct inode *inode, __u64 bits);
759 int ll_region_mapped(unsigned long addr, size_t count);
760 int ll_extent_lock(struct ll_file_data *, struct inode *,
761 struct lov_stripe_md *, int mode, ldlm_policy_data_t *,
762 struct lustre_handle *, int ast_flags);
763 int ll_extent_unlock(struct ll_file_data *, struct inode *,
764 struct lov_stripe_md *, int mode, struct lustre_handle *);
765 int ll_file_open(struct inode *inode, struct file *file);
766 int ll_file_release(struct inode *inode, struct file *file);
767 int ll_lsm_getattr(struct obd_export *, struct lov_stripe_md *, struct obdo *);
768 int ll_glimpse_ioctl(struct ll_sb_info *sbi,
769 struct lov_stripe_md *lsm, lstat_t *st);
770 int ll_glimpse_size(struct inode *inode, int ast_flags);
771 int ll_local_open(struct file *file,
772 struct lookup_intent *it, struct ll_file_data *fd,
773 struct obd_client_handle *och);
774 int ll_release_openhandle(struct dentry *, struct lookup_intent *);
775 int ll_mdc_close(struct obd_export *mdc_exp, struct inode *inode,
777 int ll_mdc_real_close(struct inode *inode, int flags);
778 extern void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid, struct file
779 *file, size_t count, int rw);
780 int ll_getattr_it(struct vfsmount *mnt, struct dentry *de,
781 struct lookup_intent *it, struct kstat *stat);
782 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat);
783 struct ll_file_data *ll_file_data_get(void);
784 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0))
785 int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd);
787 int ll_inode_permission(struct inode *inode, int mask);
789 int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
790 int flags, struct lov_user_md *lum,
792 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
793 struct lov_mds_md **lmm, int *lmm_size,
794 struct ptlrpc_request **request);
795 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
797 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmm,
798 int *lmm_size, struct ptlrpc_request **request);
799 int ll_fsync(struct file *file, struct dentry *dentry, int data);
800 int ll_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
806 * protect race ll_find_aliases vs ll_revalidate_it vs ll_unhash_aliases
808 extern spinlock_t ll_lookup_lock;
809 extern struct dentry_operations ll_d_ops;
810 void ll_intent_drop_lock(struct lookup_intent *);
811 void ll_intent_release(struct lookup_intent *);
812 extern void ll_set_dd(struct dentry *de);
813 int ll_drop_dentry(struct dentry *dentry);
814 void ll_unhash_aliases(struct inode *);
815 void ll_frob_intent(struct lookup_intent **itp, struct lookup_intent *deft);
816 void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry);
817 int ll_dcompare(struct dentry *parent, struct qstr *d_name, struct qstr *name);
818 int revalidate_it_finish(struct ptlrpc_request *request, int offset,
819 struct lookup_intent *it, struct dentry *de);
821 /* llite/llite_lib.c */
822 extern struct super_operations lustre_super_operations;
824 char *ll_read_opt(const char *opt, char *data);
825 void ll_lli_init(struct ll_inode_info *lli);
826 int ll_fill_super(struct super_block *sb);
827 void ll_put_super(struct super_block *sb);
828 void ll_kill_super(struct super_block *sb);
829 struct inode *ll_inode_from_lock(struct ldlm_lock *lock);
830 void ll_clear_inode(struct inode *inode);
831 int ll_setattr_raw(struct inode *inode, struct iattr *attr);
832 int ll_setattr(struct dentry *de, struct iattr *attr);
833 #ifndef HAVE_STATFS_DENTRY_PARAM
834 int ll_statfs(struct super_block *sb, struct kstatfs *sfs);
836 int ll_statfs(struct dentry *de, struct kstatfs *sfs);
838 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
839 __u64 max_age, __u32 flags);
840 void ll_update_inode(struct inode *inode, struct lustre_md *md);
841 void ll_read_inode2(struct inode *inode, void *opaque);
842 int ll_iocontrol(struct inode *inode, struct file *file,
843 unsigned int cmd, unsigned long arg);
844 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
845 void ll_umount_begin(struct vfsmount *vfsmnt, int flags);
847 void ll_umount_begin(struct super_block *sb);
849 int ll_remount_fs(struct super_block *sb, int *flags, char *data);
850 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs);
851 int ll_prep_inode(struct obd_export *exp, struct inode **inode,
852 struct ptlrpc_request *req, int offset, struct super_block *);
853 void lustre_dump_dentry(struct dentry *, int recur);
854 void lustre_dump_inode(struct inode *);
855 struct ll_async_page *llite_pglist_next_llap(struct list_head *head,
856 struct list_head *list);
857 int ll_obd_statfs(struct inode *inode, void *arg);
858 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *max_mdsize);
859 int ll_process_config(struct lustre_cfg *lcfg);
861 /* llite/llite_nfs.c */
862 extern struct export_operations lustre_export_operations;
863 __u32 get_uuid2int(const char *name, int len);
864 struct dentry *ll_fh_to_dentry(struct super_block *sb, __u32 *data, int len,
865 int fhtype, int parent);
866 int ll_dentry_to_fh(struct dentry *, __u32 *datap, int *lenp, int need_parent);
868 /* llite/special.c */
869 extern struct inode_operations ll_special_inode_operations;
870 extern struct file_operations ll_special_chr_inode_fops;
871 extern struct file_operations ll_special_chr_file_fops;
872 extern struct file_operations ll_special_blk_inode_fops;
873 extern struct file_operations ll_special_fifo_inode_fops;
874 extern struct file_operations ll_special_fifo_file_fops;
875 extern struct file_operations ll_special_sock_inode_fops;
877 /* llite/symlink.c */
878 extern struct inode_operations ll_fast_symlink_inode_operations;
880 /* llite/llite_close.c */
881 struct ll_close_queue {
883 struct list_head lcq_list;
884 wait_queue_head_t lcq_waitq;
885 struct completion lcq_comp;
888 #ifdef HAVE_CLOSE_THREAD
889 void llap_write_pending(struct inode *inode, struct ll_async_page *llap);
890 void llap_write_complete(struct inode *inode, struct ll_async_page *llap);
891 void ll_open_complete(struct inode *inode);
892 int ll_is_inode_dirty(struct inode *inode);
893 void ll_try_done_writing(struct inode *inode);
894 void ll_queue_done_writing(struct inode *inode);
896 static inline void llap_write_pending(struct inode *inode,
897 struct ll_async_page *llap) { return; };
898 static inline void llap_write_complete(struct inode *inode,
899 struct ll_async_page *llap) { return; };
900 static inline void ll_open_complete(struct inode *inode) { return; };
901 static inline int ll_is_inode_dirty(struct inode *inode) { return 0; };
902 static inline void ll_try_done_writing(struct inode *inode) { return; };
903 static inline void ll_queue_done_writing(struct inode *inode) { return; };
904 //static inline void ll_close_thread_shutdown(struct ll_close_queue *lcq) { return; };
905 //static inline int ll_close_thread_start(struct ll_close_queue **lcq_ret) { return 0; };
907 void ll_close_thread_shutdown(struct ll_close_queue *lcq);
908 int ll_close_thread_start(struct ll_close_queue **lcq_ret);
910 /* llite/llite_mmap.c */
911 typedef struct rb_root rb_root_t;
912 typedef struct rb_node rb_node_t;
914 struct ll_lock_tree_node;
915 struct ll_lock_tree {
917 struct list_head lt_locked_list;
918 struct ll_file_data *lt_fd;
921 int ll_teardown_mmaps(struct address_space *mapping, __u64 first, __u64 last);
922 int ll_file_mmap(struct file * file, struct vm_area_struct * vma);
923 struct ll_lock_tree_node * ll_node_from_inode(struct inode *inode, __u64 start,
924 __u64 end, ldlm_mode_t mode);
925 int ll_tree_lock(struct ll_lock_tree *tree,
926 struct ll_lock_tree_node *first_node,
927 const char *buf, size_t count, int ast_flags);
928 int ll_tree_lock_iov(struct ll_lock_tree *tree,
929 struct ll_lock_tree_node *first_node,
930 const struct iovec *iov, unsigned long nr_segs,
932 int ll_tree_unlock(struct ll_lock_tree *tree);
934 #define ll_s2sbi(sb) (s2lsi(sb)->lsi_llsbi)
936 static inline __u64 ll_ts2u64(struct timespec *time)
938 __u64 t = time->tv_sec;
942 /* don't need an addref as the sb_info should be holding one */
943 static inline struct obd_export *ll_s2obdexp(struct super_block *sb)
945 return ll_s2sbi(sb)->ll_osc_exp;
948 /* don't need an addref as the sb_info should be holding one */
949 static inline struct obd_export *ll_s2mdcexp(struct super_block *sb)
951 return ll_s2sbi(sb)->ll_mdc_exp;
954 static inline struct client_obd *sbi2mdc(struct ll_sb_info *sbi)
956 struct obd_device *obd = sbi->ll_mdc_exp->exp_obd;
962 // FIXME: replace the name of this with LL_SB to conform to kernel stuff
963 static inline struct ll_sb_info *ll_i2sbi(struct inode *inode)
965 return ll_s2sbi(inode->i_sb);
968 static inline struct obd_export *ll_i2obdexp(struct inode *inode)
970 return ll_s2obdexp(inode->i_sb);
973 static inline struct obd_export *ll_i2mdcexp(struct inode *inode)
975 return ll_s2mdcexp(inode->i_sb);
978 /** get lu_fid from inode. */
979 static inline struct lu_fid *ll_inode_lu_fid(struct inode *inode)
981 return &ll_i2info(inode)->lli_fid.f20;
984 /** get ll_fid from inode. */
985 static inline struct ll_fid *ll_inode_ll_fid(struct inode *inode)
987 return &ll_i2info(inode)->lli_fid.f16;
990 static inline void ll_inode2fid(struct ll_fid *fid, struct inode *inode)
992 *fid = *ll_inode_ll_fid(inode);
995 static inline int ll_mds_max_easize(struct super_block *sb)
997 return sbi2mdc(ll_s2sbi(sb))->cl_max_mds_easize;
1000 static inline __u64 ll_file_maxbytes(struct inode *inode)
1002 return ll_i2info(inode)->lli_maxbytes;
1006 int ll_setxattr(struct dentry *dentry, const char *name,
1007 const void *value, size_t size, int flags);
1008 ssize_t ll_getxattr(struct dentry *dentry, const char *name,
1009 void *buffer, size_t size);
1010 ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size);
1011 int ll_removexattr(struct dentry *dentry, const char *name);
1015 #define LL_SA_RPC_MIN 2
1016 #define LL_SA_RPC_DEF 32
1017 #define LL_SA_RPC_MAX 8192
1019 /* per inode struct, for dir only */
1020 struct ll_statahead_info {
1021 struct inode *sai_inode;
1022 unsigned int sai_generation; /* generation for statahead */
1023 atomic_t sai_refcount; /* when access this struct, hold
1025 unsigned int sai_sent; /* stat requests sent count */
1026 unsigned int sai_replied; /* stat requests which received
1028 unsigned int sai_max; /* max ahead of lookup */
1029 unsigned int sai_index; /* index of statahead entry */
1030 unsigned int sai_index_next; /* index for the next statahead
1031 * entry to be stated */
1032 unsigned int sai_hit; /* hit count */
1033 unsigned int sai_miss; /* miss count:
1034 * for "ls -al" case, it includes
1035 * hidden dentry miss;
1036 * for "ls -l" case, it does not
1037 * include hidden dentry miss.
1038 * "sai_miss_hidden" is used for
1041 unsigned int sai_consecutive_miss; /* consecutive miss */
1042 unsigned int sai_miss_hidden;/* "ls -al", but first dentry
1043 * is not a hidden one */
1044 unsigned int sai_skip_hidden;/* skipped hidden dentry count */
1045 unsigned int sai_ls_all:1; /* "ls -al", do stat-ahead for
1047 cfs_waitq_t sai_waitq; /* stat-ahead wait queue */
1048 struct ptlrpc_thread sai_thread; /* stat-ahead thread */
1049 struct list_head sai_entries_sent; /* entries sent out */
1050 struct list_head sai_entries_received; /* entries returned */
1051 struct list_head sai_entries_stated; /* entries stated */
1054 int do_statahead_enter(struct inode *dir, struct dentry **dentry, int lookup);
1055 void ll_statahead_exit(struct dentry *dentry, int result);
1056 void ll_stop_statahead(struct inode *inode, void *key);
1059 void ll_statahead_mark(struct dentry *dentry)
1061 struct ll_inode_info *lli = ll_i2info(dentry->d_parent->d_inode);
1062 struct ll_dentry_data *ldd = ll_d2d(dentry);
1064 /* not the same process, don't mark */
1065 if (lli->lli_opendir_pid != cfs_curproc_pid())
1068 spin_lock(&lli->lli_lock);
1069 if (likely(lli->lli_sai != NULL && ldd != NULL))
1070 ldd->lld_sa_generation = lli->lli_sai->sai_generation;
1071 spin_unlock(&lli->lli_lock);
1075 int ll_statahead_enter(struct inode *dir, struct dentry **dentryp, int lookup)
1077 struct ll_sb_info *sbi = ll_i2sbi(dir);
1078 struct ll_inode_info *lli = ll_i2info(dir);
1079 struct ll_dentry_data *ldd = ll_d2d(*dentryp);
1081 if (sbi->ll_sa_max == 0)
1084 /* temporarily disable dir stat ahead in interoperability mode */
1085 if (sbi->ll_mdc_exp->exp_connect_flags & OBD_CONNECT_FID)
1088 /* not the same process, don't statahead */
1089 if (lli->lli_opendir_pid != cfs_curproc_pid())
1093 * When "ls" a dentry, the system trigger more than once "revalidate" or
1094 * "lookup", for "getattr", for "getxattr", and maybe for others.
1095 * Under patchless client mode, the operation intent is not accurate,
1096 * it maybe misguide the statahead thread. For example:
1097 * The "revalidate" call for "getattr" and "getxattr" of a dentry maybe
1098 * have the same operation intent -- "IT_GETATTR".
1099 * In fact, one dentry should has only one chance to interact with the
1100 * statahead thread, otherwise the statahead windows will be confused.
1101 * The solution is as following:
1102 * Assign "lld_sa_generation" with "sai_generation" when a dentry
1103 * "IT_GETATTR" for the first time, and the subsequent "IT_GETATTR"
1104 * will bypass interacting with statahead thread for checking:
1105 * "lld_sa_generation == lli_sai->sai_generation"
1107 if (ldd && lli->lli_sai &&
1108 ldd->lld_sa_generation == lli->lli_sai->sai_generation)
1111 return do_statahead_enter(dir, dentryp, lookup);
1114 static void inline ll_dops_init(struct dentry *de, int block)
1116 struct ll_dentry_data *lld = ll_d2d(de);
1118 if (lld == NULL && block != 0) {
1124 lld->lld_sa_generation = 0;
1126 de->d_op = &ll_d_ops;
1129 /* llite ioctl register support rountine */
1136 #define LLIOC_MAX_CMD 256
1139 * Rules to write a callback function:
1142 * @magic: Dynamic ioctl call routine will feed this vaule with the pointer
1143 * returned to ll_iocontrol_register. Callback functions should use this
1144 * data to check the potential collasion of ioctl cmd. If collasion is
1145 * found, callback function should return LLIOC_CONT.
1146 * @rcp: The result of ioctl command.
1149 * If @magic matches the pointer returned by ll_iocontrol_data, the
1150 * callback should return LLIOC_STOP; return LLIOC_STOP otherwise.
1152 typedef enum llioc_iter (*llioc_callback_t)(struct inode *inode,
1153 struct file *file, unsigned int cmd, unsigned long arg,
1154 void *magic, int *rcp);
1156 enum llioc_iter ll_iocontrol_call(struct inode *inode, struct file *file,
1157 unsigned int cmd, unsigned long arg, int *rcp);
1159 /* export functions */
1160 /* Register ioctl block dynamatically for a regular file.
1162 * @cmd: the array of ioctl command set
1163 * @count: number of commands in the @cmd
1164 * @cb: callback function, it will be called if an ioctl command is found to
1165 * belong to the command list @cmd.
1168 * A magic pointer will be returned if success;
1169 * otherwise, NULL will be returned.
1171 void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd);
1172 void ll_iocontrol_unregister(void *magic);
1174 ino_t ll_fid_build_ino(struct ll_sb_info *sbi,
1175 struct ll_fid *fid);
1176 __u32 ll_fid_build_gen(struct ll_sb_info *sbi,
1177 struct ll_fid *fid);
1181 #endif /* LLITE_INTERNAL_H */