4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
32 #ifndef LLITE_INTERNAL_H
33 #define LLITE_INTERNAL_H
35 #include <lustre_disk.h> /* for s2sbi */
36 #include <lustre_linkea.h>
38 /* for struct cl_lock_descr and struct cl_io */
39 #include <cl_object.h>
40 #include <lustre_lmv.h>
41 #include <lustre_mdc.h>
42 #include <lustre_intent.h>
43 #include <linux/compat.h>
44 #include <linux/aio.h>
45 #include <lustre_compat.h>
46 #include <lustre_crypto.h>
47 #include <range_lock.h>
49 #include "vvp_internal.h"
51 #include "foreign_symlink.h"
57 #ifndef HAVE_VM_FAULT_RETRY
58 #define VM_FAULT_RETRY 0
61 /* Kernel 3.1 kills LOOKUP_CONTINUE, LOOKUP_PARENT is equivalent to it.
62 * seem kernel commit 49084c3bb2055c401f3493c13edae14d49128ca0 */
63 #ifndef LOOKUP_CONTINUE
64 #define LOOKUP_CONTINUE LOOKUP_PARENT
67 /** Only used on client-side for indicating the tail of dir hash/offset. */
68 #define LL_DIR_END_OFF 0x7fffffffffffffffULL
69 #define LL_DIR_END_OFF_32BIT 0x7fffffffUL
71 /* 4UL * 1024 * 1024 */
72 #define LL_MAX_BLKSIZE_BITS 22
74 #define LL_IT2STR(it) ((it) ? ldlm_it2str((it)->it_op) : "0")
76 #define TIMES_SET_FLAGS (ATTR_MTIME_SET | ATTR_ATIME_SET | ATTR_TIMES_SET)
78 struct ll_dentry_data {
79 unsigned int lld_sa_generation;
80 unsigned int lld_invalid:1;
81 unsigned int lld_nfs_dentry:1;
82 struct rcu_head lld_rcu_head;
85 #define ll_d2d(de) ((struct ll_dentry_data*)((de)->d_fsdata))
87 #define LLI_INODE_MAGIC 0x111d0de5
88 #define LLI_INODE_DEAD 0xdeadd00d
90 struct ll_getname_data {
91 #ifdef HAVE_DIR_CONTEXT
92 struct dir_context ctx;
94 char *lgd_name; /* points to a buffer with NAME_MAX+1 size */
95 struct lu_fid lgd_fid; /* target fid we are looking for */
96 int lgd_found; /* inode matched? */
100 struct lu_env *lg_env;
102 struct cl_lock *lg_lock;
103 unsigned long lg_gid;
106 /* See comment on trunc_sem_down_read_nowait */
107 struct ll_trunc_sem {
108 /* when positive, this is a count of readers, when -1, it indicates
109 * the semaphore is held for write, and 0 is unlocked
111 atomic_t ll_trunc_readers;
112 /* this tracks a count of waiting writers */
113 atomic_t ll_trunc_waiters;
116 struct ll_inode_info {
117 __u32 lli_inode_magic;
120 volatile unsigned long lli_flags;
121 struct posix_acl *lli_posix_acl;
123 /* identifying fields for both metadata and data stacks. */
124 struct lu_fid lli_fid;
125 /* master inode fid for stripe directory */
126 struct lu_fid lli_pfid;
128 /* We need all three because every inode may be opened in different
130 struct obd_client_handle *lli_mds_read_och;
131 struct obd_client_handle *lli_mds_write_och;
132 struct obd_client_handle *lli_mds_exec_och;
133 __u64 lli_open_fd_read_count;
134 __u64 lli_open_fd_write_count;
135 __u64 lli_open_fd_exec_count;
137 /* Number of times this inode was opened */
138 u64 lli_open_fd_count;
139 /* When last close was performed on this inode */
140 ktime_t lli_close_fd_time;
142 /* Protects access to och pointers and their usage counters */
143 struct mutex lli_och_mutex;
145 struct inode lli_vfs_inode;
147 /* the most recent timestamps obtained from mds */
152 spinlock_t lli_agl_lock;
154 /* Try to make the d::member and f::member are aligned. Before using
155 * these members, make clear whether it is directory or not. */
159 /* metadata statahead */
160 /* since parent-child threads can share the same @file
161 * struct, "opendir_key" is the token when dir close for
162 * case of parent exit before child -- it is me should
163 * cleanup the dir readahead. */
164 void *lli_opendir_key;
165 struct ll_statahead_info *lli_sai;
166 /* protect statahead stuff. */
167 spinlock_t lli_sa_lock;
168 /* "opendir_pid" is the token when lookup/revalid
169 * -- I am the owner of dir statahead. */
170 pid_t lli_opendir_pid;
171 /* directory depth to ROOT */
172 unsigned short lli_depth;
173 /* stat will try to access statahead entries or start
174 * statahead if this flag is set, and this flag will be
175 * set upon dir open, and cleared when dir is closed,
176 * statahead hit ratio is too low, or start statahead
178 unsigned short lli_sa_enabled:1;
179 /* generation for statahead */
180 unsigned int lli_sa_generation;
181 /* rw lock protects lli_lsm_md */
182 struct rw_semaphore lli_lsm_sem;
183 /* directory stripe information */
184 struct lmv_stripe_md *lli_lsm_md;
185 /* directory default LMV */
186 struct lmv_stripe_md *lli_default_lsm_md;
189 /* for non-directory */
191 struct mutex lli_size_mutex;
192 char *lli_symlink_name;
193 struct ll_trunc_sem lli_trunc_sem;
194 struct range_lock_tree lli_write_tree;
195 struct mutex lli_setattr_mutex;
197 struct rw_semaphore lli_glimpse_sem;
198 ktime_t lli_glimpse_time;
199 struct list_head lli_agl_list;
202 /* for writepage() only to communicate to fsync */
205 /* protect the file heat fields */
206 spinlock_t lli_heat_lock;
207 __u32 lli_heat_flags;
208 struct obd_heat_instance lli_heat_instances[OBD_HEAT_COUNT];
211 * Whenever a process try to read/write the file, the
212 * jobid of the process will be saved here, and it'll
213 * be packed into the write PRC when flush later.
215 * So the read/write statistics for jobid will not be
216 * accurate if the file is shared by different jobs.
218 char lli_jobid[LUSTRE_JOBID_SIZE];
220 struct mutex lli_pcc_lock;
221 enum lu_pcc_state_flags lli_pcc_state;
223 * @lli_pcc_generation saves the gobal PCC generation
224 * when the file was successfully attached into PCC.
225 * The flags of the PCC dataset are saved in
227 * The gobal PCC generation will be increased when add
228 * or delete a PCC backend, or change the configuration
229 * parameters for PCC.
230 * If @lli_pcc_generation is same as the gobal PCC
231 * generation, we can use the saved flags of the PCC
232 * dataset to determine whether need to try auto attach
235 __u64 lli_pcc_generation;
236 enum pcc_dataset_flags lli_pcc_dsflags;
237 struct pcc_inode *lli_pcc_inode;
239 struct mutex lli_group_mutex;
240 __u64 lli_group_users;
241 unsigned long lli_group_gid;
243 __u64 lli_attr_valid;
245 __u64 lli_lazyblocks;
249 /* XXX: For following frequent used members, although they maybe special
250 * used for non-directory object, it is some time-wasting to check
251 * whether the object is directory or not before using them. On the
252 * other hand, currently, sizeof(f) > sizeof(d), it cannot reduce
253 * the "ll_inode_info" size even if moving those members into u.f.
254 * So keep them out side.
256 * In the future, if more members are added only for directory,
257 * some of the following members can be moved into u.f.
259 struct cl_object *lli_clob;
261 /* mutex to request for layout lock exclusively. */
262 struct mutex lli_layout_mutex;
263 /* Layout version, protected by lli_layout_lock */
264 __u32 lli_layout_gen;
265 spinlock_t lli_layout_lock;
267 __u32 lli_projid; /* project id */
269 struct rw_semaphore lli_xattrs_list_rwsem;
270 struct mutex lli_xattrs_enq_lock;
271 struct list_head lli_xattrs; /* ll_xattr_entry->xe_list */
274 static inline void ll_trunc_sem_init(struct ll_trunc_sem *sem)
276 atomic_set(&sem->ll_trunc_readers, 0);
277 atomic_set(&sem->ll_trunc_waiters, 0);
280 /* This version of down read ignores waiting writers, meaning if the semaphore
281 * is already held for read, this down_read will 'join' that reader and also
282 * take the semaphore.
284 * This lets us avoid an unusual deadlock.
286 * We must take lli_trunc_sem in read mode on entry in to various i/o paths
287 * in Lustre, in order to exclude truncates. Some of these paths then need to
288 * take the mmap_lock, while still holding the trunc_sem. The problem is that
289 * page faults hold the mmap_lock when calling in to Lustre, and then must also
290 * take the trunc_sem to exclude truncate.
292 * This means the locking order for trunc_sem and mmap_lock is sometimes AB,
293 * sometimes BA. This is almost OK because in both cases, we take the trunc
294 * sem for read, so it doesn't block.
296 * However, if a write mode user (truncate, a setattr op) arrives in the
297 * middle of this, the second reader on the truncate_sem will wait behind that
300 * So we have, on our truncate sem, in order (where 'reader' and 'writer' refer
301 * to the mode in which they take the semaphore):
302 * reader (holding mmap_lock, needs truncate_sem)
304 * reader (holding truncate sem, waiting for mmap_lock)
306 * And so the readers deadlock.
308 * The solution is this modified semaphore, where this down_read ignores
309 * waiting write operations, and all waiters are woken up at once, so readers
310 * using down_read_nowait cannot get stuck behind waiting writers, regardless
311 * of the order they arrived in.
313 * down_read_nowait is only used in the page fault case, where we already hold
314 * the mmap_lock. This is because otherwise repeated read and write operations
315 * (which take the truncate sem) could prevent a truncate from ever starting.
316 * This could still happen with page faults, but without an even more complex
317 * mechanism, this is unavoidable.
321 static inline void trunc_sem_down_read_nowait(struct ll_trunc_sem *sem)
323 wait_var_event(&sem->ll_trunc_readers,
324 atomic_inc_unless_negative(&sem->ll_trunc_readers));
327 static inline void trunc_sem_down_read(struct ll_trunc_sem *sem)
329 wait_var_event(&sem->ll_trunc_readers,
330 atomic_read(&sem->ll_trunc_waiters) == 0 &&
331 atomic_inc_unless_negative(&sem->ll_trunc_readers));
334 static inline void trunc_sem_up_read(struct ll_trunc_sem *sem)
336 if (atomic_dec_return(&sem->ll_trunc_readers) == 0 &&
337 atomic_read(&sem->ll_trunc_waiters))
338 wake_up_var(&sem->ll_trunc_readers);
341 static inline void trunc_sem_down_write(struct ll_trunc_sem *sem)
343 atomic_inc(&sem->ll_trunc_waiters);
344 wait_var_event(&sem->ll_trunc_readers,
345 atomic_cmpxchg(&sem->ll_trunc_readers, 0, -1) == 0);
346 atomic_dec(&sem->ll_trunc_waiters);
349 static inline void trunc_sem_up_write(struct ll_trunc_sem *sem)
351 atomic_set(&sem->ll_trunc_readers, 0);
352 wake_up_var(&sem->ll_trunc_readers);
355 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
356 static inline void lli_clear_acl(struct ll_inode_info *lli)
358 if (lli->lli_posix_acl) {
359 posix_acl_release(lli->lli_posix_acl);
360 lli->lli_posix_acl = NULL;
364 static inline void lli_replace_acl(struct ll_inode_info *lli,
365 struct lustre_md *md)
367 spin_lock(&lli->lli_lock);
368 if (lli->lli_posix_acl)
369 posix_acl_release(lli->lli_posix_acl);
370 lli->lli_posix_acl = md->posix_acl;
371 spin_unlock(&lli->lli_lock);
374 static inline void lli_clear_acl(struct ll_inode_info *lli)
378 static inline void lli_replace_acl(struct ll_inode_info *lli,
379 struct lustre_md *md)
384 static inline __u32 ll_layout_version_get(struct ll_inode_info *lli)
388 spin_lock(&lli->lli_layout_lock);
389 gen = lli->lli_layout_gen;
390 spin_unlock(&lli->lli_layout_lock);
395 static inline void ll_layout_version_set(struct ll_inode_info *lli, __u32 gen)
397 spin_lock(&lli->lli_layout_lock);
398 lli->lli_layout_gen = gen;
399 spin_unlock(&lli->lli_layout_lock);
403 /* File data is modified. */
404 LLIF_DATA_MODIFIED = 0,
405 /* File is being restored */
406 LLIF_FILE_RESTORING = 1,
407 /* Xattr cache is attached to the file */
408 LLIF_XATTR_CACHE = 2,
409 /* Project inherit */
410 LLIF_PROJECT_INHERIT = 3,
411 /* update atime from MDS even if it's older than local inode atime. */
412 LLIF_UPDATE_ATIME = 4,
413 /* foreign file/dir can be unlinked unconditionnaly */
414 LLIF_FOREIGN_REMOVABLE = 5,
415 /* setting encryption context in progress */
416 LLIF_SET_ENC_CTX = 6,
420 int ll_xattr_cache_destroy(struct inode *inode);
422 int ll_xattr_cache_get(struct inode *inode,
428 int ll_xattr_cache_insert(struct inode *inode,
433 static inline bool obd_connect_has_secctx(struct obd_connect_data *data)
435 #ifdef CONFIG_SECURITY
436 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
437 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
443 static inline void obd_connect_set_secctx(struct obd_connect_data *data)
445 #ifdef CONFIG_SECURITY
446 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
450 int ll_dentry_init_security(struct dentry *dentry, int mode, struct qstr *name,
451 const char **secctx_name, void **secctx,
453 int ll_inode_init_security(struct dentry *dentry, struct inode *inode,
456 int ll_listsecurity(struct inode *inode, char *secctx_name,
457 size_t secctx_name_size);
459 static inline bool obd_connect_has_enc(struct obd_connect_data *data)
461 #ifdef HAVE_LUSTRE_CRYPTO
462 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
463 data->ocd_connect_flags2 & OBD_CONNECT2_ENCRYPT;
469 static inline void obd_connect_set_enc(struct obd_connect_data *data)
471 #ifdef HAVE_LUSTRE_CRYPTO
472 data->ocd_connect_flags2 |= OBD_CONNECT2_ENCRYPT;
477 * Locking to guarantee consistency of non-atomic updates to long long i_size,
478 * consistency between file size and KMS.
480 * Implemented by ->lli_size_mutex and ->lsm_lock, nested in that order.
483 void ll_inode_size_lock(struct inode *inode);
484 void ll_inode_size_unlock(struct inode *inode);
486 static inline struct ll_inode_info *ll_i2info(struct inode *inode)
488 return container_of(inode, struct ll_inode_info, lli_vfs_inode);
491 static inline struct pcc_inode *ll_i2pcci(struct inode *inode)
493 return ll_i2info(inode)->lli_pcc_inode;
496 /* default to use at least 16M for fast read if possible */
497 #define RA_REMAIN_WINDOW_MIN MiB_TO_PAGES(16UL)
499 /* default read-ahead on a given client mountpoint. */
500 #define SBI_DEFAULT_READ_AHEAD_MAX MiB_TO_PAGES(1024UL)
502 /* default read-ahead for a single file descriptor */
503 #define SBI_DEFAULT_READ_AHEAD_PER_FILE_MAX MiB_TO_PAGES(256UL)
505 /* default read-ahead full files smaller than limit on the second read */
506 #define SBI_DEFAULT_READ_AHEAD_WHOLE_MAX MiB_TO_PAGES(2UL)
508 /* default range pages */
509 #define SBI_DEFAULT_RA_RANGE_PAGES MiB_TO_PAGES(1ULL)
511 /* Min range pages */
512 #define RA_MIN_MMAP_RANGE_PAGES 16UL
517 RA_STAT_DISTANT_READPAGE,
518 RA_STAT_MISS_IN_WINDOW,
519 RA_STAT_FAILED_GRAB_PAGE,
520 RA_STAT_FAILED_MATCH,
525 RA_STAT_MAX_IN_FLIGHT,
526 RA_STAT_WRONG_GRAB_PAGE,
527 RA_STAT_FAILED_REACH_END,
529 RA_STAT_FAILED_FAST_READ,
530 RA_STAT_MMAP_RANGE_READ,
535 atomic_t ra_cur_pages;
536 unsigned long ra_max_pages;
537 unsigned long ra_max_pages_per_file;
538 unsigned long ra_range_pages;
539 unsigned long ra_max_read_ahead_whole_pages;
540 struct workqueue_struct *ll_readahead_wq;
542 * Max number of active works could be triggered
543 * for async readahead.
545 unsigned int ra_async_max_active;
546 /* how many async readahead triggered in flight */
547 atomic_t ra_async_inflight;
548 /* Threshold to control when to trigger async readahead */
549 unsigned long ra_async_pages_per_file_threshold;
552 /* ra_io_arg will be filled in the beginning of ll_readahead with
553 * ras_lock, then the following ll_read_ahead_pages will read RA
554 * pages according to this arg, all the items in this structure are
555 * counted by page index.
558 pgoff_t ria_start_idx; /* start offset of read-ahead*/
559 pgoff_t ria_end_idx; /* end offset of read-ahead*/
560 unsigned long ria_reserved; /* reserved pages for read-ahead */
561 pgoff_t ria_end_idx_min;/* minimum end to cover current read */
562 bool ria_eof; /* reach end of file */
563 /* If stride read pattern is detected, ria_stoff is the byte offset
564 * where stride read is started. Note: for normal read-ahead, the
565 * value here is meaningless, and also it will not be accessed*/
567 /* ria_length and ria_bytes are the length and pages length in the
568 * stride I/O mode. And they will also be used to check whether
569 * it is stride I/O read-ahead in the read-ahead pages*/
574 /* LL_HIST_MAX=32 causes an overflow */
575 #define LL_HIST_MAX 28
576 #define LL_HIST_START 12 /* buckets start at 2^12 = 4k */
577 #define LL_PROCESS_HIST_MAX 10
578 struct per_process_info {
580 struct obd_histogram pp_r_hist;
581 struct obd_histogram pp_w_hist;
584 /* pp_extents[LL_PROCESS_HIST_MAX] will hold the combined process info */
585 struct ll_rw_extents_info {
586 struct per_process_info pp_extents[LL_PROCESS_HIST_MAX + 1];
589 #define LL_OFFSET_HIST_MAX 100
590 struct ll_rw_process_info {
593 loff_t rw_range_start;
595 loff_t rw_last_file_pos;
597 size_t rw_smallest_extent;
598 size_t rw_largest_extent;
599 struct ll_file_data *rw_last_file;
602 enum stats_track_type {
603 STATS_TRACK_ALL = 0, /* track all processes */
604 STATS_TRACK_PID, /* track process with this pid */
605 STATS_TRACK_PPID, /* track processes with this ppid */
606 STATS_TRACK_GID, /* track processes with this gid */
610 /* flags for sbi->ll_flags */
611 #define LL_SBI_NOLCK 0x01 /* DLM locking disabled (directio-only) */
612 #define LL_SBI_CHECKSUM 0x02 /* checksum each page as it's written */
613 #define LL_SBI_FLOCK 0x04
614 #define LL_SBI_USER_XATTR 0x08 /* support user xattr */
615 #define LL_SBI_ACL 0x10 /* support ACL */
616 /* LL_SBI_RMT_CLIENT 0x40 remote client */
617 #define LL_SBI_MDS_CAPA 0x80 /* support mds capa, obsolete */
618 #define LL_SBI_OSS_CAPA 0x100 /* support oss capa, obsolete */
619 #define LL_SBI_LOCALFLOCK 0x200 /* Local flocks support by kernel */
620 #define LL_SBI_LRU_RESIZE 0x400 /* lru resize support */
621 #define LL_SBI_LAZYSTATFS 0x800 /* lazystatfs mount option */
622 /* LL_SBI_SOM_PREVIEW 0x1000 SOM preview mount option, obsolete */
623 #define LL_SBI_32BIT_API 0x2000 /* generate 32 bit inodes. */
624 #define LL_SBI_64BIT_HASH 0x4000 /* support 64-bits dir hash/offset */
625 #define LL_SBI_AGL_ENABLED 0x8000 /* enable agl */
626 #define LL_SBI_VERBOSE 0x10000 /* verbose mount/umount */
627 #define LL_SBI_LAYOUT_LOCK 0x20000 /* layout lock support */
628 #define LL_SBI_USER_FID2PATH 0x40000 /* allow fid2path by unprivileged users */
629 #define LL_SBI_XATTR_CACHE 0x80000 /* support for xattr cache */
630 #define LL_SBI_NOROOTSQUASH 0x100000 /* do not apply root squash */
631 #define LL_SBI_ALWAYS_PING 0x200000 /* always ping even if server
633 #define LL_SBI_FAST_READ 0x400000 /* fast read support */
634 #define LL_SBI_FILE_SECCTX 0x800000 /* set file security context at create */
635 /* LL_SBI_PIO 0x1000000 parallel IO support, introduced in
637 #define LL_SBI_TINY_WRITE 0x2000000 /* tiny write support */
638 #define LL_SBI_FILE_HEAT 0x4000000 /* file heat support */
639 #define LL_SBI_TEST_DUMMY_ENCRYPTION 0x8000000 /* test dummy encryption */
640 #define LL_SBI_ENCRYPT 0x10000000 /* client side encryption */
641 #define LL_SBI_FOREIGN_SYMLINK 0x20000000 /* foreign fake-symlink support */
642 /* foreign fake-symlink upcall registered */
643 #define LL_SBI_FOREIGN_SYMLINK_UPCALL 0x40000000
644 #define LL_SBI_PARALLEL_DIO 0x80000000 /* parallel (async) submission of
646 #define LL_SBI_FLAGS { \
674 "test_dummy_encryption", \
677 "foreign_symlink_upcall", \
681 /* This is embedded into llite super-blocks to keep track of connect
682 * flags (capabilities) supported by all imports given mount is
684 struct lustre_client_ocd {
685 /* This is conjunction of connect_flags across all imports
686 * (LOVs) this mount is connected to. This field is updated by
687 * cl_ocd_update() under ->lco_lock. */
689 struct mutex lco_lock;
690 struct obd_export *lco_md_exp;
691 struct obd_export *lco_dt_exp;
695 /* this protects pglist and ra_info. It isn't safe to
696 * grab from interrupt contexts */
698 spinlock_t ll_pp_extent_lock; /* pp_extent entry*/
699 spinlock_t ll_process_lock; /* ll_rw_process_info */
700 struct obd_uuid ll_sb_uuid;
701 struct obd_export *ll_md_exp;
702 struct obd_export *ll_dt_exp;
703 struct obd_device *ll_md_obd;
704 struct obd_device *ll_dt_obd;
705 struct dentry *ll_debugfs_entry;
706 struct lu_fid ll_root_fid; /* root object fid */
709 unsigned int ll_xattr_cache_enabled:1,
710 ll_xattr_cache_set:1, /* already set to 0/1 */
711 ll_client_common_fill_super_succeeded:1,
714 struct lustre_client_ocd ll_lco;
716 struct lprocfs_stats *ll_stats; /* lprocfs stats counter */
718 /* Used to track "unstable" pages on a client, and maintain a
719 * LRU list of clean pages. An "unstable" page is defined as
720 * any page which is sent to a server as part of a bulk request,
721 * but is uncommitted to stable storage. */
722 struct cl_client_cache *ll_cache;
724 struct lprocfs_stats *ll_ra_stats;
726 struct ll_ra_info ll_ra_info;
727 unsigned int ll_namelen;
728 const struct file_operations *ll_fop;
730 struct lu_site *ll_site;
731 struct cl_device *ll_cl;
733 struct ll_rw_extents_info ll_rw_extents_info;
734 int ll_extent_process_count;
735 struct ll_rw_process_info ll_rw_process_info[LL_PROCESS_HIST_MAX];
736 unsigned int ll_offset_process_count;
737 struct ll_rw_process_info ll_rw_offset_info[LL_OFFSET_HIST_MAX];
738 unsigned int ll_rw_offset_entry_count;
739 int ll_stats_track_id;
740 enum stats_track_type ll_stats_track_type;
743 /* metadata stat-ahead */
744 unsigned int ll_sa_running_max;/* max concurrent
745 * statahead instances */
746 unsigned int ll_sa_max; /* max statahead RPCs */
747 atomic_t ll_sa_total; /* statahead thread started
749 atomic_t ll_sa_wrong; /* statahead thread stopped for
751 atomic_t ll_sa_running; /* running statahead thread
753 atomic_t ll_agl_total; /* AGL thread started count */
755 dev_t ll_sdev_orig; /* save s_dev before assign for
758 struct root_squash_info ll_squash;
761 /* st_blksize returned by stat(2), when non-zero */
762 unsigned int ll_stat_blksize;
764 /* maximum relative age of cached statfs results */
765 unsigned int ll_statfs_max_age;
767 struct kset ll_kset; /* sysfs object */
768 struct completion ll_kobj_unregister;
771 unsigned int ll_heat_decay_weight;
772 unsigned int ll_heat_period_second;
774 /* Opens of the same inode before we start requesting open lock */
775 u32 ll_oc_thrsh_count;
777 /* Time in ms between last inode close and next open to be considered
778 * instant back to back and would trigger an open lock request
782 /* Time in ms after last file close that we no longer count prior opens*/
785 /* filesystem fsname */
786 char ll_fsname[LUSTRE_MAXFSNAME + 1];
788 /* Persistent Client Cache */
789 struct pcc_super ll_pcc_super;
791 /* to protect vs updates in all following foreign symlink fields */
792 struct rw_semaphore ll_foreign_symlink_sem;
793 /* foreign symlink path prefix */
794 char *ll_foreign_symlink_prefix;
795 /* full prefix size including leading '\0' */
796 size_t ll_foreign_symlink_prefix_size;
797 /* foreign symlink path upcall */
798 char *ll_foreign_symlink_upcall;
799 /* foreign symlink path upcall infos */
800 struct ll_foreign_symlink_upcall_item *ll_foreign_symlink_upcall_items;
801 /* foreign symlink path upcall nb infos */
802 unsigned int ll_foreign_symlink_upcall_nb_items;
805 #define SBI_DEFAULT_HEAT_DECAY_WEIGHT ((80 * 256 + 50) / 100)
806 #define SBI_DEFAULT_HEAT_PERIOD_SECOND (60)
808 #define SBI_DEFAULT_OPENCACHE_THRESHOLD_COUNT (5)
809 #define SBI_DEFAULT_OPENCACHE_THRESHOLD_MS (100) /* 0.1 second */
810 #define SBI_DEFAULT_OPENCACHE_THRESHOLD_MAX_MS (60000) /* 1 minute */
813 * per file-descriptor read-ahead data.
815 struct ll_readahead_state {
817 /* End byte that read(2) try to read. */
818 loff_t ras_last_read_end_bytes;
820 * number of bytes read after last read-ahead window reset. As window
821 * is reset on each seek, this is effectively a number of consecutive
822 * accesses. Maybe ->ras_accessed_in_window is better name.
824 * XXX nikita: window is also reset (by ras_update()) when Lustre
825 * believes that memory pressure evicts read-ahead pages. In that
826 * case, it probably doesn't make sense to expand window to
827 * PTLRPC_MAX_BRW_PAGES on the third access.
829 loff_t ras_consecutive_bytes;
831 * number of read requests after the last read-ahead window reset
832 * As window is reset on each seek, this is effectively the number
833 * on consecutive read request and is used to trigger read-ahead.
835 unsigned long ras_consecutive_requests;
837 * Parameters of current read-ahead window. Handled by
838 * ras_update(). On the initial access to the file or after a seek,
839 * window is reset to 0. After 3 consecutive accesses, window is
840 * expanded to PTLRPC_MAX_BRW_PAGES. Afterwards, window is enlarged by
841 * PTLRPC_MAX_BRW_PAGES chunks up to ->ra_max_pages.
843 pgoff_t ras_window_start_idx;
844 pgoff_t ras_window_pages;
846 /* Page index where min range read starts */
847 pgoff_t ras_range_min_start_idx;
848 /* Page index where mmap range read ends */
849 pgoff_t ras_range_max_end_idx;
850 /* number of mmap pages where last time detected */
851 pgoff_t ras_last_range_pages;
852 /* number of mmap range requests */
853 pgoff_t ras_range_requests;
856 * Optimal RPC size in pages.
857 * It decides how many pages will be sent for each read-ahead.
859 unsigned long ras_rpc_pages;
861 * Where next read-ahead should start at. This lies within read-ahead
862 * window. Read-ahead window is read in pieces rather than at once
863 * because: 1. lustre limits total number of pages under read-ahead by
864 * ->ra_max_pages (see ll_ra_count_get()), 2. client cannot read pages
865 * not covered by DLM lock.
867 pgoff_t ras_next_readahead_idx;
869 * Total number of ll_file_read requests issued, reads originating
870 * due to mmap are not counted in this total. This value is used to
871 * trigger full file read-ahead after multiple reads to a small file.
873 unsigned long ras_requests;
875 * The following 3 items are used for detecting the stride I/O
877 * In stride I/O mode,
878 * ...............|-----data-----|****gap*****|--------|******|....
879 * offset |-stride_bytes-|-stride_gap-|
880 * ras_stride_offset = offset;
881 * ras_stride_length = stride_bytes + stride_gap;
882 * ras_stride_bytes = stride_bytes;
883 * Note: all these three items are counted by bytes.
885 loff_t ras_stride_offset;
886 loff_t ras_stride_length;
887 loff_t ras_stride_bytes;
889 * number of consecutive stride request count, and it is similar as
890 * ras_consecutive_requests, but used for stride I/O mode.
891 * Note: only more than 2 consecutive stride request are detected,
892 * stride read-ahead will be enable
894 unsigned long ras_consecutive_stride_requests;
895 /* index of the last page that async readahead starts */
896 pgoff_t ras_async_last_readpage_idx;
897 /* whether we should increase readahead window */
898 bool ras_need_increase_window;
899 /* whether ra miss check should be skipped */
900 bool ras_no_miss_check;
903 struct ll_readahead_work {
904 /** File to readahead */
905 struct file *lrw_file;
906 pgoff_t lrw_start_idx;
909 /* async worker to handler read */
910 struct work_struct lrw_readahead_work;
911 char lrw_jobid[LUSTRE_JOBID_SIZE];
914 extern struct kmem_cache *ll_file_data_slab;
915 struct lustre_handle;
916 struct ll_file_data {
917 struct ll_readahead_state fd_ras;
918 struct ll_grouplock fd_grouplock;
922 /* openhandle if lease exists for this file.
923 * Borrow lli->lli_och_mutex to protect assignment */
924 struct obd_client_handle *fd_lease_och;
925 struct obd_client_handle *fd_och;
926 struct file *fd_file;
927 /* Indicate whether need to report failure when close.
928 * true: failure is known, not report again.
929 * false: unknown failure, should report. */
930 bool fd_write_failed;
931 bool ll_lock_no_expand;
932 rwlock_t fd_lock; /* protect lcc list */
933 struct list_head fd_lccs; /* list of ll_cl_context */
934 /* Used by mirrored file to lead IOs to a specific mirror, usually
935 * for mirror resync. 0 means default. */
936 __u32 fd_designated_mirror;
937 /* The layout version when resync starts. Resync I/O should carry this
938 * layout version for verification to OST objects */
939 __u32 fd_layout_version;
940 struct pcc_file fd_pcc_file;
943 void llite_tunables_unregister(void);
944 int llite_tunables_register(void);
946 static inline struct inode *ll_info2i(struct ll_inode_info *lli)
948 return &lli->lli_vfs_inode;
951 __u32 ll_i2suppgid(struct inode *i);
952 void ll_i2gids(__u32 *suppgids, struct inode *i1,struct inode *i2);
954 static inline int ll_need_32bit_api(struct ll_sb_info *sbi)
956 #if BITS_PER_LONG == 32
958 #elif defined(CONFIG_COMPAT)
959 if (unlikely(sbi->ll_flags & LL_SBI_32BIT_API))
962 # ifdef CONFIG_X86_X32
963 /* in_compat_syscall() returns true when called from a kthread
964 * and CONFIG_X86_X32 is enabled, which is wrong. So check
965 * whether the caller comes from a syscall (ie. not a kthread)
966 * before calling in_compat_syscall(). */
967 if (current->flags & PF_KTHREAD)
971 return unlikely(in_compat_syscall());
973 return unlikely(sbi->ll_flags & LL_SBI_32BIT_API);
977 static inline bool ll_sbi_has_fast_read(struct ll_sb_info *sbi)
979 return !!(sbi->ll_flags & LL_SBI_FAST_READ);
982 static inline bool ll_sbi_has_tiny_write(struct ll_sb_info *sbi)
984 return !!(sbi->ll_flags & LL_SBI_TINY_WRITE);
987 static inline bool ll_sbi_has_file_heat(struct ll_sb_info *sbi)
989 return !!(sbi->ll_flags & LL_SBI_FILE_HEAT);
992 static inline bool ll_sbi_has_foreign_symlink(struct ll_sb_info *sbi)
994 return !!(sbi->ll_flags & LL_SBI_FOREIGN_SYMLINK);
997 static inline bool ll_sbi_has_parallel_dio(struct ll_sb_info *sbi)
999 return !!(sbi->ll_flags & LL_SBI_PARALLEL_DIO);
1002 void ll_ras_enter(struct file *f, loff_t pos, size_t count);
1004 /* llite/lcommon_misc.c */
1005 int cl_ocd_update(struct obd_device *host, struct obd_device *watched,
1006 enum obd_notify_event ev, void *owner);
1007 int cl_get_grouplock(struct cl_object *obj, unsigned long gid, int nonblock,
1008 struct ll_grouplock *lg);
1009 void cl_put_grouplock(struct ll_grouplock *lg);
1011 /* llite/lproc_llite.c */
1012 int ll_debugfs_register_super(struct super_block *sb, const char *name);
1013 void ll_debugfs_unregister_super(struct super_block *sb);
1014 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count);
1017 LPROC_LL_READ_BYTES,
1018 LPROC_LL_WRITE_BYTES,
1045 LPROC_LL_GETXATTR_HITS,
1047 LPROC_LL_REMOVEXATTR,
1048 LPROC_LL_INODE_PERM,
1050 LPROC_LL_INODE_OCOUNT,
1051 LPROC_LL_INODE_OPCLTM,
1052 LPROC_LL_FILE_OPCODES
1056 enum get_default_layout_type {
1057 GET_DEFAULT_LAYOUT_ROOT = 1,
1060 extern const struct file_operations ll_dir_operations;
1061 extern const struct inode_operations ll_dir_inode_operations;
1062 #ifdef HAVE_DIR_CONTEXT
1063 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
1064 struct dir_context *ctx);
1066 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
1067 void *cookie, filldir_t filldir);
1069 int ll_get_mdt_idx(struct inode *inode);
1070 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid);
1071 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
1073 void ll_release_page(struct inode *inode, struct page *page, bool remove);
1074 int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl);
1077 extern const struct inode_operations ll_special_inode_operations;
1079 struct inode *ll_iget(struct super_block *sb, ino_t hash,
1080 struct lustre_md *lic);
1081 int ll_test_inode_by_fid(struct inode *inode, void *opaque);
1082 int ll_md_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1083 void *data, int flag);
1084 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de);
1085 int ll_rmdir_entry(struct inode *dir, char *name, int namelen);
1086 void ll_update_times(struct ptlrpc_request *request, struct inode *inode);
1089 int ll_writepage(struct page *page, struct writeback_control *wbc);
1090 int ll_writepages(struct address_space *, struct writeback_control *wbc);
1091 int ll_readpage(struct file *file, struct page *page);
1092 int ll_io_read_page(const struct lu_env *env, struct cl_io *io,
1093 struct cl_page *page, struct file *file);
1094 void ll_readahead_init(struct inode *inode, struct ll_readahead_state *ras);
1095 int vvp_io_write_commit(const struct lu_env *env, struct cl_io *io);
1098 void ll_cl_add(struct file *file, const struct lu_env *env, struct cl_io *io,
1099 enum lcc_type type);
1100 void ll_cl_remove(struct file *file, const struct lu_env *env);
1101 struct ll_cl_context *ll_cl_find(struct file *file);
1103 extern const struct address_space_operations ll_aops;
1106 extern const struct inode_operations ll_file_inode_operations;
1107 const struct file_operations *ll_select_file_operations(struct ll_sb_info *sbi);
1108 extern int ll_have_md_lock(struct inode *inode, __u64 *bits,
1109 enum ldlm_mode l_req_mode);
1110 extern enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
1111 struct lustre_handle *lockh, __u64 flags,
1112 enum ldlm_mode mode);
1114 int ll_file_open(struct inode *inode, struct file *file);
1115 int ll_file_release(struct inode *inode, struct file *file);
1116 int ll_release_openhandle(struct dentry *, struct lookup_intent *);
1117 int ll_md_real_close(struct inode *inode, fmode_t fmode);
1118 void ll_track_file_opens(struct inode *inode);
1119 extern void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1120 struct ll_file_data *file, loff_t pos,
1121 size_t count, int rw);
1122 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
1123 int ll_getattr(const struct path *path, struct kstat *stat,
1124 u32 request_mask, unsigned int flags);
1126 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat);
1128 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
1129 unsigned int flags, bool foreign);
1130 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1131 struct posix_acl *ll_get_acl(struct inode *inode, int type);
1132 int ll_set_acl(struct inode *inode, struct posix_acl *acl, int type);
1133 #else /* !CONFIG_LUSTRE_FS_POSIX_ACL */
1134 #define ll_get_acl NULL
1135 #define ll_set_acl NULL
1136 #endif /* CONFIG_LUSTRE_FS_POSIX_ACL */
1138 static inline int ll_xflags_to_inode_flags(int xflags)
1140 return ((xflags & FS_XFLAG_SYNC) ? S_SYNC : 0) |
1141 ((xflags & FS_XFLAG_NOATIME) ? S_NOATIME : 0) |
1142 ((xflags & FS_XFLAG_APPEND) ? S_APPEND : 0) |
1143 ((xflags & FS_XFLAG_IMMUTABLE) ? S_IMMUTABLE : 0);
1146 static inline int ll_inode_flags_to_xflags(int inode_flags)
1148 return ((inode_flags & S_SYNC) ? FS_XFLAG_SYNC : 0) |
1149 ((inode_flags & S_NOATIME) ? FS_XFLAG_NOATIME : 0) |
1150 ((inode_flags & S_APPEND) ? FS_XFLAG_APPEND : 0) |
1151 ((inode_flags & S_IMMUTABLE) ? FS_XFLAG_IMMUTABLE : 0);
1154 int ll_migrate(struct inode *parent, struct file *file,
1155 struct lmv_user_md *lum, const char *name);
1156 int ll_get_fid_by_name(struct inode *parent, const char *name,
1157 int namelen, struct lu_fid *fid, struct inode **inode);
1158 int ll_inode_permission(struct inode *inode, int mask);
1159 int ll_ioctl_check_project(struct inode *inode, __u32 xflags, __u32 projid);
1160 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
1162 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
1164 int ll_ioctl_project(struct file *file, unsigned int cmd,
1167 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
1168 __u64 flags, struct lov_user_md *lum,
1170 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1171 struct lov_mds_md **lmm, int *lmm_size,
1172 struct ptlrpc_request **request);
1173 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
1175 int ll_dir_getstripe_default(struct inode *inode, void **lmmp,
1176 int *lmm_size, struct ptlrpc_request **request,
1177 struct ptlrpc_request **root_request, u64 valid);
1178 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
1179 struct ptlrpc_request **request, u64 valid);
1180 int ll_fsync(struct file *file, loff_t start, loff_t end, int data);
1181 int ll_merge_attr(const struct lu_env *env, struct inode *inode);
1182 int ll_fid2path(struct inode *inode, void __user *arg);
1183 int ll_data_version(struct inode *inode, __u64 *data_version, int flags);
1184 int ll_hsm_release(struct inode *inode);
1185 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss);
1186 void ll_io_set_mirror(struct cl_io *io, const struct file *file);
1188 /* llite/dcache.c */
1190 int ll_d_init(struct dentry *de);
1191 extern const struct dentry_operations ll_d_ops;
1192 void ll_intent_drop_lock(struct lookup_intent *);
1193 void ll_intent_release(struct lookup_intent *);
1194 void ll_prune_aliases(struct inode *inode);
1195 void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry);
1196 int ll_revalidate_it_finish(struct ptlrpc_request *request,
1197 struct lookup_intent *it, struct dentry *de);
1199 /* llite/llite_lib.c */
1200 extern const struct super_operations lustre_super_operations;
1202 void ll_lli_init(struct ll_inode_info *lli);
1203 int ll_fill_super(struct super_block *sb);
1204 void ll_put_super(struct super_block *sb);
1205 void ll_kill_super(struct super_block *sb);
1206 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock);
1207 void ll_dir_clear_lsm_md(struct inode *inode);
1208 void ll_clear_inode(struct inode *inode);
1209 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1210 enum op_xvalid xvalid, bool hsm_import);
1211 int ll_setattr(struct dentry *de, struct iattr *attr);
1212 int ll_statfs(struct dentry *de, struct kstatfs *sfs);
1213 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1215 int ll_update_inode(struct inode *inode, struct lustre_md *md);
1216 void ll_update_inode_flags(struct inode *inode, unsigned int ext_flags);
1217 void ll_update_dir_depth(struct inode *dir, struct inode *inode);
1218 int ll_read_inode2(struct inode *inode, void *opaque);
1219 void ll_truncate_inode_pages_final(struct inode *inode);
1220 void ll_delete_inode(struct inode *inode);
1221 int ll_iocontrol(struct inode *inode, struct file *file,
1222 unsigned int cmd, unsigned long arg);
1223 int ll_flush_ctx(struct inode *inode);
1224 void ll_umount_begin(struct super_block *sb);
1225 int ll_remount_fs(struct super_block *sb, int *flags, char *data);
1226 int ll_show_options(struct seq_file *seq, struct dentry *dentry);
1227 void ll_dirty_page_discard_warn(struct page *page, int ioret);
1228 int ll_prep_inode(struct inode **inode, struct req_capsule *pill,
1229 struct super_block *sb, struct lookup_intent *it);
1230 int ll_obd_statfs(struct inode *inode, void __user *arg);
1231 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *max_mdsize);
1232 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *default_mdsize);
1233 int ll_set_default_mdsize(struct ll_sb_info *sbi, int default_mdsize);
1235 void ll_unlock_md_op_lsm(struct md_op_data *op_data);
1236 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
1237 struct inode *i1, struct inode *i2,
1238 const char *name, size_t namelen,
1239 __u32 mode, enum md_op_code opc,
1241 void ll_finish_md_op_data(struct md_op_data *op_data);
1242 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg);
1243 void ll_compute_rootsquash_state(struct ll_sb_info *sbi);
1244 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
1245 struct lov_user_md **kbuf);
1246 void ll_open_cleanup(struct super_block *sb, struct req_capsule *pill);
1248 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req);
1250 /* Compute expected user md size when passing in a md from user space */
1251 static inline ssize_t ll_lov_user_md_size(const struct lov_user_md *lum)
1253 switch (lum->lmm_magic) {
1254 case LOV_USER_MAGIC_V1:
1255 return sizeof(struct lov_user_md_v1);
1256 case LOV_USER_MAGIC_V3:
1257 return sizeof(struct lov_user_md_v3);
1258 case LOV_USER_MAGIC_SPECIFIC:
1259 if (lum->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
1262 return lov_user_md_size(lum->lmm_stripe_count,
1263 LOV_USER_MAGIC_SPECIFIC);
1264 case LOV_USER_MAGIC_COMP_V1:
1265 return ((struct lov_comp_md_v1 *)lum)->lcm_size;
1266 case LOV_USER_MAGIC_FOREIGN:
1267 return foreign_size(lum);
1273 /* llite/llite_nfs.c */
1274 extern const struct export_operations lustre_export_operations;
1275 __u32 get_uuid2int(const char *name, int len);
1276 struct inode *search_inode_for_lustre(struct super_block *sb,
1277 const struct lu_fid *fid);
1278 int ll_dir_get_parent_fid(struct inode *dir, struct lu_fid *parent_fid);
1280 /* llite/symlink.c */
1281 extern const struct inode_operations ll_fast_symlink_inode_operations;
1284 * IO arguments for various VFS I/O interfaces.
1286 struct vvp_io_args {
1287 /** normal/sendfile/splice */
1290 struct kiocb *via_iocb;
1291 struct iov_iter *via_iter;
1301 struct ll_cl_context {
1302 struct list_head lcc_list;
1304 const struct lu_env *lcc_env;
1305 struct cl_io *lcc_io;
1306 struct cl_page *lcc_page;
1307 enum lcc_type lcc_type;
1310 struct ll_thread_info {
1311 struct vvp_io_args lti_args;
1312 struct ra_io_arg lti_ria;
1313 struct ll_cl_context lti_io_ctx;
1316 extern struct lu_context_key ll_thread_key;
1318 static inline struct ll_thread_info *ll_env_info(const struct lu_env *env)
1320 struct ll_thread_info *lti;
1322 lti = lu_context_key_get(&env->le_ctx, &ll_thread_key);
1323 LASSERT(lti != NULL);
1328 static inline struct vvp_io_args *ll_env_args(const struct lu_env *env)
1330 return &ll_env_info(env)->lti_args;
1333 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1334 struct vvp_io_args *args);
1336 /* llite/llite_mmap.c */
1338 int ll_file_mmap(struct file * file, struct vm_area_struct * vma);
1339 void policy_from_vma(union ldlm_policy_data *policy, struct vm_area_struct *vma,
1340 unsigned long addr, size_t count);
1341 struct vm_area_struct *our_vma(struct mm_struct *mm, unsigned long addr,
1344 #define ll_s2sbi(sb) (s2lsi(sb)->lsi_llsbi)
1346 /* don't need an addref as the sb_info should be holding one */
1347 static inline struct obd_export *ll_s2dtexp(struct super_block *sb)
1349 return ll_s2sbi(sb)->ll_dt_exp;
1352 /* don't need an addref as the sb_info should be holding one */
1353 static inline struct obd_export *ll_s2mdexp(struct super_block *sb)
1355 return ll_s2sbi(sb)->ll_md_exp;
1358 static inline struct client_obd *sbi2mdc(struct ll_sb_info *sbi)
1360 struct obd_device *obd = sbi->ll_md_exp->exp_obd;
1366 // FIXME: replace the name of this with LL_SB to conform to kernel stuff
1367 static inline struct ll_sb_info *ll_i2sbi(struct inode *inode)
1369 return ll_s2sbi(inode->i_sb);
1372 static inline struct obd_export *ll_i2dtexp(struct inode *inode)
1374 return ll_s2dtexp(inode->i_sb);
1377 static inline struct obd_export *ll_i2mdexp(struct inode *inode)
1379 return ll_s2mdexp(inode->i_sb);
1382 static inline struct lu_fid *ll_inode2fid(struct inode *inode)
1386 LASSERT(inode != NULL);
1387 fid = &ll_i2info(inode)->lli_fid;
1392 static inline bool ll_dir_striped(struct inode *inode)
1395 return S_ISDIR(inode->i_mode) &&
1396 lmv_dir_striped(ll_i2info(inode)->lli_lsm_md);
1399 static inline loff_t ll_file_maxbytes(struct inode *inode)
1401 struct cl_object *obj = ll_i2info(inode)->lli_clob;
1404 return MAX_LFS_FILESIZE;
1406 return min_t(loff_t, cl_object_maxbytes(obj), MAX_LFS_FILESIZE);
1410 extern const struct xattr_handler *ll_xattr_handlers[];
1412 #define XATTR_USER_T 1
1413 #define XATTR_TRUSTED_T 2
1414 #define XATTR_SECURITY_T 3
1415 #define XATTR_ACL_ACCESS_T 4
1416 #define XATTR_ACL_DEFAULT_T 5
1417 #define XATTR_LUSTRE_T 6
1418 #define XATTR_OTHER_T 7
1420 ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size);
1421 int ll_xattr_list(struct inode *inode, const char *name, int type,
1422 void *buffer, size_t size, u64 valid);
1423 const struct xattr_handler *get_xattr_type(const char *name);
1426 * Common IO arguments for various VFS I/O interfaces.
1428 int cl_sb_init(struct super_block *sb);
1429 int cl_sb_fini(struct super_block *sb);
1431 enum ras_update_flags {
1435 void ll_ra_count_put(struct ll_sb_info *sbi, unsigned long len);
1436 void ll_ra_stats_inc(struct inode *inode, enum ra_stat which);
1440 #define LL_SA_RPC_MIN 2
1441 #define LL_SA_RPC_DEF 32
1442 #define LL_SA_RPC_MAX 512
1444 /* XXX: If want to support more concurrent statahead instances,
1445 * please consider to decentralize the RPC lists attached
1446 * on related import, such as imp_{sending,delayed}_list.
1448 #define LL_SA_RUNNING_MAX 256
1449 #define LL_SA_RUNNING_DEF 16
1451 #define LL_SA_CACHE_BIT 5
1452 #define LL_SA_CACHE_SIZE (1 << LL_SA_CACHE_BIT)
1453 #define LL_SA_CACHE_MASK (LL_SA_CACHE_SIZE - 1)
1455 /* per inode struct, for dir only */
1456 struct ll_statahead_info {
1457 struct dentry *sai_dentry;
1458 atomic_t sai_refcount; /* when access this struct, hold
1460 unsigned int sai_max; /* max ahead of lookup */
1461 __u64 sai_sent; /* stat requests sent count */
1462 __u64 sai_replied; /* stat requests which received
1464 __u64 sai_index; /* index of statahead entry */
1465 __u64 sai_index_wait; /* index of entry which is the
1466 * caller is waiting for */
1467 __u64 sai_hit; /* hit count */
1468 __u64 sai_miss; /* miss count:
1469 * for "ls -al" case, includes
1470 * hidden dentry miss;
1471 * for "ls -l" case, it does not
1472 * include hidden dentry miss.
1473 * "sai_miss_hidden" is used for
1476 unsigned int sai_consecutive_miss; /* consecutive miss */
1477 unsigned int sai_miss_hidden;/* "ls -al", but first dentry
1478 * is not a hidden one */
1479 unsigned int sai_skip_hidden;/* skipped hidden dentry count
1481 unsigned int sai_ls_all:1, /* "ls -al", do stat-ahead for
1483 sai_in_readpage:1;/* statahead is in readdir()*/
1484 wait_queue_head_t sai_waitq; /* stat-ahead wait queue */
1485 struct task_struct *sai_task; /* stat-ahead thread */
1486 struct task_struct *sai_agl_task; /* AGL thread */
1487 struct list_head sai_interim_entries; /* entries which got async
1488 * stat reply, but not
1490 struct list_head sai_entries; /* completed entries */
1491 struct list_head sai_agls; /* AGLs to be sent */
1492 struct list_head sai_cache[LL_SA_CACHE_SIZE];
1493 spinlock_t sai_cache_lock[LL_SA_CACHE_SIZE];
1494 atomic_t sai_cache_count; /* entry count in cache */
1497 int ll_revalidate_statahead(struct inode *dir, struct dentry **dentry,
1499 int ll_start_statahead(struct inode *dir, struct dentry *dentry, bool agl);
1500 void ll_authorize_statahead(struct inode *dir, void *key);
1501 void ll_deauthorize_statahead(struct inode *dir, void *key);
1504 blkcnt_t dirty_cnt(struct inode *inode);
1506 int cl_glimpse_size0(struct inode *inode, int agl);
1507 int cl_glimpse_lock(const struct lu_env *env, struct cl_io *io,
1508 struct inode *inode, struct cl_object *clob, int agl);
1510 static inline int cl_glimpse_size(struct inode *inode)
1512 return cl_glimpse_size0(inode, 0);
1515 /* AGL is 'asychronous glimpse lock', which is a speculative lock taken as
1516 * part of statahead */
1517 static inline int cl_agl(struct inode *inode)
1519 return cl_glimpse_size0(inode, 1);
1522 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise);
1524 int cl_io_get(struct inode *inode, struct lu_env **envout,
1525 struct cl_io **ioout, __u16 *refcheck);
1527 static inline int ll_glimpse_size(struct inode *inode)
1529 struct ll_inode_info *lli = ll_i2info(inode);
1532 down_read(&lli->lli_glimpse_sem);
1533 rc = cl_glimpse_size(inode);
1534 lli->lli_glimpse_time = ktime_get();
1535 up_read(&lli->lli_glimpse_sem);
1539 /* dentry may statahead when statahead is enabled and current process has opened
1540 * parent directory, and this dentry hasn't accessed statahead cache before */
1542 dentry_may_statahead(struct inode *dir, struct dentry *dentry)
1544 struct ll_inode_info *lli;
1545 struct ll_dentry_data *ldd;
1547 if (ll_i2sbi(dir)->ll_sa_max == 0)
1550 lli = ll_i2info(dir);
1552 /* statahead is not allowed for this dir, there may be three causes:
1553 * 1. dir is not opened.
1554 * 2. statahead hit ratio is too low.
1555 * 3. previous stat started statahead thread failed. */
1556 if (!lli->lli_sa_enabled)
1559 /* not the same process, don't statahead */
1560 if (lli->lli_opendir_pid != current->pid)
1564 * When stating a dentry, kernel may trigger 'revalidate' or 'lookup'
1565 * multiple times, eg. for 'getattr', 'getxattr' and etc.
1566 * For patchless client, lookup intent is not accurate, which may
1567 * misguide statahead. For example:
1568 * The 'revalidate' call for 'getattr' and 'getxattr' of a dentry will
1569 * have the same intent -- IT_GETATTR, while one dentry should access
1570 * statahead cache once, otherwise statahead windows is messed up.
1571 * The solution is as following:
1572 * Assign 'lld_sa_generation' with 'lli_sa_generation' when a dentry
1573 * IT_GETATTR for the first time, and subsequent IT_GETATTR will
1574 * bypass interacting with statahead cache by checking
1575 * 'lld_sa_generation == lli->lli_sa_generation'.
1577 ldd = ll_d2d(dentry);
1578 if (ldd != NULL && lli->lli_sa_generation &&
1579 ldd->lld_sa_generation == lli->lli_sa_generation)
1585 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
1586 enum cl_fsync_mode mode, int ignore_layout);
1588 static inline int ll_file_nolock(const struct file *file)
1590 struct ll_file_data *fd = file->private_data;
1591 struct inode *inode = file_inode((struct file *)file);
1593 LASSERT(fd != NULL);
1594 return ((fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
1595 (ll_i2sbi(inode)->ll_flags & LL_SBI_NOLCK));
1598 static inline void ll_set_lock_data(struct obd_export *exp, struct inode *inode,
1599 struct lookup_intent *it, __u64 *bits)
1601 if (!it->it_lock_set) {
1602 struct lustre_handle handle;
1604 /* If this inode is a remote object, it will get two
1605 * separate locks in different namespaces, Master MDT,
1606 * where the name entry is, will grant LOOKUP lock,
1607 * remote MDT, where the object is, will grant
1608 * UPDATE|PERM lock. The inode will be attched to both
1609 * LOOKUP and PERM locks, so revoking either locks will
1610 * case the dcache being cleared */
1611 if (it->it_remote_lock_mode) {
1612 handle.cookie = it->it_remote_lock_handle;
1613 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID
1614 "(%p) for remote lock %#llx\n",
1615 PFID(ll_inode2fid(inode)), inode,
1617 md_set_lock_data(exp, &handle, inode, NULL);
1620 handle.cookie = it->it_lock_handle;
1622 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID"(%p)"
1623 " for lock %#llx\n",
1624 PFID(ll_inode2fid(inode)), inode, handle.cookie);
1626 md_set_lock_data(exp, &handle, inode, &it->it_lock_bits);
1627 it->it_lock_set = 1;
1631 *bits = it->it_lock_bits;
1634 static inline int d_lustre_invalid(const struct dentry *dentry)
1636 struct ll_dentry_data *lld = ll_d2d(dentry);
1638 return (lld == NULL) || lld->lld_invalid;
1641 static inline void __d_lustre_invalidate(struct dentry *dentry)
1643 struct ll_dentry_data *lld = ll_d2d(dentry);
1646 lld->lld_invalid = 1;
1650 * Mark dentry INVALID, if dentry refcount is zero (this is normally case for
1651 * ll_md_blocking_ast), it will be pruned by ll_prune_aliases() and
1652 * ll_prune_negative_children(); otherwise dput() of the last refcount will
1653 * unhash this dentry and kill it.
1655 static inline void d_lustre_invalidate(struct dentry *dentry)
1657 CDEBUG(D_DENTRY, "invalidate dentry %pd (%p) parent %p inode %p refc %d\n",
1659 dentry->d_parent, dentry->d_inode, ll_d_count(dentry));
1661 spin_lock(&dentry->d_lock);
1662 __d_lustre_invalidate(dentry);
1663 spin_unlock(&dentry->d_lock);
1666 static inline void d_lustre_revalidate(struct dentry *dentry)
1668 spin_lock(&dentry->d_lock);
1669 LASSERT(ll_d2d(dentry) != NULL);
1670 ll_d2d(dentry)->lld_invalid = 0;
1671 spin_unlock(&dentry->d_lock);
1674 static inline dev_t ll_compat_encode_dev(dev_t dev)
1676 /* The compat_sys_*stat*() syscalls will fail unless the
1677 * device majors and minors are both less than 256. Note that
1678 * the value returned here will be passed through
1679 * old_encode_dev() in cp_compat_stat(). And so we are not
1680 * trying to return a valid compat (u16) device number, just
1681 * one that will pass the old_valid_dev() check. */
1683 return MKDEV(MAJOR(dev) & 0xff, MINOR(dev) & 0xff);
1686 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf);
1687 int ll_layout_refresh(struct inode *inode, __u32 *gen);
1688 int ll_layout_restore(struct inode *inode, loff_t start, __u64 length);
1689 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
1690 struct lu_extent *ext);
1692 int ll_xattr_init(void);
1693 void ll_xattr_fini(void);
1695 int ll_page_sync_io(const struct lu_env *env, struct cl_io *io,
1696 struct cl_page *page, enum cl_req_type crt);
1698 int ll_getparent(struct file *file, struct getparent __user *arg);
1701 int cl_setattr_ost(struct cl_object *obj, const struct iattr *attr,
1702 enum op_xvalid xvalid, unsigned int attr_flags);
1704 extern struct lu_env *cl_inode_fini_env;
1705 extern __u16 cl_inode_fini_refcheck;
1707 int cl_file_inode_init(struct inode *inode, struct lustre_md *md);
1708 void cl_inode_fini(struct inode *inode);
1710 u64 cl_fid_build_ino(const struct lu_fid *fid, int api32);
1711 u32 cl_fid_build_gen(const struct lu_fid *fid);
1713 static inline struct pcc_super *ll_i2pccs(struct inode *inode)
1715 return &ll_i2sbi(inode)->ll_pcc_super;
1718 static inline struct pcc_super *ll_info2pccs(struct ll_inode_info *lli)
1720 return ll_i2pccs(ll_info2i(lli));
1723 #ifdef HAVE_LUSTRE_CRYPTO
1725 extern const struct llcrypt_operations lustre_cryptops;
1727 /* llite/llite_foreign.c */
1728 int ll_manage_foreign(struct inode *inode, struct lustre_md *lmd);
1729 bool ll_foreign_is_openable(struct dentry *dentry, unsigned int flags);
1730 bool ll_foreign_is_removable(struct dentry *dentry, bool unset);
1732 #endif /* LLITE_INTERNAL_H */