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 <linux/parser.h>
46 #include <lustre_compat.h>
47 #include <lustre_crypto.h>
48 #include <range_lock.h>
50 #include "vvp_internal.h"
52 #include "foreign_symlink.h"
58 #ifndef HAVE_VM_FAULT_RETRY
59 #define VM_FAULT_RETRY 0
62 /* Kernel 3.1 kills LOOKUP_CONTINUE, LOOKUP_PARENT is equivalent to it.
63 * seem kernel commit 49084c3bb2055c401f3493c13edae14d49128ca0 */
64 #ifndef LOOKUP_CONTINUE
65 #define LOOKUP_CONTINUE LOOKUP_PARENT
68 /** Only used on client-side for indicating the tail of dir hash/offset. */
69 #define LL_DIR_END_OFF 0x7fffffffffffffffULL
70 #define LL_DIR_END_OFF_32BIT 0x7fffffffUL
72 /* 4UL * 1024 * 1024 */
73 #define LL_MAX_BLKSIZE_BITS 22
75 #define LL_IT2STR(it) ((it) ? ldlm_it2str((it)->it_op) : "0")
77 #define TIMES_SET_FLAGS (ATTR_MTIME_SET | ATTR_ATIME_SET | ATTR_TIMES_SET)
79 struct ll_dentry_data {
80 unsigned int lld_sa_generation;
81 unsigned int lld_invalid:1;
82 unsigned int lld_nfs_dentry:1;
83 struct rcu_head lld_rcu_head;
86 #define ll_d2d(de) ((struct ll_dentry_data*)((de)->d_fsdata))
88 #define LLI_INODE_MAGIC 0x111d0de5
89 #define LLI_INODE_DEAD 0xdeadd00d
91 struct ll_getname_data {
92 #ifdef HAVE_DIR_CONTEXT
93 struct dir_context ctx;
95 char *lgd_name; /* points to a buffer with NAME_MAX+1 size */
96 struct lu_fid lgd_fid; /* target fid we are looking for */
97 int lgd_found; /* inode matched? */
100 struct ll_grouplock {
101 struct lu_env *lg_env;
103 struct cl_lock *lg_lock;
104 unsigned long lg_gid;
107 /* See comment on trunc_sem_down_read_nowait */
108 struct ll_trunc_sem {
109 /* when positive, this is a count of readers, when -1, it indicates
110 * the semaphore is held for write, and 0 is unlocked
112 atomic_t ll_trunc_readers;
113 /* this tracks a count of waiting writers */
114 atomic_t ll_trunc_waiters;
117 struct ll_inode_info {
118 __u32 lli_inode_magic;
121 volatile unsigned long lli_flags;
122 struct posix_acl *lli_posix_acl;
124 /* identifying fields for both metadata and data stacks. */
125 struct lu_fid lli_fid;
126 /* master inode fid for stripe directory */
127 struct lu_fid lli_pfid;
129 /* We need all three because every inode may be opened in different
131 struct obd_client_handle *lli_mds_read_och;
132 struct obd_client_handle *lli_mds_write_och;
133 struct obd_client_handle *lli_mds_exec_och;
134 __u64 lli_open_fd_read_count;
135 __u64 lli_open_fd_write_count;
136 __u64 lli_open_fd_exec_count;
138 /* Number of times this inode was opened */
139 u64 lli_open_fd_count;
140 /* When last close was performed on this inode */
141 ktime_t lli_close_fd_time;
143 /* Protects access to och pointers and their usage counters */
144 struct mutex lli_och_mutex;
146 struct inode lli_vfs_inode;
148 /* the most recent timestamps obtained from mds */
153 spinlock_t lli_agl_lock;
155 /* Try to make the d::member and f::member are aligned. Before using
156 * these members, make clear whether it is directory or not. */
160 /* metadata statahead */
161 /* since parent-child threads can share the same @file
162 * struct, "opendir_key" is the token when dir close for
163 * case of parent exit before child -- it is me should
164 * cleanup the dir readahead. */
165 void *lli_opendir_key;
166 struct ll_statahead_info *lli_sai;
167 /* protect statahead stuff. */
168 spinlock_t lli_sa_lock;
169 /* "opendir_pid" is the token when lookup/revalid
170 * -- I am the owner of dir statahead. */
171 pid_t lli_opendir_pid;
172 /* directory depth to ROOT */
173 unsigned short lli_depth;
174 /* stat will try to access statahead entries or start
175 * statahead if this flag is set, and this flag will be
176 * set upon dir open, and cleared when dir is closed,
177 * statahead hit ratio is too low, or start statahead
179 unsigned short lli_sa_enabled:1;
180 /* generation for statahead */
181 unsigned int lli_sa_generation;
182 /* rw lock protects lli_lsm_md */
183 struct rw_semaphore lli_lsm_sem;
184 /* directory stripe information */
185 struct lmv_stripe_md *lli_lsm_md;
186 /* directory default LMV */
187 struct lmv_stripe_md *lli_default_lsm_md;
190 /* for non-directory */
192 struct mutex lli_size_mutex;
193 char *lli_symlink_name;
194 struct ll_trunc_sem lli_trunc_sem;
195 struct range_lock_tree lli_write_tree;
196 struct mutex lli_setattr_mutex;
198 struct rw_semaphore lli_glimpse_sem;
199 ktime_t lli_glimpse_time;
200 struct list_head lli_agl_list;
203 /* for writepage() only to communicate to fsync */
206 /* protect the file heat fields */
207 spinlock_t lli_heat_lock;
208 __u32 lli_heat_flags;
209 struct obd_heat_instance lli_heat_instances[OBD_HEAT_COUNT];
212 * Whenever a process try to read/write the file, the
213 * jobid of the process will be saved here, and it'll
214 * be packed into the write PRC when flush later.
216 * So the read/write statistics for jobid will not be
217 * accurate if the file is shared by different jobs.
219 char lli_jobid[LUSTRE_JOBID_SIZE];
221 struct mutex lli_pcc_lock;
222 enum lu_pcc_state_flags lli_pcc_state;
224 * @lli_pcc_generation saves the gobal PCC generation
225 * when the file was successfully attached into PCC.
226 * The flags of the PCC dataset are saved in
228 * The gobal PCC generation will be increased when add
229 * or delete a PCC backend, or change the configuration
230 * parameters for PCC.
231 * If @lli_pcc_generation is same as the gobal PCC
232 * generation, we can use the saved flags of the PCC
233 * dataset to determine whether need to try auto attach
236 __u64 lli_pcc_generation;
237 enum pcc_dataset_flags lli_pcc_dsflags;
238 struct pcc_inode *lli_pcc_inode;
240 struct mutex lli_group_mutex;
241 __u64 lli_group_users;
242 unsigned long lli_group_gid;
244 __u64 lli_attr_valid;
246 __u64 lli_lazyblocks;
250 /* XXX: For following frequent used members, although they maybe special
251 * used for non-directory object, it is some time-wasting to check
252 * whether the object is directory or not before using them. On the
253 * other hand, currently, sizeof(f) > sizeof(d), it cannot reduce
254 * the "ll_inode_info" size even if moving those members into u.f.
255 * So keep them out side.
257 * In the future, if more members are added only for directory,
258 * some of the following members can be moved into u.f.
260 struct cl_object *lli_clob;
262 /* mutex to request for layout lock exclusively. */
263 struct mutex lli_layout_mutex;
264 /* Layout version, protected by lli_layout_lock */
265 __u32 lli_layout_gen;
266 spinlock_t lli_layout_lock;
268 __u32 lli_projid; /* project id */
270 struct rw_semaphore lli_xattrs_list_rwsem;
271 struct mutex lli_xattrs_enq_lock;
272 struct list_head lli_xattrs; /* ll_xattr_entry->xe_list */
273 struct list_head lli_lccs; /* list of ll_cl_context */
276 static inline void ll_trunc_sem_init(struct ll_trunc_sem *sem)
278 atomic_set(&sem->ll_trunc_readers, 0);
279 atomic_set(&sem->ll_trunc_waiters, 0);
282 /* This version of down read ignores waiting writers, meaning if the semaphore
283 * is already held for read, this down_read will 'join' that reader and also
284 * take the semaphore.
286 * This lets us avoid an unusual deadlock.
288 * We must take lli_trunc_sem in read mode on entry in to various i/o paths
289 * in Lustre, in order to exclude truncates. Some of these paths then need to
290 * take the mmap_lock, while still holding the trunc_sem. The problem is that
291 * page faults hold the mmap_lock when calling in to Lustre, and then must also
292 * take the trunc_sem to exclude truncate.
294 * This means the locking order for trunc_sem and mmap_lock is sometimes AB,
295 * sometimes BA. This is almost OK because in both cases, we take the trunc
296 * sem for read, so it doesn't block.
298 * However, if a write mode user (truncate, a setattr op) arrives in the
299 * middle of this, the second reader on the truncate_sem will wait behind that
302 * So we have, on our truncate sem, in order (where 'reader' and 'writer' refer
303 * to the mode in which they take the semaphore):
304 * reader (holding mmap_lock, needs truncate_sem)
306 * reader (holding truncate sem, waiting for mmap_lock)
308 * And so the readers deadlock.
310 * The solution is this modified semaphore, where this down_read ignores
311 * waiting write operations, and all waiters are woken up at once, so readers
312 * using down_read_nowait cannot get stuck behind waiting writers, regardless
313 * of the order they arrived in.
315 * down_read_nowait is only used in the page fault case, where we already hold
316 * the mmap_lock. This is because otherwise repeated read and write operations
317 * (which take the truncate sem) could prevent a truncate from ever starting.
318 * This could still happen with page faults, but without an even more complex
319 * mechanism, this is unavoidable.
323 static inline void trunc_sem_down_read_nowait(struct ll_trunc_sem *sem)
325 wait_var_event(&sem->ll_trunc_readers,
326 atomic_inc_unless_negative(&sem->ll_trunc_readers));
329 static inline void trunc_sem_down_read(struct ll_trunc_sem *sem)
331 wait_var_event(&sem->ll_trunc_readers,
332 atomic_read(&sem->ll_trunc_waiters) == 0 &&
333 atomic_inc_unless_negative(&sem->ll_trunc_readers));
336 static inline void trunc_sem_up_read(struct ll_trunc_sem *sem)
338 if (atomic_dec_return(&sem->ll_trunc_readers) == 0 &&
339 atomic_read(&sem->ll_trunc_waiters))
340 wake_up_var(&sem->ll_trunc_readers);
343 static inline void trunc_sem_down_write(struct ll_trunc_sem *sem)
345 atomic_inc(&sem->ll_trunc_waiters);
346 wait_var_event(&sem->ll_trunc_readers,
347 atomic_cmpxchg(&sem->ll_trunc_readers, 0, -1) == 0);
348 atomic_dec(&sem->ll_trunc_waiters);
351 static inline void trunc_sem_up_write(struct ll_trunc_sem *sem)
353 atomic_set(&sem->ll_trunc_readers, 0);
354 wake_up_var(&sem->ll_trunc_readers);
357 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
358 static inline void lli_clear_acl(struct ll_inode_info *lli)
360 if (lli->lli_posix_acl) {
361 posix_acl_release(lli->lli_posix_acl);
362 lli->lli_posix_acl = NULL;
366 static inline void lli_replace_acl(struct ll_inode_info *lli,
367 struct lustre_md *md)
369 write_lock(&lli->lli_lock);
370 if (lli->lli_posix_acl)
371 posix_acl_release(lli->lli_posix_acl);
372 lli->lli_posix_acl = md->posix_acl;
373 write_unlock(&lli->lli_lock);
376 static inline void lli_clear_acl(struct ll_inode_info *lli)
380 static inline void lli_replace_acl(struct ll_inode_info *lli,
381 struct lustre_md *md)
386 static inline __u32 ll_layout_version_get(struct ll_inode_info *lli)
390 spin_lock(&lli->lli_layout_lock);
391 gen = lli->lli_layout_gen;
392 spin_unlock(&lli->lli_layout_lock);
397 static inline void ll_layout_version_set(struct ll_inode_info *lli, __u32 gen)
399 spin_lock(&lli->lli_layout_lock);
400 lli->lli_layout_gen = gen;
401 spin_unlock(&lli->lli_layout_lock);
405 /* File data is modified. */
406 LLIF_DATA_MODIFIED = 0,
407 /* File is being restored */
408 LLIF_FILE_RESTORING = 1,
409 /* Xattr cache is attached to the file */
410 LLIF_XATTR_CACHE = 2,
411 /* Project inherit */
412 LLIF_PROJECT_INHERIT = 3,
413 /* update atime from MDS even if it's older than local inode atime. */
414 LLIF_UPDATE_ATIME = 4,
415 /* foreign file/dir can be unlinked unconditionnaly */
416 LLIF_FOREIGN_REMOVABLE = 5,
417 /* setting encryption context in progress */
418 LLIF_SET_ENC_CTX = 6,
419 /* Xattr cache is filled */
420 LLIF_XATTR_CACHE_FILLED = 7,
424 int ll_xattr_cache_destroy(struct inode *inode);
425 int ll_xattr_cache_empty(struct inode *inode);
427 int ll_xattr_cache_get(struct inode *inode,
433 int ll_xattr_cache_insert(struct inode *inode,
438 static inline bool obd_connect_has_secctx(struct obd_connect_data *data)
440 #ifdef CONFIG_SECURITY
441 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
442 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
448 static inline void obd_connect_set_secctx(struct obd_connect_data *data)
450 #ifdef CONFIG_SECURITY
451 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
455 int ll_dentry_init_security(struct dentry *dentry, int mode, struct qstr *name,
456 const char **secctx_name, void **secctx,
458 int ll_inode_init_security(struct dentry *dentry, struct inode *inode,
461 int ll_listsecurity(struct inode *inode, char *secctx_name,
462 size_t secctx_name_size);
464 static inline bool obd_connect_has_enc(struct obd_connect_data *data)
466 #ifdef HAVE_LUSTRE_CRYPTO
467 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
468 data->ocd_connect_flags2 & OBD_CONNECT2_ENCRYPT;
474 static inline void obd_connect_set_enc(struct obd_connect_data *data)
476 #ifdef HAVE_LUSTRE_CRYPTO
477 data->ocd_connect_flags2 |= OBD_CONNECT2_ENCRYPT;
482 * Locking to guarantee consistency of non-atomic updates to long long i_size,
483 * consistency between file size and KMS.
485 * Implemented by ->lli_size_mutex and ->lsm_lock, nested in that order.
488 void ll_inode_size_lock(struct inode *inode);
489 void ll_inode_size_unlock(struct inode *inode);
491 static inline struct ll_inode_info *ll_i2info(struct inode *inode)
493 return container_of(inode, struct ll_inode_info, lli_vfs_inode);
496 static inline struct pcc_inode *ll_i2pcci(struct inode *inode)
498 return ll_i2info(inode)->lli_pcc_inode;
501 /* default to use at least 16M for fast read if possible */
502 #define RA_REMAIN_WINDOW_MIN MiB_TO_PAGES(16UL)
504 /* default read-ahead on a given client mountpoint. */
505 #define SBI_DEFAULT_READ_AHEAD_MAX MiB_TO_PAGES(1024UL)
507 /* default read-ahead for a single file descriptor */
508 #define SBI_DEFAULT_READ_AHEAD_PER_FILE_MAX MiB_TO_PAGES(256UL)
510 /* default read-ahead full files smaller than limit on the second read */
511 #define SBI_DEFAULT_READ_AHEAD_WHOLE_MAX MiB_TO_PAGES(2UL)
513 /* default range pages */
514 #define SBI_DEFAULT_RA_RANGE_PAGES MiB_TO_PAGES(1ULL)
516 /* Min range pages */
517 #define RA_MIN_MMAP_RANGE_PAGES 16UL
522 RA_STAT_DISTANT_READPAGE,
523 RA_STAT_MISS_IN_WINDOW,
524 RA_STAT_FAILED_GRAB_PAGE,
525 RA_STAT_FAILED_MATCH,
530 RA_STAT_MAX_IN_FLIGHT,
531 RA_STAT_WRONG_GRAB_PAGE,
532 RA_STAT_FAILED_REACH_END,
534 RA_STAT_FAILED_FAST_READ,
535 RA_STAT_MMAP_RANGE_READ,
540 atomic_t ra_cur_pages;
541 unsigned long ra_max_pages;
542 unsigned long ra_max_pages_per_file;
543 unsigned long ra_range_pages;
544 unsigned long ra_max_read_ahead_whole_pages;
545 struct workqueue_struct *ll_readahead_wq;
547 * Max number of active works could be triggered
548 * for async readahead.
550 unsigned int ra_async_max_active;
551 /* how many async readahead triggered in flight */
552 atomic_t ra_async_inflight;
553 /* Threshold to control when to trigger async readahead */
554 unsigned long ra_async_pages_per_file_threshold;
557 /* ra_io_arg will be filled in the beginning of ll_readahead with
558 * ras_lock, then the following ll_read_ahead_pages will read RA
559 * pages according to this arg, all the items in this structure are
560 * counted by page index.
563 pgoff_t ria_start_idx; /* start offset of read-ahead*/
564 pgoff_t ria_end_idx; /* end offset of read-ahead*/
565 unsigned long ria_reserved; /* reserved pages for read-ahead */
566 pgoff_t ria_end_idx_min;/* minimum end to cover current read */
567 bool ria_eof; /* reach end of file */
568 /* If stride read pattern is detected, ria_stoff is the byte offset
569 * where stride read is started. Note: for normal read-ahead, the
570 * value here is meaningless, and also it will not be accessed*/
572 /* ria_length and ria_bytes are the length and pages length in the
573 * stride I/O mode. And they will also be used to check whether
574 * it is stride I/O read-ahead in the read-ahead pages*/
579 /* LL_HIST_MAX=32 causes an overflow */
580 #define LL_HIST_MAX 28
581 #define LL_HIST_START 12 /* buckets start at 2^12 = 4k */
582 #define LL_PROCESS_HIST_MAX 10
583 struct per_process_info {
585 struct obd_histogram pp_r_hist;
586 struct obd_histogram pp_w_hist;
589 /* pp_extents[LL_PROCESS_HIST_MAX] will hold the combined process info */
590 struct ll_rw_extents_info {
592 struct per_process_info pp_extents[LL_PROCESS_HIST_MAX + 1];
595 #define LL_OFFSET_HIST_MAX 100
596 struct ll_rw_process_info {
599 loff_t rw_range_start;
601 loff_t rw_last_file_pos;
603 size_t rw_smallest_extent;
604 size_t rw_largest_extent;
605 struct ll_file_data *rw_last_file;
608 enum stats_track_type {
609 STATS_TRACK_ALL = 0, /* track all processes */
610 STATS_TRACK_PID, /* track process with this pid */
611 STATS_TRACK_PPID, /* track processes with this ppid */
612 STATS_TRACK_GID, /* track processes with this gid */
616 /* flags for sbi->ll_flags */
618 LL_SBI_NOLCK, /* DLM locking disabled directio-only */
619 LL_SBI_CHECKSUM, /* checksum each page as it's written */
620 LL_SBI_LOCALFLOCK, /* local flocks instead of fs-wide */
621 LL_SBI_FLOCK, /* flock enabled */
622 LL_SBI_USER_XATTR, /* support user xattr */
623 LL_SBI_LRU_RESIZE, /* lru resize support */
624 LL_SBI_LAZYSTATFS, /* lazystatfs mount option */
625 LL_SBI_32BIT_API, /* generate 32 bit inodes. */
626 LL_SBI_USER_FID2PATH, /* fid2path by unprivileged users */
627 LL_SBI_VERBOSE, /* verbose mount/umount */
628 LL_SBI_ALWAYS_PING, /* ping even if server suppress_pings */
629 LL_SBI_TEST_DUMMY_ENCRYPTION, /* test dummy encryption */
630 LL_SBI_ENCRYPT, /* client side encryption */
631 LL_SBI_FOREIGN_SYMLINK, /* foreign fake-symlink support */
632 LL_SBI_FOREIGN_SYMLINK_UPCALL, /* foreign fake-symlink upcall set */
633 LL_SBI_NUM_MOUNT_OPT,
635 LL_SBI_ACL, /* support ACL */
636 LL_SBI_AGL_ENABLED, /* enable agl */
637 LL_SBI_64BIT_HASH, /* support 64-bits dir hash/offset */
638 LL_SBI_LAYOUT_LOCK, /* layout lock support */
639 LL_SBI_XATTR_CACHE, /* support for xattr cache */
640 LL_SBI_NOROOTSQUASH, /* do not apply root squash */
641 LL_SBI_FAST_READ, /* fast read support */
642 LL_SBI_FILE_SECCTX, /* file security context at create */
643 LL_SBI_TINY_WRITE, /* tiny write support */
644 LL_SBI_FILE_HEAT, /* file heat support */
645 LL_SBI_PARALLEL_DIO, /* parallel (async) O_DIRECT RPCs */
649 int ll_sbi_flags_seq_show(struct seq_file *m, void *v);
651 /* This is embedded into llite super-blocks to keep track of connect
652 * flags (capabilities) supported by all imports given mount is
654 struct lustre_client_ocd {
655 /* This is conjunction of connect_flags across all imports
656 * (LOVs) this mount is connected to. This field is updated by
657 * cl_ocd_update() under ->lco_lock. */
659 struct mutex lco_lock;
660 struct obd_export *lco_md_exp;
661 struct obd_export *lco_dt_exp;
665 /* this protects pglist and ra_info. It isn't safe to
666 * grab from interrupt contexts */
668 spinlock_t ll_pp_extent_lock; /* pp_extent entry*/
669 spinlock_t ll_process_lock; /* ll_rw_process_info */
670 struct obd_uuid ll_sb_uuid;
671 struct obd_export *ll_md_exp;
672 struct obd_export *ll_dt_exp;
673 struct obd_device *ll_md_obd;
674 struct obd_device *ll_dt_obd;
675 struct dentry *ll_debugfs_entry;
676 struct lu_fid ll_root_fid; /* root object fid */
678 DECLARE_BITMAP(ll_flags, LL_SBI_NUM_FLAGS); /* enum ll_sbi_flags */
679 unsigned int ll_xattr_cache_enabled:1,
680 ll_xattr_cache_set:1, /* already set to 0/1 */
681 ll_client_common_fill_super_succeeded:1,
684 struct lustre_client_ocd ll_lco;
686 struct lprocfs_stats *ll_stats; /* lprocfs stats counter */
688 /* Used to track "unstable" pages on a client, and maintain a
689 * LRU list of clean pages. An "unstable" page is defined as
690 * any page which is sent to a server as part of a bulk request,
691 * but is uncommitted to stable storage. */
692 struct cl_client_cache *ll_cache;
694 struct lprocfs_stats *ll_ra_stats;
696 struct ll_ra_info ll_ra_info;
697 unsigned int ll_namelen;
698 const struct file_operations *ll_fop;
700 struct lu_site *ll_site;
701 struct cl_device *ll_cl;
704 struct ll_rw_extents_info ll_rw_extents_info;
705 int ll_extent_process_count;
706 unsigned int ll_offset_process_count;
707 ktime_t ll_process_stats_init;
708 struct ll_rw_process_info ll_rw_process_info[LL_PROCESS_HIST_MAX];
709 struct ll_rw_process_info ll_rw_offset_info[LL_OFFSET_HIST_MAX];
710 unsigned int ll_rw_offset_entry_count;
711 int ll_stats_track_id;
712 enum stats_track_type ll_stats_track_type;
715 /* metadata stat-ahead */
716 unsigned int ll_sa_running_max;/* max concurrent
717 * statahead instances */
718 unsigned int ll_sa_max; /* max statahead RPCs */
719 atomic_t ll_sa_total; /* statahead thread started
721 atomic_t ll_sa_wrong; /* statahead thread stopped for
723 atomic_t ll_sa_running; /* running statahead thread
725 atomic_t ll_agl_total; /* AGL thread started count */
727 dev_t ll_sdev_orig; /* save s_dev before assign for
730 struct root_squash_info ll_squash;
733 /* st_blksize returned by stat(2), when non-zero */
734 unsigned int ll_stat_blksize;
736 /* maximum relative age of cached statfs results */
737 unsigned int ll_statfs_max_age;
739 struct kset ll_kset; /* sysfs object */
740 struct completion ll_kobj_unregister;
743 unsigned int ll_heat_decay_weight;
744 unsigned int ll_heat_period_second;
746 /* Opens of the same inode before we start requesting open lock */
747 u32 ll_oc_thrsh_count;
749 /* Time in ms between last inode close and next open to be considered
750 * instant back to back and would trigger an open lock request
754 /* Time in ms after last file close that we no longer count prior opens*/
757 /* filesystem fsname */
758 char ll_fsname[LUSTRE_MAXFSNAME + 1];
760 /* Persistent Client Cache */
761 struct pcc_super ll_pcc_super;
763 /* to protect vs updates in all following foreign symlink fields */
764 struct rw_semaphore ll_foreign_symlink_sem;
765 /* foreign symlink path prefix */
766 char *ll_foreign_symlink_prefix;
767 /* full prefix size including leading '\0' */
768 size_t ll_foreign_symlink_prefix_size;
769 /* foreign symlink path upcall */
770 char *ll_foreign_symlink_upcall;
771 /* foreign symlink path upcall infos */
772 struct ll_foreign_symlink_upcall_item *ll_foreign_symlink_upcall_items;
773 /* foreign symlink path upcall nb infos */
774 unsigned int ll_foreign_symlink_upcall_nb_items;
777 #define SBI_DEFAULT_HEAT_DECAY_WEIGHT ((80 * 256 + 50) / 100)
778 #define SBI_DEFAULT_HEAT_PERIOD_SECOND (60)
780 #define SBI_DEFAULT_OPENCACHE_THRESHOLD_COUNT (5)
781 #define SBI_DEFAULT_OPENCACHE_THRESHOLD_MS (100) /* 0.1 second */
782 #define SBI_DEFAULT_OPENCACHE_THRESHOLD_MAX_MS (60000) /* 1 minute */
785 * per file-descriptor read-ahead data.
787 struct ll_readahead_state {
789 /* End byte that read(2) try to read. */
790 loff_t ras_last_read_end_bytes;
792 * number of bytes read after last read-ahead window reset. As window
793 * is reset on each seek, this is effectively a number of consecutive
794 * accesses. Maybe ->ras_accessed_in_window is better name.
796 * XXX nikita: window is also reset (by ras_update()) when Lustre
797 * believes that memory pressure evicts read-ahead pages. In that
798 * case, it probably doesn't make sense to expand window to
799 * PTLRPC_MAX_BRW_PAGES on the third access.
801 loff_t ras_consecutive_bytes;
803 * number of read requests after the last read-ahead window reset
804 * As window is reset on each seek, this is effectively the number
805 * on consecutive read request and is used to trigger read-ahead.
807 unsigned long ras_consecutive_requests;
809 * Parameters of current read-ahead window. Handled by
810 * ras_update(). On the initial access to the file or after a seek,
811 * window is reset to 0. After 3 consecutive accesses, window is
812 * expanded to PTLRPC_MAX_BRW_PAGES. Afterwards, window is enlarged by
813 * PTLRPC_MAX_BRW_PAGES chunks up to ->ra_max_pages.
815 pgoff_t ras_window_start_idx;
816 pgoff_t ras_window_pages;
818 /* Page index where min range read starts */
819 pgoff_t ras_range_min_start_idx;
820 /* Page index where mmap range read ends */
821 pgoff_t ras_range_max_end_idx;
822 /* number of mmap pages where last time detected */
823 pgoff_t ras_last_range_pages;
824 /* number of mmap range requests */
825 pgoff_t ras_range_requests;
828 * Optimal RPC size in pages.
829 * It decides how many pages will be sent for each read-ahead.
831 unsigned long ras_rpc_pages;
833 * Where next read-ahead should start at. This lies within read-ahead
834 * window. Read-ahead window is read in pieces rather than at once
835 * because: 1. lustre limits total number of pages under read-ahead by
836 * ->ra_max_pages (see ll_ra_count_get()), 2. client cannot read pages
837 * not covered by DLM lock.
839 pgoff_t ras_next_readahead_idx;
841 * Total number of ll_file_read requests issued, reads originating
842 * due to mmap are not counted in this total. This value is used to
843 * trigger full file read-ahead after multiple reads to a small file.
845 unsigned long ras_requests;
847 * The following 3 items are used for detecting the stride I/O
849 * In stride I/O mode,
850 * ...............|-----data-----|****gap*****|--------|******|....
851 * offset |-stride_bytes-|-stride_gap-|
852 * ras_stride_offset = offset;
853 * ras_stride_length = stride_bytes + stride_gap;
854 * ras_stride_bytes = stride_bytes;
855 * Note: all these three items are counted by bytes.
857 loff_t ras_stride_offset;
858 loff_t ras_stride_length;
859 loff_t ras_stride_bytes;
861 * number of consecutive stride request count, and it is similar as
862 * ras_consecutive_requests, but used for stride I/O mode.
863 * Note: only more than 2 consecutive stride request are detected,
864 * stride read-ahead will be enable
866 unsigned long ras_consecutive_stride_requests;
867 /* index of the last page that async readahead starts */
868 pgoff_t ras_async_last_readpage_idx;
869 /* whether we should increase readahead window */
870 bool ras_need_increase_window;
871 /* whether ra miss check should be skipped */
872 bool ras_no_miss_check;
875 struct ll_readahead_work {
876 /** File to readahead */
877 struct file *lrw_file;
878 pgoff_t lrw_start_idx;
881 /* async worker to handler read */
882 struct work_struct lrw_readahead_work;
883 char lrw_jobid[LUSTRE_JOBID_SIZE];
886 extern struct kmem_cache *ll_file_data_slab;
887 struct lustre_handle;
888 struct ll_file_data {
889 struct ll_readahead_state fd_ras;
890 struct ll_grouplock fd_grouplock;
894 /* openhandle if lease exists for this file.
895 * Borrow lli->lli_och_mutex to protect assignment */
896 struct obd_client_handle *fd_lease_och;
897 struct obd_client_handle *fd_och;
898 struct file *fd_file;
899 /* Indicate whether need to report failure when close.
900 * true: failure is known, not report again.
901 * false: unknown failure, should report. */
902 bool fd_write_failed;
903 bool ll_lock_no_expand;
904 /* Used by mirrored file to lead IOs to a specific mirror, usually
905 * for mirror resync. 0 means default. */
906 __u32 fd_designated_mirror;
907 /* The layout version when resync starts. Resync I/O should carry this
908 * layout version for verification to OST objects */
909 __u32 fd_layout_version;
910 struct pcc_file fd_pcc_file;
913 void llite_tunables_unregister(void);
914 int llite_tunables_register(void);
916 static inline struct inode *ll_info2i(struct ll_inode_info *lli)
918 return &lli->lli_vfs_inode;
921 __u32 ll_i2suppgid(struct inode *i);
922 void ll_i2gids(__u32 *suppgids, struct inode *i1,struct inode *i2);
924 static inline int ll_need_32bit_api(struct ll_sb_info *sbi)
926 #if BITS_PER_LONG == 32
928 #elif defined(CONFIG_COMPAT)
929 if (unlikely(test_bit(LL_SBI_32BIT_API, sbi->ll_flags)))
932 # ifdef CONFIG_X86_X32
933 /* in_compat_syscall() returns true when called from a kthread
934 * and CONFIG_X86_X32 is enabled, which is wrong. So check
935 * whether the caller comes from a syscall (ie. not a kthread)
936 * before calling in_compat_syscall(). */
937 if (current->flags & PF_KTHREAD)
941 return unlikely(in_compat_syscall());
943 return unlikely(test_bit(LL_SBI_32BIT_API, sbi->ll_flags));
947 static inline bool ll_sbi_has_fast_read(struct ll_sb_info *sbi)
949 return test_bit(LL_SBI_FAST_READ, sbi->ll_flags);
952 static inline bool ll_sbi_has_tiny_write(struct ll_sb_info *sbi)
954 return test_bit(LL_SBI_TINY_WRITE, sbi->ll_flags);
957 static inline bool ll_sbi_has_file_heat(struct ll_sb_info *sbi)
959 return test_bit(LL_SBI_FILE_HEAT, sbi->ll_flags);
962 static inline bool ll_sbi_has_foreign_symlink(struct ll_sb_info *sbi)
964 return test_bit(LL_SBI_FOREIGN_SYMLINK, sbi->ll_flags);
967 static inline bool ll_sbi_has_parallel_dio(struct ll_sb_info *sbi)
969 return test_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags);
972 void ll_ras_enter(struct file *f, loff_t pos, size_t count);
974 /* llite/lcommon_misc.c */
975 int cl_ocd_update(struct obd_device *host, struct obd_device *watched,
976 enum obd_notify_event ev, void *owner);
977 int cl_get_grouplock(struct cl_object *obj, unsigned long gid, int nonblock,
978 struct ll_grouplock *lg);
979 void cl_put_grouplock(struct ll_grouplock *lg);
981 /* llite/lproc_llite.c */
982 int ll_debugfs_register_super(struct super_block *sb, const char *name);
983 void ll_debugfs_unregister_super(struct super_block *sb);
984 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count);
988 LPROC_LL_WRITE_BYTES,
1015 LPROC_LL_GETXATTR_HITS,
1017 LPROC_LL_REMOVEXATTR,
1018 LPROC_LL_INODE_PERM,
1020 LPROC_LL_INODE_OCOUNT,
1021 LPROC_LL_INODE_OPCLTM,
1022 LPROC_LL_FILE_OPCODES
1026 enum get_default_layout_type {
1027 GET_DEFAULT_LAYOUT_ROOT = 1,
1030 extern const struct file_operations ll_dir_operations;
1031 extern const struct inode_operations ll_dir_inode_operations;
1032 #ifdef HAVE_DIR_CONTEXT
1033 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
1034 struct dir_context *ctx);
1036 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
1037 void *cookie, filldir_t filldir);
1039 int ll_get_mdt_idx(struct inode *inode);
1040 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid);
1041 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
1043 void ll_release_page(struct inode *inode, struct page *page, bool remove);
1044 int quotactl_ioctl(struct super_block *sb, struct if_quotactl *qctl);
1047 extern const struct inode_operations ll_special_inode_operations;
1049 struct inode *ll_iget(struct super_block *sb, ino_t hash,
1050 struct lustre_md *lic);
1051 int ll_test_inode_by_fid(struct inode *inode, void *opaque);
1052 int ll_md_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1053 void *data, int flag);
1054 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de);
1055 int ll_rmdir_entry(struct inode *dir, char *name, int namelen);
1056 void ll_update_times(struct ptlrpc_request *request, struct inode *inode);
1059 int ll_writepage(struct page *page, struct writeback_control *wbc);
1060 int ll_writepages(struct address_space *, struct writeback_control *wbc);
1061 int ll_readpage(struct file *file, struct page *page);
1062 int ll_io_read_page(const struct lu_env *env, struct cl_io *io,
1063 struct cl_page *page, struct file *file);
1064 void ll_readahead_init(struct inode *inode, struct ll_readahead_state *ras);
1065 int vvp_io_write_commit(const struct lu_env *env, struct cl_io *io);
1068 void ll_cl_add(struct inode *inode, const struct lu_env *env, struct cl_io *io,
1069 enum lcc_type type);
1070 void ll_cl_remove(struct inode *inode, const struct lu_env *env);
1071 struct ll_cl_context *ll_cl_find(struct inode *inode);
1073 extern const struct address_space_operations ll_aops;
1076 extern const struct inode_operations ll_file_inode_operations;
1077 const struct file_operations *ll_select_file_operations(struct ll_sb_info *sbi);
1078 extern int ll_have_md_lock(struct inode *inode, __u64 *bits,
1079 enum ldlm_mode l_req_mode);
1080 extern enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
1081 struct lustre_handle *lockh, __u64 flags,
1082 enum ldlm_mode mode);
1084 int ll_file_open(struct inode *inode, struct file *file);
1085 int ll_file_release(struct inode *inode, struct file *file);
1086 int ll_release_openhandle(struct dentry *, struct lookup_intent *);
1087 int ll_md_real_close(struct inode *inode, fmode_t fmode);
1088 void ll_track_file_opens(struct inode *inode);
1089 extern void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1090 struct ll_file_data *file, loff_t pos,
1091 size_t count, int rw);
1092 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
1093 int ll_getattr(const struct path *path, struct kstat *stat,
1094 u32 request_mask, unsigned int flags);
1096 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat);
1098 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
1099 unsigned int flags, bool foreign);
1100 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1101 struct posix_acl *ll_get_acl(struct inode *inode, int type);
1102 int ll_set_acl(struct inode *inode, struct posix_acl *acl, int type);
1103 #else /* !CONFIG_LUSTRE_FS_POSIX_ACL */
1104 #define ll_get_acl NULL
1105 #define ll_set_acl NULL
1106 #endif /* CONFIG_LUSTRE_FS_POSIX_ACL */
1108 static inline int ll_xflags_to_inode_flags(int xflags)
1110 return ((xflags & FS_XFLAG_SYNC) ? S_SYNC : 0) |
1111 ((xflags & FS_XFLAG_NOATIME) ? S_NOATIME : 0) |
1112 ((xflags & FS_XFLAG_APPEND) ? S_APPEND : 0) |
1113 ((xflags & FS_XFLAG_IMMUTABLE) ? S_IMMUTABLE : 0);
1116 static inline int ll_inode_flags_to_xflags(int inode_flags)
1118 return ((inode_flags & S_SYNC) ? FS_XFLAG_SYNC : 0) |
1119 ((inode_flags & S_NOATIME) ? FS_XFLAG_NOATIME : 0) |
1120 ((inode_flags & S_APPEND) ? FS_XFLAG_APPEND : 0) |
1121 ((inode_flags & S_IMMUTABLE) ? FS_XFLAG_IMMUTABLE : 0);
1124 int ll_migrate(struct inode *parent, struct file *file,
1125 struct lmv_user_md *lum, const char *name);
1126 int ll_get_fid_by_name(struct inode *parent, const char *name,
1127 int namelen, struct lu_fid *fid, struct inode **inode);
1128 int ll_inode_permission(struct inode *inode, int mask);
1129 int ll_ioctl_check_project(struct inode *inode, __u32 xflags, __u32 projid);
1130 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
1132 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
1134 int ll_ioctl_project(struct file *file, unsigned int cmd,
1137 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
1138 __u64 flags, struct lov_user_md *lum,
1140 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1141 struct lov_mds_md **lmm, int *lmm_size,
1142 struct ptlrpc_request **request);
1143 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
1145 int ll_dir_getstripe_default(struct inode *inode, void **lmmp,
1146 int *lmm_size, struct ptlrpc_request **request,
1147 struct ptlrpc_request **root_request, u64 valid);
1148 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
1149 struct ptlrpc_request **request, u64 valid);
1150 int ll_fsync(struct file *file, loff_t start, loff_t end, int data);
1151 int ll_merge_attr(const struct lu_env *env, struct inode *inode);
1152 int ll_fid2path(struct inode *inode, void __user *arg);
1153 int ll_data_version(struct inode *inode, __u64 *data_version, int flags);
1154 int ll_hsm_release(struct inode *inode);
1155 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss);
1156 void ll_io_set_mirror(struct cl_io *io, const struct file *file);
1158 /* llite/dcache.c */
1160 extern const struct dentry_operations ll_d_ops;
1162 bool ll_d_setup(struct dentry *de, bool do_put);
1164 static inline bool lld_is_init(struct dentry *dentry)
1166 return ll_d2d(dentry);
1169 #define ll_d_setup(de, do_put) (true)
1170 #define lld_is_init(dentry) (true)
1173 void ll_intent_drop_lock(struct lookup_intent *);
1174 void ll_intent_release(struct lookup_intent *);
1175 void ll_prune_aliases(struct inode *inode);
1176 void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry);
1177 int ll_revalidate_it_finish(struct ptlrpc_request *request,
1178 struct lookup_intent *it, struct dentry *de);
1180 /* llite/llite_lib.c */
1181 extern const struct super_operations lustre_super_operations;
1183 void ll_lli_init(struct ll_inode_info *lli);
1184 int ll_fill_super(struct super_block *sb);
1185 void ll_put_super(struct super_block *sb);
1186 void ll_kill_super(struct super_block *sb);
1187 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock);
1188 void ll_dir_clear_lsm_md(struct inode *inode);
1189 void ll_clear_inode(struct inode *inode);
1190 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1191 enum op_xvalid xvalid, bool hsm_import);
1192 int ll_setattr(struct dentry *de, struct iattr *attr);
1193 int ll_statfs(struct dentry *de, struct kstatfs *sfs);
1194 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1196 int ll_update_inode(struct inode *inode, struct lustre_md *md);
1197 void ll_update_inode_flags(struct inode *inode, unsigned int ext_flags);
1198 void ll_update_dir_depth(struct inode *dir, struct inode *inode);
1199 int ll_read_inode2(struct inode *inode, void *opaque);
1200 void ll_truncate_inode_pages_final(struct inode *inode);
1201 void ll_delete_inode(struct inode *inode);
1202 int ll_iocontrol(struct inode *inode, struct file *file,
1203 unsigned int cmd, unsigned long arg);
1204 int ll_flush_ctx(struct inode *inode);
1205 void ll_umount_begin(struct super_block *sb);
1206 int ll_remount_fs(struct super_block *sb, int *flags, char *data);
1207 int ll_show_options(struct seq_file *seq, struct dentry *dentry);
1208 void ll_dirty_page_discard_warn(struct page *page, int ioret);
1209 int ll_prep_inode(struct inode **inode, struct req_capsule *pill,
1210 struct super_block *sb, struct lookup_intent *it);
1211 int ll_obd_statfs(struct inode *inode, void __user *arg);
1212 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *max_mdsize);
1213 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *default_mdsize);
1214 int ll_set_default_mdsize(struct ll_sb_info *sbi, int default_mdsize);
1216 void ll_unlock_md_op_lsm(struct md_op_data *op_data);
1217 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
1218 struct inode *i1, struct inode *i2,
1219 const char *name, size_t namelen,
1220 __u32 mode, enum md_op_code opc,
1222 void ll_finish_md_op_data(struct md_op_data *op_data);
1223 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg);
1224 void ll_compute_rootsquash_state(struct ll_sb_info *sbi);
1225 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
1226 struct lov_user_md **kbuf);
1227 void ll_open_cleanup(struct super_block *sb, struct req_capsule *pill);
1229 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req);
1231 /* Compute expected user md size when passing in a md from user space */
1232 static inline ssize_t ll_lov_user_md_size(const struct lov_user_md *lum)
1234 switch (lum->lmm_magic) {
1235 case LOV_USER_MAGIC_V1:
1236 return sizeof(struct lov_user_md_v1);
1237 case LOV_USER_MAGIC_V3:
1238 return sizeof(struct lov_user_md_v3);
1239 case LOV_USER_MAGIC_SPECIFIC:
1240 if (lum->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
1243 return lov_user_md_size(lum->lmm_stripe_count,
1244 LOV_USER_MAGIC_SPECIFIC);
1245 case LOV_USER_MAGIC_COMP_V1:
1246 return ((struct lov_comp_md_v1 *)lum)->lcm_size;
1247 case LOV_USER_MAGIC_FOREIGN:
1248 return foreign_size(lum);
1254 /* llite/llite_nfs.c */
1255 extern const struct export_operations lustre_export_operations;
1256 __u32 get_uuid2int(const char *name, int len);
1257 struct inode *search_inode_for_lustre(struct super_block *sb,
1258 const struct lu_fid *fid);
1259 int ll_dir_get_parent_fid(struct inode *dir, struct lu_fid *parent_fid);
1261 /* llite/symlink.c */
1262 extern const struct inode_operations ll_fast_symlink_inode_operations;
1265 * IO arguments for various VFS I/O interfaces.
1267 struct vvp_io_args {
1268 /** normal/sendfile/splice */
1271 struct kiocb *via_iocb;
1272 struct iov_iter *via_iter;
1282 struct ll_cl_context {
1283 struct list_head lcc_list;
1285 const struct lu_env *lcc_env;
1286 struct cl_io *lcc_io;
1287 struct cl_page *lcc_page;
1288 enum lcc_type lcc_type;
1290 * Get encryption context operation in progress,
1291 * allow getxattr of LL_XATTR_NAME_ENCRYPTION_CONTEXT xattr
1293 unsigned lcc_getencctx:1;
1296 struct ll_thread_info {
1297 struct vvp_io_args lti_args;
1298 struct ra_io_arg lti_ria;
1299 struct ll_cl_context lti_io_ctx;
1302 extern struct lu_context_key ll_thread_key;
1304 static inline struct ll_thread_info *ll_env_info(const struct lu_env *env)
1306 struct ll_thread_info *lti;
1308 lti = lu_context_key_get(&env->le_ctx, &ll_thread_key);
1309 LASSERT(lti != NULL);
1314 static inline struct vvp_io_args *ll_env_args(const struct lu_env *env)
1316 return &ll_env_info(env)->lti_args;
1319 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1320 struct vvp_io_args *args);
1322 /* llite/llite_mmap.c */
1324 int ll_file_mmap(struct file * file, struct vm_area_struct * vma);
1325 void policy_from_vma(union ldlm_policy_data *policy, struct vm_area_struct *vma,
1326 unsigned long addr, size_t count);
1327 struct vm_area_struct *our_vma(struct mm_struct *mm, unsigned long addr,
1330 #define ll_s2sbi(sb) (s2lsi(sb)->lsi_llsbi)
1332 /* don't need an addref as the sb_info should be holding one */
1333 static inline struct obd_export *ll_s2dtexp(struct super_block *sb)
1335 return ll_s2sbi(sb)->ll_dt_exp;
1338 /* don't need an addref as the sb_info should be holding one */
1339 static inline struct obd_export *ll_s2mdexp(struct super_block *sb)
1341 return ll_s2sbi(sb)->ll_md_exp;
1344 static inline struct client_obd *sbi2mdc(struct ll_sb_info *sbi)
1346 struct obd_device *obd = sbi->ll_md_exp->exp_obd;
1352 // FIXME: replace the name of this with LL_SB to conform to kernel stuff
1353 static inline struct ll_sb_info *ll_i2sbi(struct inode *inode)
1355 return ll_s2sbi(inode->i_sb);
1358 static inline struct obd_export *ll_i2dtexp(struct inode *inode)
1360 return ll_s2dtexp(inode->i_sb);
1363 static inline struct obd_export *ll_i2mdexp(struct inode *inode)
1365 return ll_s2mdexp(inode->i_sb);
1368 static inline struct lu_fid *ll_inode2fid(struct inode *inode)
1372 LASSERT(inode != NULL);
1373 fid = &ll_i2info(inode)->lli_fid;
1378 static inline bool ll_dir_striped(struct inode *inode)
1381 return S_ISDIR(inode->i_mode) &&
1382 lmv_dir_striped(ll_i2info(inode)->lli_lsm_md);
1385 static inline loff_t ll_file_maxbytes(struct inode *inode)
1387 struct cl_object *obj = ll_i2info(inode)->lli_clob;
1390 return MAX_LFS_FILESIZE;
1392 return min_t(loff_t, cl_object_maxbytes(obj), MAX_LFS_FILESIZE);
1396 extern const struct xattr_handler *ll_xattr_handlers[];
1398 #define XATTR_USER_T 1
1399 #define XATTR_TRUSTED_T 2
1400 #define XATTR_SECURITY_T 3
1401 #define XATTR_ACL_ACCESS_T 4
1402 #define XATTR_ACL_DEFAULT_T 5
1403 #define XATTR_LUSTRE_T 6
1404 #define XATTR_OTHER_T 7
1406 ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size);
1407 int ll_xattr_list(struct inode *inode, const char *name, int type,
1408 void *buffer, size_t size, u64 valid);
1409 const struct xattr_handler *get_xattr_type(const char *name);
1412 * Common IO arguments for various VFS I/O interfaces.
1414 int cl_sb_init(struct super_block *sb);
1415 int cl_sb_fini(struct super_block *sb);
1417 enum ras_update_flags {
1421 void ll_ra_count_put(struct ll_sb_info *sbi, unsigned long len);
1422 void ll_ra_stats_inc(struct inode *inode, enum ra_stat which);
1426 #define LL_SA_RPC_MIN 2
1427 #define LL_SA_RPC_DEF 32
1428 #define LL_SA_RPC_MAX 512
1430 /* XXX: If want to support more concurrent statahead instances,
1431 * please consider to decentralize the RPC lists attached
1432 * on related import, such as imp_{sending,delayed}_list.
1434 #define LL_SA_RUNNING_MAX 256
1435 #define LL_SA_RUNNING_DEF 16
1437 #define LL_SA_CACHE_BIT 5
1438 #define LL_SA_CACHE_SIZE (1 << LL_SA_CACHE_BIT)
1439 #define LL_SA_CACHE_MASK (LL_SA_CACHE_SIZE - 1)
1441 /* per inode struct, for dir only */
1442 struct ll_statahead_info {
1443 struct dentry *sai_dentry;
1444 atomic_t sai_refcount; /* when access this struct, hold
1446 unsigned int sai_max; /* max ahead of lookup */
1447 __u64 sai_sent; /* stat requests sent count */
1448 __u64 sai_replied; /* stat requests which received
1450 __u64 sai_index; /* index of statahead entry */
1451 __u64 sai_index_wait; /* index of entry which is the
1452 * caller is waiting for */
1453 __u64 sai_hit; /* hit count */
1454 __u64 sai_miss; /* miss count:
1455 * for "ls -al" case, includes
1456 * hidden dentry miss;
1457 * for "ls -l" case, it does not
1458 * include hidden dentry miss.
1459 * "sai_miss_hidden" is used for
1462 unsigned int sai_consecutive_miss; /* consecutive miss */
1463 unsigned int sai_miss_hidden;/* "ls -al", but first dentry
1464 * is not a hidden one */
1465 unsigned int sai_skip_hidden;/* skipped hidden dentry count
1467 unsigned int sai_ls_all:1, /* "ls -al", do stat-ahead for
1469 sai_in_readpage:1;/* statahead is in readdir()*/
1470 wait_queue_head_t sai_waitq; /* stat-ahead wait queue */
1471 struct task_struct *sai_task; /* stat-ahead thread */
1472 struct task_struct *sai_agl_task; /* AGL thread */
1473 struct list_head sai_interim_entries; /* entries which got async
1474 * stat reply, but not
1476 struct list_head sai_entries; /* completed entries */
1477 struct list_head sai_agls; /* AGLs to be sent */
1478 struct list_head sai_cache[LL_SA_CACHE_SIZE];
1479 spinlock_t sai_cache_lock[LL_SA_CACHE_SIZE];
1480 atomic_t sai_cache_count; /* entry count in cache */
1483 int ll_revalidate_statahead(struct inode *dir, struct dentry **dentry,
1485 int ll_start_statahead(struct inode *dir, struct dentry *dentry, bool agl);
1486 void ll_authorize_statahead(struct inode *dir, void *key);
1487 void ll_deauthorize_statahead(struct inode *dir, void *key);
1490 blkcnt_t dirty_cnt(struct inode *inode);
1492 int cl_glimpse_size0(struct inode *inode, int agl);
1493 int cl_glimpse_lock(const struct lu_env *env, struct cl_io *io,
1494 struct inode *inode, struct cl_object *clob, int agl);
1496 static inline int cl_glimpse_size(struct inode *inode)
1498 return cl_glimpse_size0(inode, 0);
1501 /* AGL is 'asychronous glimpse lock', which is a speculative lock taken as
1502 * part of statahead */
1503 static inline int cl_agl(struct inode *inode)
1505 return cl_glimpse_size0(inode, 1);
1508 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise);
1510 int cl_io_get(struct inode *inode, struct lu_env **envout,
1511 struct cl_io **ioout, __u16 *refcheck);
1513 static inline int ll_glimpse_size(struct inode *inode)
1515 struct ll_inode_info *lli = ll_i2info(inode);
1518 down_read(&lli->lli_glimpse_sem);
1519 rc = cl_glimpse_size(inode);
1520 lli->lli_glimpse_time = ktime_get();
1521 up_read(&lli->lli_glimpse_sem);
1525 /* dentry may statahead when statahead is enabled and current process has opened
1526 * parent directory, and this dentry hasn't accessed statahead cache before */
1528 dentry_may_statahead(struct inode *dir, struct dentry *dentry)
1530 struct ll_inode_info *lli;
1531 struct ll_dentry_data *ldd;
1533 if (ll_i2sbi(dir)->ll_sa_max == 0)
1536 lli = ll_i2info(dir);
1538 /* statahead is not allowed for this dir, there may be three causes:
1539 * 1. dir is not opened.
1540 * 2. statahead hit ratio is too low.
1541 * 3. previous stat started statahead thread failed. */
1542 if (!lli->lli_sa_enabled)
1545 /* not the same process, don't statahead */
1546 if (lli->lli_opendir_pid != current->pid)
1550 * When stating a dentry, kernel may trigger 'revalidate' or 'lookup'
1551 * multiple times, eg. for 'getattr', 'getxattr' and etc.
1552 * For patchless client, lookup intent is not accurate, which may
1553 * misguide statahead. For example:
1554 * The 'revalidate' call for 'getattr' and 'getxattr' of a dentry will
1555 * have the same intent -- IT_GETATTR, while one dentry should access
1556 * statahead cache once, otherwise statahead windows is messed up.
1557 * The solution is as following:
1558 * Assign 'lld_sa_generation' with 'lli_sa_generation' when a dentry
1559 * IT_GETATTR for the first time, and subsequent IT_GETATTR will
1560 * bypass interacting with statahead cache by checking
1561 * 'lld_sa_generation == lli->lli_sa_generation'.
1563 ldd = ll_d2d(dentry);
1564 if (ldd != NULL && lli->lli_sa_generation &&
1565 ldd->lld_sa_generation == lli->lli_sa_generation)
1571 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
1572 enum cl_fsync_mode mode, int ignore_layout);
1574 static inline int ll_file_nolock(const struct file *file)
1576 struct ll_file_data *fd = file->private_data;
1577 struct inode *inode = file_inode((struct file *)file);
1579 LASSERT(fd != NULL);
1580 return ((fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
1581 test_bit(LL_SBI_NOLCK, ll_i2sbi(inode)->ll_flags));
1584 static inline void ll_set_lock_data(struct obd_export *exp, struct inode *inode,
1585 struct lookup_intent *it, __u64 *bits)
1587 if (!it->it_lock_set) {
1588 struct lustre_handle handle;
1590 /* If this inode is a remote object, it will get two
1591 * separate locks in different namespaces, Master MDT,
1592 * where the name entry is, will grant LOOKUP lock,
1593 * remote MDT, where the object is, will grant
1594 * UPDATE|PERM lock. The inode will be attched to both
1595 * LOOKUP and PERM locks, so revoking either locks will
1596 * case the dcache being cleared */
1597 if (it->it_remote_lock_mode) {
1598 handle.cookie = it->it_remote_lock_handle;
1599 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID
1600 "(%p) for remote lock %#llx\n",
1601 PFID(ll_inode2fid(inode)), inode,
1603 md_set_lock_data(exp, &handle, inode, NULL);
1606 handle.cookie = it->it_lock_handle;
1608 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID"(%p)"
1609 " for lock %#llx\n",
1610 PFID(ll_inode2fid(inode)), inode, handle.cookie);
1612 md_set_lock_data(exp, &handle, inode, &it->it_lock_bits);
1613 it->it_lock_set = 1;
1617 *bits = it->it_lock_bits;
1620 static inline int d_lustre_invalid(const struct dentry *dentry)
1622 return !ll_d2d(dentry) || ll_d2d(dentry)->lld_invalid;
1626 * Mark dentry INVALID, if dentry refcount is zero (this is normally case for
1627 * ll_md_blocking_ast), it will be pruned by ll_prune_aliases() and
1628 * ll_prune_negative_children(); otherwise dput() of the last refcount will
1629 * unhash this dentry and kill it.
1631 static inline void d_lustre_invalidate(struct dentry *dentry)
1633 CDEBUG(D_DENTRY, "invalidate dentry %pd (%p) parent %p inode %p refc %d\n",
1635 dentry->d_parent, dentry->d_inode, ll_d_count(dentry));
1637 spin_lock(&dentry->d_lock);
1638 if (lld_is_init(dentry))
1639 ll_d2d(dentry)->lld_invalid = 1;
1640 spin_unlock(&dentry->d_lock);
1643 static inline void d_lustre_revalidate(struct dentry *dentry)
1645 spin_lock(&dentry->d_lock);
1646 LASSERT(ll_d2d(dentry));
1647 ll_d2d(dentry)->lld_invalid = 0;
1648 spin_unlock(&dentry->d_lock);
1651 static inline dev_t ll_compat_encode_dev(dev_t dev)
1653 /* The compat_sys_*stat*() syscalls will fail unless the
1654 * device majors and minors are both less than 256. Note that
1655 * the value returned here will be passed through
1656 * old_encode_dev() in cp_compat_stat(). And so we are not
1657 * trying to return a valid compat (u16) device number, just
1658 * one that will pass the old_valid_dev() check. */
1660 return MKDEV(MAJOR(dev) & 0xff, MINOR(dev) & 0xff);
1663 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf);
1664 int ll_layout_refresh(struct inode *inode, __u32 *gen);
1665 int ll_layout_restore(struct inode *inode, loff_t start, __u64 length);
1666 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
1667 struct lu_extent *ext);
1669 int ll_xattr_init(void);
1670 void ll_xattr_fini(void);
1672 int ll_page_sync_io(const struct lu_env *env, struct cl_io *io,
1673 struct cl_page *page, enum cl_req_type crt);
1675 int ll_getparent(struct file *file, struct getparent __user *arg);
1678 int cl_setattr_ost(struct cl_object *obj, const struct iattr *attr,
1679 enum op_xvalid xvalid, unsigned int attr_flags);
1681 extern struct lu_env *cl_inode_fini_env;
1682 extern __u16 cl_inode_fini_refcheck;
1684 int cl_file_inode_init(struct inode *inode, struct lustre_md *md);
1685 void cl_inode_fini(struct inode *inode);
1687 u64 cl_fid_build_ino(const struct lu_fid *fid, int api32);
1688 u32 cl_fid_build_gen(const struct lu_fid *fid);
1690 static inline struct pcc_super *ll_i2pccs(struct inode *inode)
1692 return &ll_i2sbi(inode)->ll_pcc_super;
1695 static inline struct pcc_super *ll_info2pccs(struct ll_inode_info *lli)
1697 return ll_i2pccs(ll_info2i(lli));
1701 /* The digested form is made of a FID (16 bytes) followed by the second-to-last
1702 * ciphertext block (16 bytes), so a total length of 32 bytes.
1703 * That way, llcrypt does not compute a digested form of this digest.
1705 struct ll_digest_filename {
1706 struct lu_fid ldf_fid;
1707 char ldf_excerpt[LLCRYPT_FNAME_DIGEST_SIZE];
1710 int ll_setup_filename(struct inode *dir, const struct qstr *iname,
1711 int lookup, struct llcrypt_name *fname,
1712 struct lu_fid *fid);
1713 int ll_fname_disk_to_usr(struct inode *inode,
1714 u32 hash, u32 minor_hash,
1715 struct llcrypt_str *iname, struct llcrypt_str *oname,
1716 struct lu_fid *fid);
1717 int ll_revalidate_d_crypto(struct dentry *dentry, unsigned int flags);
1718 #ifdef HAVE_LUSTRE_CRYPTO
1719 extern const struct llcrypt_operations lustre_cryptops;
1722 /* llite/llite_foreign.c */
1723 int ll_manage_foreign(struct inode *inode, struct lustre_md *lmd);
1724 bool ll_foreign_is_openable(struct dentry *dentry, unsigned int flags);
1725 bool ll_foreign_is_removable(struct dentry *dentry, bool unset);
1727 #endif /* LLITE_INTERNAL_H */