diff --git a/[refs] b/[refs] index ac323a241b52..73f032511f67 100644 --- a/[refs] +++ b/[refs] @@ -1,2 +1,2 @@ --- -refs/heads/master: 0b1fc335d2f2c1206ac4048e5f6d8971f2aae6be +refs/heads/master: 3d6e48f43340343d97839eadb1ab7b6a3ea98797 diff --git a/trunk/Documentation/filesystems/Locking b/trunk/Documentation/filesystems/Locking index 8362860e21a7..680fb566b928 100644 --- a/trunk/Documentation/filesystems/Locking +++ b/trunk/Documentation/filesystems/Locking @@ -144,8 +144,8 @@ prototypes: void (*kill_sb) (struct super_block *); locking rules: may block BKL -get_sb yes no -kill_sb yes no +get_sb yes yes +kill_sb yes yes ->get_sb() returns error or 0 with locked superblock attached to the vfsmount (exclusive on ->s_umount). @@ -409,12 +409,12 @@ ioctl: yes (see below) unlocked_ioctl: no (see below) compat_ioctl: no mmap: no -open: no +open: maybe (see below) flush: no release: no fsync: no (see below) aio_fsync: no -fasync: no +fasync: yes (see below) lock: yes readv: no writev: no @@ -431,6 +431,13 @@ For many filesystems, it is probably safe to acquire the inode semaphore. Note some filesystems (i.e. remote ones) provide no protection for i_size so you will need to use the BKL. +->open() locking is in-transit: big lock partially moved into the methods. +The only exception is ->open() in the instances of file_operations that never +end up in ->i_fop/->proc_fops, i.e. ones that belong to character devices +(chrdev_open() takes lock before replacing ->f_op and calling the secondary +method. As soon as we fix the handling of module reference counters all +instances of ->open() will be called without the BKL. + Note: ext2_release() was *the* source of contention on fs-intensive loads and dropping BKL on ->release() helps to get rid of that (we still grab BKL for cases when we close a file that had been opened r/w, but that diff --git a/trunk/MAINTAINERS b/trunk/MAINTAINERS index 186be3ba5069..b3e92fbe336c 100644 --- a/trunk/MAINTAINERS +++ b/trunk/MAINTAINERS @@ -750,13 +750,11 @@ P: Ville Syrjala M: syrjala@sci.fi S: Maintained -ATLX ETHERNET DRIVERS +ATL1 ETHERNET DRIVER P: Jay Cliburn M: jcliburn@gmail.com P: Chris Snook M: csnook@redhat.com -P: Jie Yang -M: jie.yang@atheros.com L: atl1-devel@lists.sourceforge.net W: http://sourceforge.net/projects/atl1 W: http://atl1.sourceforge.net diff --git a/trunk/arch/ia64/include/asm/sections.h b/trunk/arch/ia64/include/asm/sections.h index a7acad2bc2f0..7286e4a9fe84 100644 --- a/trunk/arch/ia64/include/asm/sections.h +++ b/trunk/arch/ia64/include/asm/sections.h @@ -21,8 +21,5 @@ extern char __start_gate_brl_fsys_bubble_down_patchlist[], __end_gate_brl_fsys_b extern char __start_unwind[], __end_unwind[]; extern char __start_ivt_text[], __end_ivt_text[]; -#undef dereference_function_descriptor -void *dereference_function_descriptor(void *); - #endif /* _ASM_IA64_SECTIONS_H */ diff --git a/trunk/arch/ia64/kernel/module.c b/trunk/arch/ia64/kernel/module.c index 545626f66a4c..29aad349e0c4 100644 --- a/trunk/arch/ia64/kernel/module.c +++ b/trunk/arch/ia64/kernel/module.c @@ -31,11 +31,9 @@ #include #include #include -#include #include #include -#include #include #define ARCH_MODULE_DEBUG 0 @@ -943,13 +941,3 @@ module_arch_cleanup (struct module *mod) if (mod->arch.core_unw_table) unw_remove_unwind_table(mod->arch.core_unw_table); } - -void *dereference_function_descriptor(void *ptr) -{ - struct fdesc *desc = ptr; - void *p; - - if (!probe_kernel_address(&desc->ip, p)) - ptr = p; - return ptr; -} diff --git a/trunk/arch/mips/sgi-ip22/ip22-platform.c b/trunk/arch/mips/sgi-ip22/ip22-platform.c index 52486c4d2b01..60141235ec40 100644 --- a/trunk/arch/mips/sgi-ip22/ip22-platform.c +++ b/trunk/arch/mips/sgi-ip22/ip22-platform.c @@ -150,7 +150,7 @@ static int __init sgiseeq_devinit(void) return res; /* Second HPC is missing? */ - if (ip22_is_fullhouse() || + if (!ip22_is_fullhouse() || get_dbe(tmp, (unsigned int *)&hpc3c1->pbdma[1])) return 0; diff --git a/trunk/arch/parisc/kernel/module.c b/trunk/arch/parisc/kernel/module.c index 44138c3e6ea7..fdacdd4341c9 100644 --- a/trunk/arch/parisc/kernel/module.c +++ b/trunk/arch/parisc/kernel/module.c @@ -47,9 +47,7 @@ #include #include #include -#include -#include #include #if 0 @@ -862,15 +860,3 @@ void module_arch_cleanup(struct module *mod) deregister_unwind_table(mod); module_bug_cleanup(mod); } - -#ifdef CONFIG_64BIT -void *dereference_function_descriptor(void *ptr) -{ - Elf64_Fdesc *desc = ptr; - void *p; - - if (!probe_kernel_address(&desc->addr, p)) - ptr = p; - return ptr; -} -#endif diff --git a/trunk/arch/powerpc/include/asm/sections.h b/trunk/arch/powerpc/include/asm/sections.h index 7710e9e6660f..916018e425c4 100644 --- a/trunk/arch/powerpc/include/asm/sections.h +++ b/trunk/arch/powerpc/include/asm/sections.h @@ -16,9 +16,6 @@ static inline int in_kernel_text(unsigned long addr) return 0; } -#undef dereference_function_descriptor -void *dereference_function_descriptor(void *); - #endif #endif /* __KERNEL__ */ diff --git a/trunk/arch/powerpc/kernel/module_64.c b/trunk/arch/powerpc/kernel/module_64.c index ad79de272ff3..ee6a2982d567 100644 --- a/trunk/arch/powerpc/kernel/module_64.c +++ b/trunk/arch/powerpc/kernel/module_64.c @@ -21,9 +21,8 @@ #include #include #include -#include #include -#include +#include #include #include #include @@ -452,13 +451,3 @@ int apply_relocate_add(Elf64_Shdr *sechdrs, return 0; } - -void *dereference_function_descriptor(void *ptr) -{ - struct ppc64_opd_entry *desc = ptr; - void *p; - - if (!probe_kernel_address(&desc->funcaddr, p)) - ptr = p; - return ptr; -} diff --git a/trunk/arch/s390/kernel/compat_ptrace.h b/trunk/arch/s390/kernel/compat_ptrace.h index cde81fa64f89..a2be3a978d5c 100644 --- a/trunk/arch/s390/kernel/compat_ptrace.h +++ b/trunk/arch/s390/kernel/compat_ptrace.h @@ -42,6 +42,7 @@ struct user_regs_struct32 u32 gprs[NUM_GPRS]; u32 acrs[NUM_ACRS]; u32 orig_gpr2; + /* nb: there's a 4-byte hole here */ s390_fp_regs fp_regs; /* * These per registers are in here so that gdb can modify them diff --git a/trunk/arch/s390/kernel/ptrace.c b/trunk/arch/s390/kernel/ptrace.c index 2815bfe348a6..c8b08289eb87 100644 --- a/trunk/arch/s390/kernel/ptrace.c +++ b/trunk/arch/s390/kernel/ptrace.c @@ -170,6 +170,13 @@ static unsigned long __peek_user(struct task_struct *child, addr_t addr) */ tmp = (addr_t) task_pt_regs(child)->orig_gpr2; + } else if (addr < (addr_t) &dummy->regs.fp_regs) { + /* + * prevent reads of padding hole between + * orig_gpr2 and fp_regs on s390. + */ + tmp = 0; + } else if (addr < (addr_t) (&dummy->regs.fp_regs + 1)) { /* * floating point regs. are stored in the thread structure @@ -270,6 +277,13 @@ static int __poke_user(struct task_struct *child, addr_t addr, addr_t data) */ task_pt_regs(child)->orig_gpr2 = data; + } else if (addr < (addr_t) &dummy->regs.fp_regs) { + /* + * prevent writes of padding hole between + * orig_gpr2 and fp_regs on s390. + */ + return 0; + } else if (addr < (addr_t) (&dummy->regs.fp_regs + 1)) { /* * floating point regs. are stored in the thread structure @@ -428,6 +442,13 @@ static u32 __peek_user_compat(struct task_struct *child, addr_t addr) */ tmp = *(__u32*)((addr_t) &task_pt_regs(child)->orig_gpr2 + 4); + } else if (addr < (addr_t) &dummy32->regs.fp_regs) { + /* + * prevent reads of padding hole between + * orig_gpr2 and fp_regs on s390. + */ + tmp = 0; + } else if (addr < (addr_t) (&dummy32->regs.fp_regs + 1)) { /* * floating point regs. are stored in the thread structure @@ -514,6 +535,13 @@ static int __poke_user_compat(struct task_struct *child, */ *(__u32*)((addr_t) &task_pt_regs(child)->orig_gpr2 + 4) = tmp; + } else if (addr < (addr_t) &dummy32->regs.fp_regs) { + /* + * prevent writess of padding hole between + * orig_gpr2 and fp_regs on s390. + */ + return 0; + } else if (addr < (addr_t) (&dummy32->regs.fp_regs + 1)) { /* * floating point regs. are stored in the thread structure diff --git a/trunk/fs/ubifs/budget.c b/trunk/fs/ubifs/budget.c index 73db464cd08b..154098157473 100644 --- a/trunk/fs/ubifs/budget.c +++ b/trunk/fs/ubifs/budget.c @@ -302,6 +302,18 @@ long long ubifs_calc_available(const struct ubifs_info *c, int min_idx_lebs) int subtract_lebs; long long available; + /* + * Force the amount available to the total size reported if the used + * space is zero. + */ + if (c->lst.total_used <= UBIFS_INO_NODE_SZ && + c->budg_data_growth + c->budg_dd_growth == 0) { + /* Do the same calculation as for c->block_cnt */ + available = c->main_lebs - 2; + available *= c->leb_size - c->dark_wm; + return available; + } + available = c->main_bytes - c->lst.total_used; /* @@ -702,106 +714,34 @@ void ubifs_release_dirty_inode_budget(struct ubifs_info *c, } /** - * ubifs_reported_space - calculate reported free space. - * @c: the UBIFS file-system description object - * @free: amount of free space - * - * This function calculates amount of free space which will be reported to - * user-space. User-space application tend to expect that if the file-system - * (e.g., via the 'statfs()' call) reports that it has N bytes available, they - * are able to write a file of size N. UBIFS attaches node headers to each data - * node and it has to write indexind nodes as well. This introduces additional - * overhead, and UBIFS it has to report sligtly less free space to meet the - * above expectetion. - * - * This function assumes free space is made up of uncompressed data nodes and - * full index nodes (one per data node, tripled because we always allow enough - * space to write the index thrice). - * - * Note, the calculation is pessimistic, which means that most of the time - * UBIFS reports less space than it actually has. - */ -long long ubifs_reported_space(const struct ubifs_info *c, uint64_t free) -{ - int divisor, factor, f; - - /* - * Reported space size is @free * X, where X is UBIFS block size - * divided by UBIFS block size + all overhead one data block - * introduces. The overhead is the node header + indexing overhead. - * - * Indexing overhead calculations are based on the following formula: - * I = N/(f - 1) + 1, where I - number of indexing nodes, N - number - * of data nodes, f - fanout. Because effective UBIFS fanout is twice - * as less than maximum fanout, we assume that each data node - * introduces 3 * @c->max_idx_node_sz / (@c->fanout/2 - 1) bytes. - * Note, the multiplier 3 is because UBIFS reseves thrice as more space - * for the index. - */ - f = c->fanout > 3 ? c->fanout >> 1 : 2; - factor = UBIFS_BLOCK_SIZE; - divisor = UBIFS_MAX_DATA_NODE_SZ; - divisor += (c->max_idx_node_sz * 3) / (f - 1); - free *= factor; - do_div(free, divisor); - return free; -} - -/** - * ubifs_get_free_space - return amount of free space. + * ubifs_budg_get_free_space - return amount of free space. * @c: UBIFS file-system description object * - * This function calculates amount of free space to report to user-space. - * - * Because UBIFS may introduce substantial overhead (the index, node headers, - * alighment, wastage at the end of eraseblocks, etc), it cannot report real - * amount of free flash space it has (well, because not all dirty space is - * reclamable, UBIFS does not actually know the real amount). If UBIFS did so, - * it would bread user expectetion about what free space is. Users seem to - * accustomed to assume that if the file-system reports N bytes of free space, - * they would be able to fit a file of N bytes to the FS. This almost works for - * traditional file-systems, because they have way less overhead than UBIFS. - * So, to keep users happy, UBIFS tries to take the overhead into account. + * This function returns amount of free space on the file-system. */ -long long ubifs_get_free_space(struct ubifs_info *c) +long long ubifs_budg_get_free_space(struct ubifs_info *c) { - int min_idx_lebs, rsvd_idx_lebs, lebs; + int min_idx_lebs, rsvd_idx_lebs; long long available, outstanding, free; + /* Do exactly the same calculations as in 'do_budget_space()' */ spin_lock(&c->space_lock); min_idx_lebs = ubifs_calc_min_idx_lebs(c); - outstanding = c->budg_data_growth + c->budg_dd_growth; - /* - * Force the amount available to the total size reported if the used - * space is zero. - */ - if (c->lst.total_used <= UBIFS_INO_NODE_SZ && !outstanding) { - spin_unlock(&c->space_lock); - return (long long)c->block_cnt << UBIFS_BLOCK_SHIFT; - } - - available = ubifs_calc_available(c, min_idx_lebs); - - /* - * When reporting free space to user-space, UBIFS guarantees that it is - * possible to write a file of free space size. This means that for - * empty LEBs we may use more precise calculations than - * 'ubifs_calc_available()' is using. Namely, we know that in empty - * LEBs we would waste only @c->leb_overhead bytes, not @c->dark_wm. - * Thus, amend the available space. - * - * Note, the calculations below are similar to what we have in - * 'do_budget_space()', so refer there for comments. - */ if (min_idx_lebs > c->lst.idx_lebs) rsvd_idx_lebs = min_idx_lebs - c->lst.idx_lebs; else rsvd_idx_lebs = 0; - lebs = c->lst.empty_lebs + c->freeable_cnt + c->idx_gc_cnt - - c->lst.taken_empty_lebs; - lebs -= rsvd_idx_lebs; - available += lebs * (c->dark_wm - c->leb_overhead); + + if (rsvd_idx_lebs > c->lst.empty_lebs + c->freeable_cnt + c->idx_gc_cnt + - c->lst.taken_empty_lebs) { + spin_unlock(&c->space_lock); + return 0; + } + + available = ubifs_calc_available(c, min_idx_lebs); + outstanding = c->budg_data_growth + c->budg_dd_growth; + c->min_idx_lebs = min_idx_lebs; spin_unlock(&c->space_lock); if (available > outstanding) diff --git a/trunk/fs/ubifs/dir.c b/trunk/fs/ubifs/dir.c index 2b267c9a1806..5c96f1fb7016 100644 --- a/trunk/fs/ubifs/dir.c +++ b/trunk/fs/ubifs/dir.c @@ -587,6 +587,7 @@ static int ubifs_unlink(struct inode *dir, struct dentry *dentry) if (err) { if (err != -ENOSPC) return err; + err = 0; budgeted = 0; } diff --git a/trunk/fs/ubifs/file.c b/trunk/fs/ubifs/file.c index 3d698e2022b1..4071d1cae29f 100644 --- a/trunk/fs/ubifs/file.c +++ b/trunk/fs/ubifs/file.c @@ -793,7 +793,7 @@ static int do_truncation(struct ubifs_info *c, struct inode *inode, int err; struct ubifs_budget_req req; loff_t old_size = inode->i_size, new_size = attr->ia_size; - int offset = new_size & (UBIFS_BLOCK_SIZE - 1), budgeted = 1; + int offset = new_size & (UBIFS_BLOCK_SIZE - 1); struct ubifs_inode *ui = ubifs_inode(inode); dbg_gen("ino %lu, size %lld -> %lld", inode->i_ino, old_size, new_size); @@ -811,15 +811,8 @@ static int do_truncation(struct ubifs_info *c, struct inode *inode, /* A funny way to budget for truncation node */ req.dirtied_ino_d = UBIFS_TRUN_NODE_SZ; err = ubifs_budget_space(c, &req); - if (err) { - /* - * Treat truncations to zero as deletion and always allow them, - * just like we do for '->unlink()'. - */ - if (new_size || err != -ENOSPC) - return err; - budgeted = 0; - } + if (err) + return err; err = vmtruncate(inode, new_size); if (err) @@ -876,12 +869,7 @@ static int do_truncation(struct ubifs_info *c, struct inode *inode, err = ubifs_jnl_truncate(c, inode, old_size, new_size); mutex_unlock(&ui->ui_mutex); out_budg: - if (budgeted) - ubifs_release_budget(c, &req); - else { - c->nospace = c->nospace_rp = 0; - smp_wmb(); - } + ubifs_release_budget(c, &req); return err; } diff --git a/trunk/fs/ubifs/find.c b/trunk/fs/ubifs/find.c index e045c8b55423..adee7b5ddeab 100644 --- a/trunk/fs/ubifs/find.c +++ b/trunk/fs/ubifs/find.c @@ -211,8 +211,14 @@ static const struct ubifs_lprops *scan_for_dirty(struct ubifs_info *c, * dirty index heap, and it falls-back to LPT scanning if the heaps are empty * or do not have an LEB which satisfies the @min_space criteria. * - * Note, LEBs which have less than dead watermark of free + dirty space are - * never picked by this function. + * Note: + * o LEBs which have less than dead watermark of dirty space are never picked + * by this function; + * + * Returns zero and the LEB properties of + * found dirty LEB in case of success, %-ENOSPC if no dirty LEB was found and a + * negative error code in case of other failures. The returned LEB is marked as + * "taken". * * The additional @pick_free argument controls if this function has to return a * free or freeable LEB if one is present. For example, GC must to set it to %1, @@ -225,10 +231,6 @@ static const struct ubifs_lprops *scan_for_dirty(struct ubifs_info *c, * * In addition @pick_free is set to %2 by the recovery process in order to * recover gc_lnum in which case an index LEB must not be returned. - * - * This function returns zero and the LEB properties of found dirty LEB in case - * of success, %-ENOSPC if no dirty LEB was found and a negative error code in - * case of other failures. The returned LEB is marked as "taken". */ int ubifs_find_dirty_leb(struct ubifs_info *c, struct ubifs_lprops *ret_lp, int min_space, int pick_free) @@ -243,7 +245,7 @@ int ubifs_find_dirty_leb(struct ubifs_info *c, struct ubifs_lprops *ret_lp, int lebs, rsvd_idx_lebs = 0; spin_lock(&c->space_lock); - lebs = c->lst.empty_lebs + c->idx_gc_cnt; + lebs = c->lst.empty_lebs; lebs += c->freeable_cnt - c->lst.taken_empty_lebs; /* @@ -315,7 +317,7 @@ int ubifs_find_dirty_leb(struct ubifs_info *c, struct ubifs_lprops *ret_lp, lp = idx_lp; if (lp) { - ubifs_assert(lp->free + lp->dirty >= c->dead_wm); + ubifs_assert(lp->dirty >= c->dead_wm); goto found; } diff --git a/trunk/fs/ubifs/gc.c b/trunk/fs/ubifs/gc.c index 13f1019c859f..d0f3dac29081 100644 --- a/trunk/fs/ubifs/gc.c +++ b/trunk/fs/ubifs/gc.c @@ -344,12 +344,6 @@ int ubifs_garbage_collect_leb(struct ubifs_info *c, struct ubifs_lprops *lp) if (err) goto out; - /* Allow for races with TNC */ - c->gced_lnum = lnum; - smp_wmb(); - c->gc_seq += 1; - smp_wmb(); - if (c->gc_lnum == -1) { c->gc_lnum = lnum; err = LEB_RETAINED; diff --git a/trunk/fs/ubifs/misc.h b/trunk/fs/ubifs/misc.h index 4c12a9215d7f..87dabf9fe742 100644 --- a/trunk/fs/ubifs/misc.h +++ b/trunk/fs/ubifs/misc.h @@ -284,30 +284,45 @@ static inline void *ubifs_idx_key(const struct ubifs_info *c, } /** - * ubifs_current_time - round current time to time granularity. - * @inode: inode + * ubifs_reported_space - calculate reported free space. + * @c: the UBIFS file-system description object + * @free: amount of free space + * + * This function calculates amount of free space which will be reported to + * user-space. User-space application tend to expect that if the file-system + * (e.g., via the 'statfs()' call) reports that it has N bytes available, they + * are able to write a file of size N. UBIFS attaches node headers to each data + * node and it has to write indexind nodes as well. This introduces additional + * overhead, and UBIFS it has to report sligtly less free space to meet the + * above expectetion. + * + * This function assumes free space is made up of uncompressed data nodes and + * full index nodes (one per data node, doubled because we always allow enough + * space to write the index twice). + * + * Note, the calculation is pessimistic, which means that most of the time + * UBIFS reports less space than it actually has. */ -static inline struct timespec ubifs_current_time(struct inode *inode) +static inline long long ubifs_reported_space(const struct ubifs_info *c, + uint64_t free) { - return (inode->i_sb->s_time_gran < NSEC_PER_SEC) ? - current_fs_time(inode->i_sb) : CURRENT_TIME_SEC; + int divisor, factor; + + divisor = UBIFS_MAX_DATA_NODE_SZ + (c->max_idx_node_sz * 3); + factor = UBIFS_MAX_DATA_NODE_SZ - UBIFS_DATA_NODE_SZ; + do_div(free, divisor); + + return free * factor; } /** - * ubifs_tnc_lookup - look up a file-system node. - * @c: UBIFS file-system description object - * @key: node key to lookup - * @node: the node is returned here - * - * This function look up and reads node with key @key. The caller has to make - * sure the @node buffer is large enough to fit the node. Returns zero in case - * of success, %-ENOENT if the node was not found, and a negative error code in - * case of failure. + * ubifs_current_time - round current time to time granularity. + * @inode: inode */ -static inline int ubifs_tnc_lookup(struct ubifs_info *c, - const union ubifs_key *key, void *node) +static inline struct timespec ubifs_current_time(struct inode *inode) { - return ubifs_tnc_locate(c, key, node, NULL, NULL); + return (inode->i_sb->s_time_gran < NSEC_PER_SEC) ? + current_fs_time(inode->i_sb) : CURRENT_TIME_SEC; } #endif /* __UBIFS_MISC_H__ */ diff --git a/trunk/fs/ubifs/super.c b/trunk/fs/ubifs/super.c index 7562464ac83f..f71e6b8822c4 100644 --- a/trunk/fs/ubifs/super.c +++ b/trunk/fs/ubifs/super.c @@ -370,9 +370,8 @@ static int ubifs_statfs(struct dentry *dentry, struct kstatfs *buf) { struct ubifs_info *c = dentry->d_sb->s_fs_info; unsigned long long free; - __le32 *uuid = (__le32 *)c->uuid; - free = ubifs_get_free_space(c); + free = ubifs_budg_get_free_space(c); dbg_gen("free space %lld bytes (%lld blocks)", free, free >> UBIFS_BLOCK_SHIFT); @@ -387,8 +386,7 @@ static int ubifs_statfs(struct dentry *dentry, struct kstatfs *buf) buf->f_files = 0; buf->f_ffree = 0; buf->f_namelen = UBIFS_MAX_NLEN; - buf->f_fsid.val[0] = le32_to_cpu(uuid[0]) ^ le32_to_cpu(uuid[2]); - buf->f_fsid.val[1] = le32_to_cpu(uuid[1]) ^ le32_to_cpu(uuid[3]); + return 0; } @@ -532,12 +530,6 @@ static int init_constants_early(struct ubifs_info *c) c->dead_wm = ALIGN(MIN_WRITE_SZ, c->min_io_size); c->dark_wm = ALIGN(UBIFS_MAX_NODE_SZ, c->min_io_size); - /* - * Calculate how many bytes would be wasted at the end of LEB if it was - * fully filled with data nodes of maximum size. This is used in - * calculations when reporting free space. - */ - c->leb_overhead = c->leb_size % UBIFS_MAX_DATA_NODE_SZ; return 0; } @@ -655,11 +647,13 @@ static int init_constants_late(struct ubifs_info *c) * internally because it does not make much sense for UBIFS, but it is * necessary to report something for the 'statfs()' call. * - * Subtract the LEB reserved for GC, the LEB which is reserved for - * deletions, and assume only one journal head is available. + * Subtract the LEB reserved for GC and the LEB which is reserved for + * deletions. + * + * Review 'ubifs_calc_available()' if changing this calculation. */ - tmp64 = c->main_lebs - 2 - c->jhead_cnt + 1; - tmp64 *= (uint64_t)c->leb_size - c->leb_overhead; + tmp64 = c->main_lebs - 2; + tmp64 *= (uint64_t)c->leb_size - c->dark_wm; tmp64 = ubifs_reported_space(c, tmp64); c->block_cnt = tmp64 >> UBIFS_BLOCK_SHIFT; diff --git a/trunk/fs/ubifs/tnc.c b/trunk/fs/ubifs/tnc.c index 7da209ab9378..e909f4a96443 100644 --- a/trunk/fs/ubifs/tnc.c +++ b/trunk/fs/ubifs/tnc.c @@ -506,7 +506,7 @@ static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key, if (keys_cmp(c, key, &node_key) != 0) ret = 0; } - if (ret == 0 && c->replaying) + if (ret == 0) dbg_mnt("dangling branch LEB %d:%d len %d, key %s", zbr->lnum, zbr->offs, zbr->len, DBGKEY(key)); return ret; @@ -1382,39 +1382,50 @@ static int lookup_level0_dirty(struct ubifs_info *c, const union ubifs_key *key, } /** - * maybe_leb_gced - determine if a LEB may have been garbage collected. + * ubifs_tnc_lookup - look up a file-system node. * @c: UBIFS file-system description object - * @lnum: LEB number - * @gc_seq1: garbage collection sequence number + * @key: node key to lookup + * @node: the node is returned here * - * This function determines if @lnum may have been garbage collected since - * sequence number @gc_seq1. If it may have been then %1 is returned, otherwise - * %0 is returned. + * This function look up and reads node with key @key. The caller has to make + * sure the @node buffer is large enough to fit the node. Returns zero in case + * of success, %-ENOENT if the node was not found, and a negative error code in + * case of failure. */ -static int maybe_leb_gced(struct ubifs_info *c, int lnum, int gc_seq1) +int ubifs_tnc_lookup(struct ubifs_info *c, const union ubifs_key *key, + void *node) { - int gc_seq2, gced_lnum; + int found, n, err; + struct ubifs_znode *znode; + struct ubifs_zbranch zbr, *zt; - gced_lnum = c->gced_lnum; - smp_rmb(); - gc_seq2 = c->gc_seq; - /* Same seq means no GC */ - if (gc_seq1 == gc_seq2) - return 0; - /* Different by more than 1 means we don't know */ - if (gc_seq1 + 1 != gc_seq2) - return 1; - /* - * We have seen the sequence number has increased by 1. Now we need to - * be sure we read the right LEB number, so read it again. - */ - smp_rmb(); - if (gced_lnum != c->gced_lnum) - return 1; - /* Finally we can check lnum */ - if (gced_lnum == lnum) - return 1; - return 0; + mutex_lock(&c->tnc_mutex); + found = ubifs_lookup_level0(c, key, &znode, &n); + if (!found) { + err = -ENOENT; + goto out; + } else if (found < 0) { + err = found; + goto out; + } + zt = &znode->zbranch[n]; + if (is_hash_key(c, key)) { + /* + * In this case the leaf node cache gets used, so we pass the + * address of the zbranch and keep the mutex locked + */ + err = tnc_read_node_nm(c, zt, node); + goto out; + } + zbr = znode->zbranch[n]; + mutex_unlock(&c->tnc_mutex); + + err = ubifs_tnc_read_node(c, &zbr, node); + return err; + +out: + mutex_unlock(&c->tnc_mutex); + return err; } /** @@ -1425,19 +1436,16 @@ static int maybe_leb_gced(struct ubifs_info *c, int lnum, int gc_seq1) * @lnum: LEB number is returned here * @offs: offset is returned here * - * This function look up and reads node with key @key. The caller has to make - * sure the @node buffer is large enough to fit the node. Returns zero in case - * of success, %-ENOENT if the node was not found, and a negative error code in - * case of failure. The node location can be returned in @lnum and @offs. + * This function is the same as 'ubifs_tnc_lookup()' but it returns the node + * location also. See 'ubifs_tnc_lookup()'. */ int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key, void *node, int *lnum, int *offs) { - int found, n, err, safely = 0, gc_seq1; + int found, n, err; struct ubifs_znode *znode; struct ubifs_zbranch zbr, *zt; -again: mutex_lock(&c->tnc_mutex); found = ubifs_lookup_level0(c, key, &znode, &n); if (!found) { @@ -1448,43 +1456,24 @@ int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key, goto out; } zt = &znode->zbranch[n]; - if (lnum) { - *lnum = zt->lnum; - *offs = zt->offs; - } if (is_hash_key(c, key)) { /* * In this case the leaf node cache gets used, so we pass the * address of the zbranch and keep the mutex locked */ + *lnum = zt->lnum; + *offs = zt->offs; err = tnc_read_node_nm(c, zt, node); goto out; } - if (safely) { - err = ubifs_tnc_read_node(c, zt, node); - goto out; - } - /* Drop the TNC mutex prematurely and race with garbage collection */ zbr = znode->zbranch[n]; - gc_seq1 = c->gc_seq; mutex_unlock(&c->tnc_mutex); - if (ubifs_get_wbuf(c, zbr.lnum)) { - /* We do not GC journal heads */ - err = ubifs_tnc_read_node(c, &zbr, node); - return err; - } + *lnum = zbr.lnum; + *offs = zbr.offs; - err = fallible_read_node(c, key, &zbr, node); - if (maybe_leb_gced(c, zbr.lnum, gc_seq1)) { - /* - * The node may have been GC'ed out from under us so try again - * while keeping the TNC mutex locked. - */ - safely = 1; - goto again; - } - return 0; + err = ubifs_tnc_read_node(c, &zbr, node); + return err; out: mutex_unlock(&c->tnc_mutex); @@ -1509,6 +1498,7 @@ static int do_lookup_nm(struct ubifs_info *c, const union ubifs_key *key, { int found, n, err; struct ubifs_znode *znode; + struct ubifs_zbranch zbr; dbg_tnc("name '%.*s' key %s", nm->len, nm->name, DBGKEY(key)); mutex_lock(&c->tnc_mutex); @@ -1532,7 +1522,11 @@ static int do_lookup_nm(struct ubifs_info *c, const union ubifs_key *key, goto out_unlock; } - err = tnc_read_node_nm(c, &znode->zbranch[n], node); + zbr = znode->zbranch[n]; + mutex_unlock(&c->tnc_mutex); + + err = tnc_read_node_nm(c, &zbr, node); + return err; out_unlock: mutex_unlock(&c->tnc_mutex); diff --git a/trunk/fs/ubifs/ubifs-media.h b/trunk/fs/ubifs/ubifs-media.h index a9ecbd9af20d..bd2121f3426e 100644 --- a/trunk/fs/ubifs/ubifs-media.h +++ b/trunk/fs/ubifs/ubifs-media.h @@ -87,7 +87,7 @@ #define UBIFS_SK_LEN 8 /* Minimum index tree fanout */ -#define UBIFS_MIN_FANOUT 3 +#define UBIFS_MIN_FANOUT 2 /* Maximum number of levels in UBIFS indexing B-tree */ #define UBIFS_MAX_LEVELS 512 diff --git a/trunk/fs/ubifs/ubifs.h b/trunk/fs/ubifs/ubifs.h index 17c620b93eec..d7f706f7a302 100644 --- a/trunk/fs/ubifs/ubifs.h +++ b/trunk/fs/ubifs/ubifs.h @@ -995,9 +995,6 @@ struct ubifs_mount_opts { * @max_idx_node_sz: maximum indexing node aligned on 8-bytes boundary * @max_inode_sz: maximum possible inode size in bytes * @max_znode_sz: size of znode in bytes - * - * @leb_overhead: how many bytes are wasted in an LEB when it is filled with - * data nodes of maximum size - used in free space reporting * @dead_wm: LEB dead space watermark * @dark_wm: LEB dark space watermark * @block_cnt: count of 4KiB blocks on the FS @@ -1031,8 +1028,6 @@ struct ubifs_mount_opts { * @sbuf: a buffer of LEB size used by GC and replay for scanning * @idx_gc: list of index LEBs that have been garbage collected * @idx_gc_cnt: number of elements on the idx_gc list - * @gc_seq: incremented for every non-index LEB garbage collected - * @gced_lnum: last non-index LEB that was garbage collected * * @infos_list: links all 'ubifs_info' objects * @umount_mutex: serializes shrinker and un-mount @@ -1229,8 +1224,6 @@ struct ubifs_info { int max_idx_node_sz; long long max_inode_sz; int max_znode_sz; - - int leb_overhead; int dead_wm; int dark_wm; int block_cnt; @@ -1264,8 +1257,6 @@ struct ubifs_info { void *sbuf; struct list_head idx_gc; int idx_gc_cnt; - volatile int gc_seq; - volatile int gced_lnum; struct list_head infos_list; struct mutex umount_mutex; @@ -1443,10 +1434,9 @@ void ubifs_release_ino_dirty(struct ubifs_info *c, struct inode *inode, struct ubifs_budget_req *req); void ubifs_cancel_ino_op(struct ubifs_info *c, struct inode *inode, struct ubifs_budget_req *req); -long long ubifs_get_free_space(struct ubifs_info *c); +long long ubifs_budg_get_free_space(struct ubifs_info *c); int ubifs_calc_min_idx_lebs(struct ubifs_info *c); void ubifs_convert_page_budget(struct ubifs_info *c); -long long ubifs_reported_space(const struct ubifs_info *c, uint64_t free); long long ubifs_calc_available(const struct ubifs_info *c, int min_idx_lebs); /* find.c */ @@ -1461,6 +1451,8 @@ int ubifs_save_dirty_idx_lnums(struct ubifs_info *c); /* tnc.c */ int ubifs_lookup_level0(struct ubifs_info *c, const union ubifs_key *key, struct ubifs_znode **zn, int *n); +int ubifs_tnc_lookup(struct ubifs_info *c, const union ubifs_key *key, + void *node); int ubifs_tnc_lookup_nm(struct ubifs_info *c, const union ubifs_key *key, void *node, const struct qstr *nm); int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key, diff --git a/trunk/include/asm-generic/sections.h b/trunk/include/asm-generic/sections.h index 79a7ff925bf8..8feeae1f2369 100644 --- a/trunk/include/asm-generic/sections.h +++ b/trunk/include/asm-generic/sections.h @@ -14,10 +14,4 @@ extern char __kprobes_text_start[], __kprobes_text_end[]; extern char __initdata_begin[], __initdata_end[]; extern char __start_rodata[], __end_rodata[]; -/* function descriptor handling (if any). Override - * in asm/sections.h */ -#ifndef dereference_function_descriptor -#define dereference_function_descriptor(p) (p) -#endif - #endif /* _ASM_GENERIC_SECTIONS_H_ */ diff --git a/trunk/include/asm-parisc/sections.h b/trunk/include/asm-parisc/sections.h index 9d13c3507ad6..fdd43ec42ec5 100644 --- a/trunk/include/asm-parisc/sections.h +++ b/trunk/include/asm-parisc/sections.h @@ -4,9 +4,4 @@ /* nothing to see, move along */ #include -#ifdef CONFIG_64BIT -#undef dereference_function_descriptor -void *dereference_function_descriptor(void *); -#endif - #endif diff --git a/trunk/lib/vsprintf.c b/trunk/lib/vsprintf.c index c399bc1093cb..d8d1d1142248 100644 --- a/trunk/lib/vsprintf.c +++ b/trunk/lib/vsprintf.c @@ -27,7 +27,6 @@ #include /* for PAGE_SIZE */ #include -#include /* for dereference_function_descriptor() */ /* Works only for digits and letters, but small and fast */ #define TOLOWER(x) ((x) | 0x20) @@ -514,6 +513,16 @@ static char *string(char *buf, char *end, char *s, int field_width, int precisio return buf; } +static inline void *dereference_function_descriptor(void *ptr) +{ +#if defined(CONFIG_IA64) || defined(CONFIG_PPC64) + void *p; + if (!probe_kernel_address(ptr, p)) + ptr = p; +#endif + return ptr; +} + static char *symbol_string(char *buf, char *end, void *ptr, int field_width, int precision, int flags) { unsigned long value = (unsigned long) ptr;